<%BANNER%>

Using Detector Arrays to Improve the Efficiency of Linear Accelerator Quality Assurance and Radiation Data Collection

Permanent Link: http://ufdc.ufl.edu/UFE0041926/00001

Material Information

Title: Using Detector Arrays to Improve the Efficiency of Linear Accelerator Quality Assurance and Radiation Data Collection
Physical Description: 1 online resource (135 p.)
Language: english
Creator: Simon, Thomas
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2010

Subjects

Subjects / Keywords: array, assurance, calibration, detector, dosimetry, linear, quality, radiation, radiotherapy, relative, therapy
Nuclear and Radiological Engineering -- Dissertations, Academic -- UF
Genre: Nuclear Engineering Sciences thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: The complexity of radiation therapy is continually increasing as new treatment modalities are implemented in the clinic. While these advances often benefit tumor dose localization, they also increase pressure on departmental resources as the new modality is adopted. This driving force comes at a time of increased pressure to perform quality assurance (QA) of the entire treatment process. The effect is a work force with too many measurements to do and not enough time in which to do them. The purpose of this work is to establish the use of detector arrays to improve the automation and efficiency of linear accelerator (LINAC) quality assurance and radiation data collection. Two traditionally time consuming measurement processes were evaluated for the potential for increased efficiency and automation: multi-leaf collimator (MLC) calibration and scanning water tank measurements. Using traditional measurement techniques, MLC calibration can take hours to accomplish with mixed results or require a significant investment of time to write in-house software. We developed a quantitative and efficient (less than 30 minutes for both leaf banks) MLC calibration method that we termed the radiation defined reference line (RDRL) method. The method uses a detector array PROFILER 2(tm); Sun Nuclear Corporation (SNC), Melbourne, FL USA to measure the penumbral position of each leaf relative to a known reference point (or line). Profile measurements are typically obtained with a scanning water tank. While time tested, the system requires above average skill and time to properly setup and acquire data. We extensively characterized and assessed the potential of a multi-axis ionization chamber array IC PROFILER(tm); SNC to measure water tank equivalent profiles. The IC PROFILER(tm) had an error spread of approximately (+/-) 0.75% relative to a water scan, with the potential of a positive offset in that error. During the characterization, the array calibration method was found to be susceptible to the LINACs symmetry stability. Symmetry variations of (+/-) 0.1% can cause calibration errors of (+/-) 2%. The cause was investigated and corrective measures were developed. Finally, a time efficient QA program was developed to determine the operation of the detector arrays.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Thomas Simon.
Thesis: Thesis (Ph.D.)--University of Florida, 2010.
Local: Adviser: Liu, Chihray.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2010
System ID: UFE0041926:00001

Permanent Link: http://ufdc.ufl.edu/UFE0041926/00001

Material Information

Title: Using Detector Arrays to Improve the Efficiency of Linear Accelerator Quality Assurance and Radiation Data Collection
Physical Description: 1 online resource (135 p.)
Language: english
Creator: Simon, Thomas
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2010

Subjects

Subjects / Keywords: array, assurance, calibration, detector, dosimetry, linear, quality, radiation, radiotherapy, relative, therapy
Nuclear and Radiological Engineering -- Dissertations, Academic -- UF
Genre: Nuclear Engineering Sciences thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: The complexity of radiation therapy is continually increasing as new treatment modalities are implemented in the clinic. While these advances often benefit tumor dose localization, they also increase pressure on departmental resources as the new modality is adopted. This driving force comes at a time of increased pressure to perform quality assurance (QA) of the entire treatment process. The effect is a work force with too many measurements to do and not enough time in which to do them. The purpose of this work is to establish the use of detector arrays to improve the automation and efficiency of linear accelerator (LINAC) quality assurance and radiation data collection. Two traditionally time consuming measurement processes were evaluated for the potential for increased efficiency and automation: multi-leaf collimator (MLC) calibration and scanning water tank measurements. Using traditional measurement techniques, MLC calibration can take hours to accomplish with mixed results or require a significant investment of time to write in-house software. We developed a quantitative and efficient (less than 30 minutes for both leaf banks) MLC calibration method that we termed the radiation defined reference line (RDRL) method. The method uses a detector array PROFILER 2(tm); Sun Nuclear Corporation (SNC), Melbourne, FL USA to measure the penumbral position of each leaf relative to a known reference point (or line). Profile measurements are typically obtained with a scanning water tank. While time tested, the system requires above average skill and time to properly setup and acquire data. We extensively characterized and assessed the potential of a multi-axis ionization chamber array IC PROFILER(tm); SNC to measure water tank equivalent profiles. The IC PROFILER(tm) had an error spread of approximately (+/-) 0.75% relative to a water scan, with the potential of a positive offset in that error. During the characterization, the array calibration method was found to be susceptible to the LINACs symmetry stability. Symmetry variations of (+/-) 0.1% can cause calibration errors of (+/-) 2%. The cause was investigated and corrective measures were developed. Finally, a time efficient QA program was developed to determine the operation of the detector arrays.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Thomas Simon.
Thesis: Thesis (Ph.D.)--University of Florida, 2010.
Local: Adviser: Liu, Chihray.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2010
System ID: UFE0041926:00001


This item has the following downloads:


Full Text





USING DETECTOR ARRAYS TO IMPROVE THE EFFICIENCY OF LINEAR
ACCELERATOR QUALITY ASSURANCE AND RADIATION DATA COLLECTION




















By

THOMAS ALLAN SIMON


A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY

UNIVERSITY OF FLORIDA

2010

































2010 Thomas Allan Simon

































To my darling wife









ACKNOWLEDGMENTS

First, I would like to thank my advisor, Dr. Chihray Liu, for continually showing me

how to be a good physicist and a good person through his direct demonstration. Next, I

would also like to thank my committee members, Drs. Jonathan Li, Sanjiv Samant, and

William Mendenhall for their guidance and support. Additionally, I would like to thank Dr.

Darren Kahler for help in preparing this manuscript and Phil Basset for useful

discussions about linear accelerator operation.

Then there are my fellow graduate students and colleagues. Useful discussions

and even more useful distractions were a necessity during eight long years of graduate

school. I would particularly like to acknowledge Kevin Segall, Christopher Fox, Bart

Lynch, Heeteak Chung, Guanghua Yan, Jean Peng, and Matthew Williams.

Finally, I am indebted to my family for their continued support. My wife, Naomi,

deserves more credit that words can provide for always encouraging and supporting

me. My brother, Jeff, and his wife and my friend, Mariel, were always there to push me.

This leaves my parents, Bill and Cathy. Bill's guidance and encouragement was my

guiding light. Cathy's pride in me, despite too many years in graduate school, kept me

from realizing that it was indeed too many years.









TABLE OF CONTENTS

page

ACKNOW LEDGMENTS ........................................ ...............

LIST O F TA BLES ......... ................ ..................... ...... ............... 8

L IS T O F F IG U R E S .......................................................................................................... 9

LIST OF ABBREVIATIONS ......... ........... ......................... ............... 11

ABSTRACT ........... ......... ..... ..... ..... ...................... ......... 13

CHAPTER

1 INTRODUCTION ........ .......... .............. .................. ... ............ 15

G general Introduction .......................... .. ...................... ...... ........... .... 15
New Treatment Modalities Increase a Facility's Resource Requirements........ 16
How Do We Solve this Problem? ............................................ 17
Streamlining Data Collection ................. ............... ............... 19
S tudy A im s.............................. .............. ...... 2 0

2 MLC CALIBRATION USING A DETECTOR ARRAY.................................... 23

Intro d uctio n ......... ...... ............ ................................. ........................... 2 3
Materials and Methods....................................... .......... 24
M materials .......................................... .............. 24
Linear accelerator and MLC...... .................................... 24
Detector arrays .................... ......... ................ 25
M ethods................................ .....................26
Measuring minor leaf offsets ...... ..................... ............... 27
Measuring major leaf offsets ...... ..................... ............... 33
MLC calibration .................... ........ ................. 36
O their M LC types ............................ ......... ................ .... ........ 37
Results .................... ..... ............ ............................... 37
Detector Offsets ........................... ............... .. ......................... 37
MLC Measurement Comparison and Reproducibility ................................... 38
MLC Service Issues ............... .................................. 39
D discussion ........................................................................................................... 39
MLC Calibration Stability and QA ......................... ......... .... ........... 39
Device Com prisons ............. ......... ....... .................................... 40
Conclusion ................ .................................... ... ..... ........... ......... 42

3 CHARACTERIZATION OF A MULTI-AXIS IONIZATION CHAMBER ARRAY ....... 47

Introduction ............... ...... ............................................................ ...... 47









Materials and Methods................................ ............... 48
M aterials................................................. ............... 48
M ethods............................................... 49
Reproducibility ................................. .. ............ .............. 50
Dose and instantaneous dose rate dependence................................. 50
PRF dependence ............................................... 52
Energy dependence..... ....... ........................ .. 53
Response to power being applied to the electronics.............. .......... 54
Calibration constancy............................................... .................... 55
Backscatter dependence ........................................... ......... .... ............... 56
Beam profile measurements and output factors............................... 57
Results and Discussion.......................................... ............... 58
Reproducibility ................. ..................... ....... .. ............... 58
Dose and Instantaneous Dose Rate Dependence ..................................... 58
PR F Dependence .. ................................. ........................................... 61
Energy Dependence........................................ ........ ............. 63
Response to Power Applied to the Electronics......... .. ....... .............. 64
C alibration C onstancy ...................... ....... ......... .. ............................ 64
Backscatter Dependence ............................................... ..................... 65
Beam Profile Measurements and Output Factors.............................. 66
C o nclusio n ......... ...... ............ .................................. ........................... 6 7

4 WIDE FIELD ARRAY CALIBRATION DEPENDENCE ON THE STABILITY OF
MEASURED DOSE DISTRIBUTIONS.................... ........ ... ............... 78

Introduction ....................................... ................. .................. 78
Materials and Methods................................................... 79
M materials .......................................... .......... 79
M ethods.......................................................................... 80
W ide field calibration theory......................................... .......................... 80
Lim iting ca liberation error............................ ........ ........... ... 83
Effects of postulate failure............................................ 83
Lim iting violations of the first postulate ................................ ... ................ 84
Lim iting violations of the second postulate........................ .. .... ............ 85
Lim iting violations of the third postulate ............................... ... ................ 86
Evaluating calibration factors ...... ..................... .............. 87
R results and D discussion ................................................ ............. 90
Effects of Postulate Failure........................... .... ................... 90
Limiting Violations of the First Calibration Postulate.................... .......... 91
Lim iting Violations of the Third Calibration Postulate .................................... 92
Evaluating Calibration Factors........................... .... .......................... 93
Other Factors Affecting the W F Calibration................................ ... ................ 94
Other Arrays and IMRT ........ ................................ ... ... ............. 95
Conclusion ........................... ...... ........ ... .................. 95

5 A QUALITY ASSURANCE PROGRAM FOR A DETECTOR ARRAY................... 102



6









In tro d u ctio n ................ 1..........................02....................
Materials and Methods......................................... 103
Materials............................. ............... 103
M etho d s............................................................................... ............... 104
P physical ................................................................ 104
Firm w are and softw are ....................................................... .......... 105
E le ctro n ics .................................................. 10 8
A rra y c a lib ra tio n ................................................... ....................... 1 1 1
Results and Discussion....................................... 114
Physical ............................ ................. ............... 114
Firm w are and Softw are ......................................... .......... .................... 115
E le c tro n ic s ................ .................................. ................................. 1 1 5
Array Calibration.......................... .................. 119
C o n c lu s io n .................................................................................................. 1 2 0

6 SUMMARY AND FUTURE WORK .............. ............ ................... 126

LIST OF REFERENCES ......................... ......... .... ... ............... 130

BIOGRAPHICAL SKETCH .......................................... 135









LIST OF TABLES


Table page

2-1 List of collim ator settings for PRO FILER 2TM ................................ ................ 43

2-2 List of additional collimator settings for PROFILER 2TM measurements of the
m major leaf offsets. .............. ..... ............ ........... .. ............ ...... ......... 43

3-1 List of subscripts, variables, and equations ...... .... ........ ............................ 69

3-2 The panel's short and long-term reproducibility were evaluated on a 60Co
teletherapy unit ........... ......... .......... .. ... ........ ............... .............. 70

3-3 The short and long term reproducibility of the relative detector calibration
factors............... ..................................................................... 70

4-1 The short term reproducibility of WF calibrations performed on the Elekta ........ 97

5-1 Quality assurance program for the IC PROFILERTM ............ ................. 122

5-2 Percent of detectors rejecting Ho ............. ........ ................... ............. 122









LIST OF FIGURES


Figure page

2-1 Minor leaf offsets are defined as the spatial offset of each leaf in a leaf........... 44

2-2 (A) The radiation defined reference line (RDRL) method requires.................. 44

2-3 The measurements required to radiographically align the array...................... 45

2-4 (A) Detector offsets relative to a reference detector. .......................................... 45

2-5 MLC leaf offsets relative to leaf number 20, the reference leaf. ......................... 46

2-6 Change in relative MLC offsets after replacement of the primary....................... 46

2-7 Calibration of the X1 MLC leaf bank ............ ............. ....... ..... .......... 46

3-1 Overlay of the IC PROFILERTM (panel) showing the multiple detector............ 71

3-2 The panel's dose response relative to the Farmer-type chamber's dose .......... 71

3-3 The panel's instantaneous dose rate response relative to the Farmer-............. 72

3-4 The panel's PRF response relative to the Farmer-type chamber's PRF........... 72

3-5 The off-axis PRF response for the x-axis detectors relative to the center......... 73

3-6 The difference between the 1800 and 0 OA_Response(PRF) values for......... 73

3-7 The energy response of the panel's center detector presented as a ratio......... 73

3-8 The accuracy of the calibration factors ................................. .................. ..... 74

3-9 The energy response of the calibration factors........................... ... ............... 74

3-10 The buildup response of the calibration factors for 6 and 18 MV.................... 74

3-11 Off axis backscatter response of the x-axis detectors for (A) a 6 MV ................. 75

3-12 Normalized cross-plane measurements with a CC13TM and the panel .............. 75

3-13 Profile agreement (over 80% of the field width) between the panel and........... 76

3-14 FDDs for a 6 MV 10 x 10 cm2 field. ............. ......... ................. 76

3-15 Output factors measured with the panel's center chamber and three.............. 77

4-1 Wide field (WF) calibration reproducibility on LINACs with beam ................. 98









4-2 Oblique view of the panel's arrays and electronics. The panel's y-axis is .......... 98

4-3 (A) The perturbation that was applied to the hypothetical calibration ................ 99

4-4 The percentage error between ten consecutive measurements and their.......... 99

4-5 Calibration reproducibility using a continuous beam during............................ 100

4-6 The effect of additional side-scatter on (A) beam measurements and (B)........ 100

4-7 The agreement between calibration factors obtained with side-scatter ............ 100

4-8 The calibration accuracy was evaluated for four measurement...................... 101

4-9 The calibration accuracy expressed as the ratio between water tank............. 101

5-1 Cable damage as a result of (A) disconnecting the cable by pulling the........... 123

5-2 The measurement reproducibility for the panel on the Elekta Synergy.......... 123

5-3 (A) An example of two background measurements that were separated ........ 124

5-4 The PDF of UC for the (-) 12.5 cm x-axis detector (X8)................................... 124

5-5 The accuracy of the panel's calibration factors were determined using............ 125

5-6 (A) Profile measurements with the panel and a scanning water tank ............. 125









LIST OF ABBREVIATIONS

AAPM American Association of Physicists in Medicine

CAX Central axis of the LINAC coordinate system

CCD Charge coupled device camera used in the control of the MLC

EPID Electronic portal imaging device

EDW Enhanced dynamic wedge

FDD Fractional depth dose

FMEA Failure mode and effects analysis

IBA Ion Beam Applications (equipment manufacturer)

IMRT Intensity modulated radiation therapy

LINAC Linear accelerator

LSI Local standard instrument

MALO Major leaf offset

MILO Minor leaf offset

MLC Multi-leaf collimator

NIST Nation Institute of Standards and Technology

panel IC PROFILERTM

PTW Physikalisch-Technische Werkstatten (equipment manufacturer)

QA Quality assurance

RDRL Radiation defined reference line method

SA1 Specific aim 1

SA2 Specific aim 2

SA3 Specific aim 3

SA4 Specific aim 4

SNC Sun Nuclear Corporation (equipment manufacturer)









SDD Source to detector distance

SSD Source to surface distance

TG Task group

TPS Treatment planning system









Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy

USING DETECTOR ARRAYS TO IMPROVE THE EFFICIENCY OF LINEAR
ACCELERATOR QUALITY ASSURANCE AND RADIATION DATA COLLECTION

By

Thomas Allan Simon

August 2010

Chair: Chihray Liu
Major: Nuclear Engineering Sciences

The complexity of radiation therapy is continually increasing as new treatment

modalities are implemented in the clinic. While these advances often benefit tumor dose

localization, they also increase pressure on departmental resources as the new

modality is adopted. This driving force comes at a time of increased pressure to perform

quality assurance (QA) of the entire treatment process. The effect is a work force with

too many measurements to do and not enough time in which to do them. The purpose

of this work is to establish the use of detector arrays to improve the automation and

efficiency of linear accelerator (LINAC) quality assurance and radiation data collection.

Two traditionally time consuming measurement processes were evaluated for the

potential for increased efficiency and automation: multi-leaf collimator (MLC) calibration

and scanning water tank measurements. Using traditional measurement techniques,

MLC calibration can take hours to accomplish with mixed results or require a significant

investment of time to write in-house software. We developed a quantitative and efficient

(less than 30 minutes for both leaf banks) MLC calibration method that we termed the

radiation defined reference line (RDRL) method. The method uses a detector array









[PROFILER 2TM; Sun Nuclear Corporation (SNC), Melbourne, FL USA] to measure the

penumbral position of each leaf relative to a known reference point (or line).

Profile measurements are typically obtained with a scanning water tank. While

time tested, the system requires above average skill and time to properly setup and

acquire data. We extensively characterized and assessed the potential of a multi-axis

ionization chamber array (IC PROFILERTM; SNC) to measure water tank equivalent

profiles. The IC PROFILERTM had an error spread of approximately () 0.75% relative to

a water scan, with the potential of a positive offset in that error. During the

characterization, the array calibration method was found to be susceptible to the

LINACs symmetry stability. Symmetry variations of () 0.1% can cause calibration

errors of () 2%. The cause was investigated and corrective measures were developed.

Finally, a time efficient QA program was developed to determine the operation of the

detector arrays.









CHAPTER 1
INTRODUCTION

General Introduction

The complexity of radiation therapy is continually increasing as new treatment

modalities are implemented in the clinic. While these advances often benefit tumor dose

localization, they also increase pressure on departmental resources as the new

modality is adopted.1 An example of this cause and effect is the use of dynamic wedges

over the more traditional physical wedge.

The enhanced dynamic wedge (EDW) increased the efficiency of radiation

treatment by eliminating the need to enter the treatment vault to install a wedge. Instead

of a physical beam attenuator, the wedge intensity pattern is created by dynamically

sweeping the collimator across the field while delivering radiation. The EDW increased

treatment efficiency; but it also complicated the radiation data collection process.

Measuring an EDW with a scanning water tank requires each profile point to be

integrated through an entire delivery. Measuring a range of field sizes, depths, and

energies could extend to several days.

To cope with the increased measurement requirements, the clinical physicist relied

on a new measurement technology detector arrays. The use of detector arrays greatly

increased the efficiency of EDW measurements due to their simultaneous measurement

of multiple points. What previously took scanning water tanks several days to measure

was reduced to a few hours with detector arrays.

Just as dynamic wedges increased the complexity of radiation therapy, an

abundance of new treatment modalities are poised to do the same.2-4 This comes at a

time when the demands on the physics department are already high.5' 6 To adapt and









evolve with the shifting clinical environment, we need to reevaluate how quality

assurance (QA) and radiation data collection is performed. The purpose of this work is

to establish the use of detector arrays to improve the automation and efficiency of

LINAC QA and radiation data collection.

New Treatment Modalities Increase a Facility's Resource Requirements

New treatment modalities in radiation therapy are becoming increasingly complex

through the incorporation of various imaging, treatment planning system (TPS), patient

localization, and delivery options.24 While the level of patient care is improved, the

overall complexity and verification requirements for those treatments increase as well.

This cause and effect is usually accompanied with a corresponding lag in the evolution

of QA technology.1 An example lies in the introduction of intensity modulated radiation

therapy (IMRT).

The radiation therapy process tree grew in complexity with the adoption of IMRT. It

affected all aspects of the clinic, but it greatly increased the time demand on physicists

in the form of treatment verification.7 During its initial introduction, the primary method

for verifying two dimensional dose distributions was film dosimetry.7'8 While this

verification process provided high-spatial-density-dosimetric-information, it was

inefficient considering the multi-field, multi-film nature of IMRT QA. A simple seven field

prostate plan could take up to two hours to verify.

An emerging modality that threatens to further increase complexity is single arc

IMRT. This combines the principles of dynamic IMRT with the added complexity of

continuously moving gantry components and a varying dose rate.9 Just as film was

initially used to verify IMRT plans, the tools and methods that were developed for IMRT

are now being used to verify arc therapies.10-12 While these technologies may prove to









be just as arduous as film dosimetry was to IMRT, new dosimetry systems are already

emerging that promise to efficiently handle the QA requirements of single arc IMRT.13' 14

How Do We Solve this Problem?

With each new treatment modality, new verification requirements are added to the

existing regime. So many tests already exist that it would be nearly impossible to do

them all.15 This begs the question "is all of this QA needed?" For example, if a physicist

performs annual QA that takes 10 hours and no problems were discovered, then were

those 10 hours wasted? Could they have been spent on an item that is more likely to

fail? Does this mean we should not perform annual QA? No; it means we need to

approach certain aspects of QA and data collection differently.

Looking at QA in a different light is already being addressed. The American

Association of Physicists in Medicine (AAPM) is taking two separate approaches, led by

Task Groups (TGs) 100 and 142.16,17 The approach of TG-100 is to apply Failure Mode

and Effects Analysis (FMEA) to the radiation therapy process. This process is a

powerful tool that aids in the allocation of resources to a system in an effort to detect

modes of failure before they occur. The goal of TG-100 is for each radiation oncology

clinic to compose a personalized FMEA analysis for each treatment process tree in its

clinic. While a completed FMEA analysis would help to focus departmental resources,

the creation of the FMEA analysis will greatly increase the time requirements on the

clinic and its physicists.

The approach of TG-142 is somewhat different. Similar to TG-40, it provides

updated machine item tolerances. Unlike TG-40, it recommends different tolerances for

different modalities. While addressing that conventional radiotherapy LINACs do not

need the same mechanical tolerances that stereotactic-radio-surgery LINACs do, TG-









142 fails to address the associated time load that already exists. As new modalities are

introduced, the time load will only increase.

Both TG-100 and 142 provide useful QA recommendations and insights. However,

the overall result is likely to be an increase in time and measurement requirements. We

feel that measurement automation and increased efficiency is the key to aid the clinical

physicist.

Again, dosimetric verification of IMRT illustrates an example. A typical seven field

prostate plan requires nearly 2 hours for verification using film. There is also a high

potential for error in the film dosimetry process that can be a source of inconsistency

throughout the community.18-21 The introduction of two dimensional detector arrays

streamlined the IMRT QA process. The effect was a reduction in the time requirements

from nearly two hours to 45 minutes for a typical seven field prostate exam. The

acquisition of the results was also standardized to a certain extent with the 2D arrays.22

This provided the community with a much more consistent reporting of results and also

an increase in care.

A second example exists with the measurement of EDWs. Initially, scanning water

tanks were used to measure EDWs with a series of point measurements.23 Each

measurement point represented the duration of an EDW delivery. This was a time

intensive undertaking even without factoring in the time of tank setup and break-down.

The whole process could easily take several days to complete. The introduction of

detector arrays that are mounted in a scanning water tank decreased the time

requirements by providing multiple integration points per measurement.23 24 However,









the process still required a scanning water tank. The introduction of a detector array that

lay on the treatment reduced the required time to minutes.25

Streamlining Data Collection

Automating and streamlining the QA and data collection regimen is a daunting

task due to the variety of measurements and delivery systems. It is nearly impossible for

one agency or group to solve all of these problems. However, a published group of tests

that the agencies could recommend would help to solve this problem. This would allow

for a much faster response time as QA and data collection demands shifted.

As an example, two methods that have traditionally been resource intensive were

investigated for the possibility of streamlining. They are the calibration of MLCs and

scanning water tank measurements.

The current methods for calibrating MLCs include the use of graph paper, film,

electronic portal imaging devices (EPIDs), and scanning water tanks. Each of these

measurement methods has certain advantages and disadvantages. Graph paper and

film are both time tested and intuitive. However, they are resource intensive and may

only provide qualitative results. Water tanks are well understood with highly accurate

and repeatable mechanics, but require a large amount of resources and suffer from

detector volume averaging unless a pinpoint (diamond,26 diode detector,27 etc.) detector

is used. The calibration of MLCs with any of these methods takes many hours to

complete. A dosimeter that is more efficient is the EPID. Unfortunately, they require

user written code due to a relative lack of commercial software. Detector arrays have

the potential to reduce that time requirement from hours to minutes.

Scanning water tanks are the gold standard of radiation therapy measurements.

They are time tested, precise, and reproducible. They do however require large









amounts of time for setup and data collection.28 They also require a higher degree of

skill to accurately use.28 The clinical physicist therefore rarely uses the scanning water

tank. Most are only used during the LINAC beam commissioning and annual QA.

Advances in computer technology have improved their efficiency. However, they

still require large amounts of preparation time (~ 1 12 hours to setup and break down)

and actual scanning time [several days to measure linear accelerator (LINAC) beam

commissioning data].28 Infrequent use by the clinical physicist adds to these

inefficiencies and decreases the likelihood of obtaining quality data. The skill required to

obtain quality beam profile measurements with a detector array is less. This is due to

physicists being more familiar with detector panels and also due to their lack of

mechanical parts and liquid water.

Study Aims

For this dissertation, two time consuming measurement processes were chosen

as a showcase for the potential of detector arrays to increase measurement automation

and efficiency. The first specific aim covers the calibration of multi-leaf-collimators

(MLC). The remaining specific aims deal with using the IC PROFILERTM as a water tank

alternative.

Specific Aim 1 (SA1) MLC calibration with detector arrays: Each leaf end creates

a penumbral position that corresponds to its actual position. An array that uses a

detector with minimal volume averaging can accurately measure these leaf positions for

QA or calibration purposes. The purpose of this aim is to create an efficient and

quantitative MLC calibration method that uses a commercially available detector array

(e.g. the PROFILER 2TM or an EPID).









Specific Aim 2 (SA2) Characterize the IC PROFILERTM: The IC PROFILERTM is

a multi-axis ion chamber array and therefore does not suffer from the undesirable

detector characteristics that diode detectors possess. However, it's potential as a water

tank alternative underlies the importance of fully understanding the device and how it

reacts in a radiation environment. The purpose of this aim is to do just that; extensively

characterize the IC PROFILERTM in a radiation environment and establish its ability to

measure LINAC beam parameters.

Specific Aim 3 (SA3) Increase the reproducibility of the wide field calibration

theory for use in unstable beams: The wide field calibration theory has become a

prominent fixture in the radiation oncology environment. It is used to correct the intra-

detector-sensitivity-variation in a wide variety of detector arrays, including the

MapCHECKTM, PROFILER 2TM, and IC PROFILERT. Accurate measurements with

these systems require confidence in the individual detectors' calibrations. However, the

calibration theory requires a perfectly reproducible LINAC beam; otherwise

unacceptable error levels are encountered. The purpose of this specific aim is to

minimize the effects of beam instability in the wide field calibration theory.

Specific Aim 4 (SA4) Establish a quality assurance program for the IC

PROFILERTM: The potential importance of the data provided by the IC PROFILERTM

(i.e. annual QA and LINAC commissioning) requires the highest level of confidence in

the array and it's measured data. The purpose of this specific aim is to develop a series

of field tests that indicate proper function of the IC PROFILERTM and detector arrays in

general.

In summary, this dissertation is organized into four specific aims:









SA1: Create an efficient MLC calibration method using detector arrays (Chapter

2).

SA2: Characterize the IC PROFILERTM in the radiation environment (Chapter 3).

SA3: Increase the reproducibility of the Wide Field Calibration theory when

operated in beams with micro instabilities (Chapter 4).

SA4: Establish a QA program for the IC PROFILERTM and detector arrays in

general (Chapter 5).









CHAPTER 2
MLC CALIBRATION USING A DETECTOR ARRAY

Introduction

Intensity modulated radiation therapy (IMRT) is a treatment modality that is used

to deliver a dose prescription to a tumor site while minimizing the exposure to

surrounding healthy tissues. A popular method of implementing IMRT is to superpose a

series of irregular fields that are shaped with a multi leaf collimator (MLC) to create a

complex radiation fluence map. The principles of radiation transport then govern the

conversion of the fluence map to a dose map.

The correct placement of high-gradient dose regions in and near the target volume

is dependent on an accurate positioning of the MLC leaves during the delivery of the

IMRT fields. In a recent study, Mu et al. demonstrated that a systematic leaf positioning

error of 1 mm in IMRT plans can result in dose errors of up to 7.6 % and 12.2 % for the

target and critical structures, respectively.29 Errors of this size can have a biologically

significant effect on the outcome of the therapy.30 For this reason the MLC must be

accurately calibrated and periodically tested.

MLC calibration requires the ability to precisely measure individual leaf positions.

Traditional methods of calibration are time consuming and/or non-reproducible in

nature. These methods include the use of graph paper,31 radiosensitive film,31'32

scanning water tanks, electronic portal imaging devices (EPID)'s,33-35 detector arrays,36

and manufacturer's proprietary methods. While each of these techniques has

advantages and disadvantages, the current trend is toward more efficient and

reproducible methods.









Recent publications have shown a refinement in measurement and calibration

techniques. In 2006 Parent et al. used an EPID to measure and predict the positions of

individual leaves on an Elekta MLC.33 In 2007 Lopes et al. used an ion-chamber array

mounted in a water tank to calibrate the individual leaf positions for a Siemens MLC.36

While both of these techniques are an improvement on the traditional methods, they still

require a significant investment of time. Using an EPID to measure leaf positions

requires mechanical and/or software corrections as well as user-written code. The ion

chamber array-based approach is susceptible to the volume averaging of the ion

chambers and requires the setup of a scanning water tank along with ancillary

equipment.

An integrated technique for MLC calibration exists as a proprietary method for

Elekta MLCs. The AutoCAL (Elekta Oncology Systems, Crawley, UK) software suite

uses EPID measurements to calibrate various machine items. This software has only

recently become available and is used exclusively with the Elekta EPID. While it

represents an important step toward more efficient and quantitative calibration

techniques, our initial uses have shown that it is prone to delays and calibrations with

unacceptable leaf positions.

The purpose of our research was to develop a more efficient and reproducible

method for calibrating an MLC.

Materials and Methods

Materials

Linear accelerator and MLC

All tests were performed with an Elekta Synergy (Elekta Oncology Systems,

Crawley, UK) linear accelerator (LINAC) using the 6 MV photon beam. The LINAC's









MLC is a 40 leaf-pair device that has been described in detail.37' 38 Each leaf projects to

a width of 1 cm at the isocentric plane (100 cm from source). Each MLC leaf bank is

located above a backup jaw that aids in beam collimation and reduces MLC radiation

transmission. An MLC leaf bank and its' associated backup jaw have parallel leading

edges that travel in the cross-plane direction when the collimator is set to 0 degrees

(IEC 1217 convention).39

Elekta leaf positions are controlled through an optical system that uses field light

reflected from a marker on top of each leaf. Reference reflectors are located in the

machine head outside of the largest obtainable field and are used to define the MLC

coordinate system. The reflected light rays trace through a series of mirrors to a charge

coupled device (CCD) camera that is interfaced to a control computer.

Detector arrays

The PROFILER 2TM is a two-axis detector array (Sun Nuclear Corporation,

Melbourne, FL) that consists of 139 diode detectors. The y-axis of the device contains

83 detectors over a length of 32.8 cm and the x-axis contains 57 detectors over a length

of 22.4 cm. Both axes have a detector spacing of 4 mm and share a central detector.

The inherent buildup of the device is 1 g/cm2 of water-equivalent material. The device

was chosen due to the detector spacing and the spatial measurement resolution of each

detector (0.8 x 0.8 mm2). Data collected using the PROFILER 2TM software can be

transferred (using copy and paste) to a spreadsheet program such as Excel (Microsoft,

Redmond, Washington) for analysis.

The EPID used in this study is an Elekta iView GT. It has a pixel dimension of 0.4

x 0.4 mm2 and a sensitive area of 41 x 41 cm2. It operates at a fixed source to surface

distance (SSD) of 160 cm. The iView software automatically projects collected images









to the isocentric plane by scaling the pixel and field dimensions to 0.25 x 0.25 mm2 and

25.6 x 25.6 cm2, respectively. Since an MLC leaf bank projects to a maximum length of

40 cm at isocenter, it is necessary to shift the EPID in order to fully image one MLC

bank.

The method described herein is specific to the Elekta MLC. However, the principle

is general and can be applied to other manufacturers' MLCs provided that appropriate

conditions for measurements are met.

Methods

We have termed this measurement technique the radiation defined reference line

(RDRL) method. Application of the method operates under three assumptions. First, the

leading edge of an MLC leaf bank is parallel to its backup jaw's leading edge. Second,

the backup jaw can provide a reproducible and uniform radiation field edge. This field

edge defines the RDRL. Third, the measured radiation field edge created by each leaf

end is representative of that leaf's position.

The third assumption of the RDRL method requires detectors with a spatial

measurement resolution that does not suffer from signal averaging in the high spatial

frequency of the penumbra. Dempsey has shown that measurements with a detector

size of 2 mm or smaller is sufficient for IMRT fields shaped with MLCs.40 The

PROFILER 2TM's detector size satisfies this requirement, but the detector location must

also be known with a precision better than the desired leaf position accuracy.

Elekta MLC leaf banks have traditionally been calibrated using standard

measurement tools, e.g. film and scanning water tanks, to determine what are termed

major and minor leaf offsets, as illustrated in Fig. 2-1. A reference leaf pair, leaf pair 20,

is used in the control of the MLC. The major leaf offset (MALO) is a calibration value









that defines the field size created by the reference leaf pair. Minor leaf offsets (MILO)

are the position alignment errors of the other leaves in relation to the reference leaf and

are the first focus of this method.

Measuring minor leaf offsets

The method described below uses the jaw edge to precisely locate all of the y-axis

detectors' offsets relative to the reference detector, as illustrated in Fig. 2-2A; the

reference detector is located in the reference leaf's direction of travel. These relative

detector positions are termed RDOj, where 'j' is the y-axis diode number 1 < j < 83; they

effectively create a uniform RDRL. Once these detector positions are known, the

detector array is used to measure the position of each leaf that results in a field edge at

detector 'j' as seen in Fig. 2-2B.

PROFILER 2TM

The procedure that follows describes measuring the minor leaf offsets for the X1

leaf bank; the procedure is repeated for the X2 leaf bank but with appropriate collimator

configurations. Three main steps were required to measure the minor leaf offsets with

the PROFILER 2TM and the RDRL method: device setup, detector offset correction, and

MLC measurement.

Step 1- Device setup: The collimator is rotated to 1800 and the PROFILER 2TM is

set on the treatment table at a source to surface distance, SSD, of 79 cm and a

corresponding source to detector distance (SDD) of 80 cm. The PROFILER 2TM's x and

y axes are then aligned with the collimator crosshair shadow such that the positive y-

axis of the PROFILER 2TM points toward the gantry.

This orientation places the y-axis column of 83 detectors perpendicular to the

direction of leaf movement. The orientation also directionally matches the ascending









numerical order of the detectors (1-83) with that of the leaves (1-40). This directional

match makes the on-screen evaluation easier during data collection with the PROFILER

2TM software. The same effect could have been achieved by rotating the PROFILER 2TM

instead of the collimator.

Next, the PROFILER 2TM is shifted by 2 mm in the direction of its negative y-axis.

This centers the crosshairs between the central y-axis detector and its immediate upper

neighbor. Since the projected leaf width at 80 cm from the source is 8 mm, this shift

locates two detectors in the projection of each leaf, as seen in Fig. 2-2B, and positions

the x-axis of the PROFILER 2TM in the projection of the reference leaf.

Small rotational errors between the crosshair and the field edge of the backup jaw

are tested by moving one of the MLC's backup jaws to 3 mm before the central axis. A

small consistent gap between the detector markers and the backup-jaws-field-edge

indicates a proper alignment, while a divergence between the two is corrected by

rotating the PROFILER 2TM.

Since a visual alignment of the array was inadequate for a precise MLC

calibration, an alignment to the LINAC's radiation coordinate system is also performed.

This is accomplished by measuring an alignment profile with the column of y-axis

detectors. Before measuring the alignment profile, a y-axis profile is measured with both

jaws opened. This open field profile is used to reference the slope of the alignment

profile. The jaw and leaf bank settings that were used for the open field are shown in

Table 2-1. The alignment profile measurement is taken with the leaves retracted and

with the backup jaw positioned at the central axis (CAX) so that the measured profile









lies within the jaw's penumbra. The jaw and leaf bank settings that were used to

measure the alignment profile for the X1 backup jaw are also shown in Table 2-1.

A tilt in the initial alignment profile indicates that the PROFILER 2TM was not

aligned with the backup jaw, as seen in Fig. 2-3. For this case, a slight manual rotation

of the PROFILER 2TM is made and the measurement is repeated, which minimizes the

misalignment. Corrective rotations and re-measurement of the alignment profile can be

performed until the profile tilt is minimized.

Step 2- Determining detector offset corrections: Although the PROFILER 2TM's

diode detectors are precisely attached to the circuit board, small inherent errors in their

fixed positions cause spikes and dips in the measured profiles when the measurements

are taken in a region of high dose gradient. This is apparent in the alignment profiles of

Fig. 2-3. A correction for these detector positions is necessary before accurate

measurements of the leaf edge positions can be performed. Detector position

corrections are determined as follows.

Immediately following the PROFILER 2TM setup, three y-axis profiles are

measured with the X1 backup jaws positioned at -1, 0, and +1 mm relative to the CAX.

To avoid mechanical hysteresis, the backup jaw is retracted in the same direction

before each measurement. The MLC and backup jaw positions that are used to

measure these three profiles are shown in Table 2-1, where reference line (RL) 30%,

RL 50%, and RL 70% correspond to the measurements taken with the X1 backup jaw

positioned at -1 mm, 0 mm, and +1 mm of the CAX, respectively. The three profiles lie

in the high dose gradient region at approximately 30%, 50% and 70% of the values









measured at the center of the open field, OF, profile of Table 2-1. These three

measurement positions were chosen because the penumbra within this region is linear.

The data from these measurements is transferred from the PROFILER 2TM

software to an Excel worksheet for analysis. The data obtained with each detector is

then normalized to the OF measurement using

MRL
NRL, J (2-1)
OF'

where NRLn,j is the normalized reference line measurement for backup jaw positions 'n'

(-1, 0, and +1 mm) and y-axis detectors 'j' (1 < j < 83); MRLn,j is the reference line

measurement for a single detector 'j'; OFj is the open field value for corresponding

detector 'j'. Linear regression on the three data pairs for each detector 'j' (NRL1 ,j,

NRL ,j, and NRL+ ,j) produces slopes, mj, and intercepts, bj, for each detector. A linear

interpolation is then used to obtain the 50% dose position, DPj, for each detector using

0.5-b
DP, (2-2)


This value is a linear measure of the jaw position that results in a field edge at each

detector.

The detector offset correction for each detector relative to a reference detector,

ref, is termed the relative detector offset, RDOj, and is calculated using

RDO, = DP, DP,, (2-3)

The calculation of the RDO values sets a common origin for each detector that allows

for a spatially unbiased measurement of MLC positions. Note that although two

detectors are in the projection of the reference leaf, either one can be used as the

reference detector.









Step 3- Determining relative leaf positions: To measure the leaf positions for an

MLC leaf bank, the procedure followed in the previous step is repeated using the MLC

leaf bank instead of the backup jaw. The collimator configurations used for the fields

MLC 30%, MLC 50% and MLC 70% are shown in Table 2-1, which correspond to

measurements taken with the X1 leaf bank positioned at -1 mm, 0 mm and +1 mm of

the CAX, respectively. The resulting profiles in the high dose gradient region are again

approximately 30%, 50% and 70% of the value measured at the center OF profile.

Since the array is aligned with the backup jaw in step 1, a tilt in these profiles indicates

that the MLC bank may be in need of calibration. It is possible to calculate the degree of

tilt in the reference line measurement and subsequently correct further measurements;

however, this increases the complexity of the algorithm and is not used.

The detector data is copied from the PROFILER 2TM software to the Excel

spreadsheet and is normalized to an open field measurement using

MMLC
NMLCj n, (2-4)
OFJ

where NMLCn,j is the normalized MLC measurements for MLC leaf bank positions 'n'

and y-axis detector 'j'; MMLCn,j is the leaf bank measurement for a single detector.

As in the DPj calculations, the 50% dose-position for each leaf and detector

combination, MLCPj, is linearly interpolated using

0.5 b
MLCP, (2-5)


where the y-intercept, bj, and slope, mj, for each detector is determined by using linear

regression on the three data pairs for each detector j. This value is a linear measure of

the leaf position that would result in a field edge at the corresponding detector.









Next, the leaf offset correction for each leaf and detector combination relative to

the reference leaf, ref, and reference detector is termed the relative leaf offset, RLOij,

and is calculated using

RLO,, = MLCPJ MLCPfe, (2-6)

The relative detector offsets are then removed from the relative leaf offsets to provide

the minor leaf offsets, MILO, using

MILOJ = RLO,,J RDO (2-7)

The MILOj values are averaged for the two detectors lying in the projection of each leaf

to provide MILOi. The MILO| values are then linearly scaled to the isocentric plane at a

100 cm distance from the source.

EPID

The PROFILER 2TM MILO results were checked using an EPID based version of

the RDRL method. With this approach, the EPID was used to image the fields RL 50%

and MLC 50%, shown in Table 2-1. The data density of the EPID made it unnecessary

to image the 30 and 70% RL and MLC fields.

Since the active field size and fixed SDD of the EPID limited its field of view to only

25 leaves, it was necessary to shift the EPID panel to fully measure a leaf bank. The

EPID was first positioned so that the projected edge of the backup jaw for the RL 50%

field falls at the center of the EPID in the in-plane direction. It was then shifted in the in-

plane direction to a position that allows the RL 50% and MLC 50% fields to be imaged

for leaves 1-25. After these images were taken, the EPID was shifted in the in-plane to

a position that allowed for the RL 50% and MLC 50% fields to be imaged for leaves 16-

40.









The RL 50% and MLC 50% fields were exported from the EPID, in tiff format, into

the MATLAB environment (The MathWorksTM, Natick, Massachusetts) for analysis. An

in-house edge detection algorithm was then used to locate the field edges for the

reference line, DPv, and each leaf, MLCPi,v, where 'i' is the leaf number and 'v' is the

pixel number. Once these are known, the pixel distance of each leaf end's position to

the reference line was calculated and converted to a spatial distance, SDi,v, using

mm
SD, =(MLC -DPJ 0.25 (2-8)
pixel)

where 0.25 mm/pixel was the inherent pixel gain. The reference leaf's spatial distance

was then subtracted from the spatial distance for each of the other leaves to determine

the minor leaf offsets using

MILO,, = SD,, SDre (2-9)

The MILO values from the two EPID positions were accepted based on the difference in

leaf positions for leaves, 16 25, and the published leaf reproducibility. These leaves

were chosen since they fall within the RL 50% and MLC 50% images taken for both of

the EPID positions described above. The EPID was not used to measure major leaf

offsets; its' purpose was to check the results obtained with the PROFILER 2TM

Measuring major leaf offsets

The method described below determines the position of the reference leaf pair,

and therefore the leaf bank, relative to previously measured baseline positions at three

MLC locations- retracted, CAX, and extended. The setup geometry for the array is the

same as was used during the minor leaf offset measurements; the field configurations

are similar. Note the projection of the center of the reference leaf for the measured leaf

bank, X1 or X2, lies along the x-axis of the array. Determination of the major leaf offsets









with the array involves three steps in addition to those required for determining the

minor leaf offsets, they are: off axis MLC measurement, array offset correction, and

baseline comparison.

Step 1- Off axis MLC measurement: The procedure for measuring the MLC at off-

axis positions is similar to Step 3 of measuring the minor leaf offsets. For the minor leaf

offsets, the MLC is measured at the CAX; therefore no additional measurements are

required at this position. For the major leaf offsets, retracted and extended reference

leaf positions are measured at 7.5 cm on the x-axis. These positions were chosen

because they correspond to the largest IMRT fields our clinic regularly uses (~20 cm).

Since the SDD of the array is 80 cm, these positions corresponded to the 6 cm x-axis

detectors on the array.

The retracted and extended reference leaf positions are measured with three MLC

geometries, three fields per geometry. The retracted reference leaf positions are

measured with symmetric MLC fields. The collimator configurations used for the

symmetric fields SYM 30%, SYM 50%, and SYM 70% are shown in Table 2-2, these

correspond to measurements taken with the reference leaf pair incrementally being

retracted over the -6 cm and +6 cm detectors. The resulting measurements in the high

dose gradient region are approximately 30%, 50%, and 70% of the value measured with

the OF field at the same detector.

The extended leaf positions are measured, in turn, by extending the X1 or X2 leaf

bank over the center of the array to an off-axis position of -7.5 or +7.5 cm, respectively.

The collimator configurations used to measure the extended X1 MLC fields X1 30%, X1

50%, and X1 70% are show in Table 2-2, these correspond to measurements taken with









the X1 leaf bank respectively positioned at -7.6 cm, -7.5 cm, and -7.4 cm. Once again,

the resulting measurements in the high dose gradient region are approximately 30%,

50%, and 70% of the value measured with the OF field at the same detector. Collimator

configurations for the extended X2 MLC fields follow the same pattern but for the +7.5

cm position.

The measured data are then copied from the PROFILER 2TM software and pasted

into the Excel worksheet. The worksheet contains the same linear-interpolation method,

as previously described in Step 3 of measuring the minor leaf offsets, for determining

the MLC position that results in a 50% open field value for the specific detector. These

positions are labeled MLCk,m where 'k' represents the -6 cm, 0 cm, or +6 cm x-axis

detector position and 'm' is the X1 or X2 reference leaf. Depending on the setup

precision of the array relative to the CAX, all of the normalized values from the three

measurements may have been either above or below the 50% open field value, if this

was the case then shifted collimator configurations were necessary.

Step 2- Array offset correction: The offset of the array relative to the radiation CAX

is determined using

AO (DP42,X1 +DP42,X2) (2-10)
2

where DP42,X1 and DP42,x2 are the respective locations of the backup jaws X1 and X2 as

determined by the detector array, which has an offset AO, the calculated midpoint

between jaws.

Each time the array was set up to measure the MLC, its position relative to the

CAX is slightly different from the previous measurement setup. This uncertainty does

not affect the MILO values since they are relative values. However, the setup









uncertainty does affect the major leaf offsets and is removed from the reference leaf

positions using

MALOk,m MLCk,m AO, (2-11)

where MALOk,m are the major leaf offset values for the X1 and X2 reference leaf.

Step 3- Baseline comparison: Before the baseline measurements are made,

proper calibration of the MLC should be verified. Therefore, the ideal time to establish

baseline values is during the LINAC acceptance-testing / commissioning phase and / or

following the annual LINAC QA. The initial set of major leaf offset measurements

establishes baseline values; subsequent measurements provide a comparison to the

baseline values using

BOk, = MALOk, baselinek,m (2-12)

where BOk is the baseline offset for x-axis detector position 'k' and reference leaf 'm'

and baselinek,m is the initial set of MALO values. Major leaf offsets that differ from the

baseline by more than a preset tolerance are then in need of calibration.

MLC calibration

Elekta MLCs have a leaf-offset gain of 14 units/mm (0.071 mm/unit). Multiplying

this gain by the minor leaf offsets and/or baseline offsets gives the adjustment values to

bring each quantity into tolerance. These values are then entered into the Elekta LINAC

software to complete the calibration for the minor and major leaf offsets. The calibration

process is repeated if the leaf positions are not within our target tolerance of +/- 0.3 mm

(+/-4 units), the published Elekta MLC reproducibility.37









Other MLC types

Multi leaf collimators with leaves of mixed width require a variation of the

described method. For example, an MLC that has leaf widths of 5 and 10 mm requires

the same setup as described for the Elekta MLC used in this work. However, instead of

two detectors being in the projection of each leaf, there are either two (for the 10 mm

leaves) or one (for the 5 mm leaves). The "backup" jaw and MLC movements need to

be altered due to sharper penumbras produced by the jaw and MLC. Multi leaf

collimators with leaf widths other than 5 or 10 mm require a different SDD for the

PROFILER 2TM. For example, an MLC with a 4 mm leaf width at isocenter would require

a 100 cm SDD and a 2 mm shift. This would locate one detector in the projection of

each leaf.

Results

Detector Offsets

The linear dose gradient measured across the field edge (defined as 50% of the

open field value) for each backup jaw was 15.2% per mm. The RDO values measured

using the X1 backup jaw are shown in Fig. 2-4A as a solid black line. This quantity was

measured 10 consecutive times to determine the short-term reproducibility. The

maximum difference between the measured values was 0.10 mm with an average

standard deviation of 0.01 mm. This level of reproducibility indicates that the PROFILER

2TM is stable for measurements over a short period of time.

There were three outliers in the RDO results in Fig. 2-4A, at -0.4 cm, 4.4 cm, and

10 cm on the y-axis, whose positions were approximately 1 mm off the nominal axis.

We speculated that the cause was a detector placement error in the off axis direction.

While placement errors of this magnitude are not a factor in the intended use of the









product, i.e. profile measurements, they do play a factor in measuring leaf positions.

Figure 2-4B shows measurements of the X1 backup jaw, the X2 backup jaw, and the

OF field. The mirrored behavior of the spikes at these three positions verified the

placement error speculation. The detectors at these three locations were replaced by

the manufacturer, which brought the RDO values closer to the population average. The

RDO values determined for the new detectors are shown in Fig. 2-4A as the dashed

line. In practice, the detector placement error is compensated by the correction

techniques described herein; ongoing replacement of outliers is not necessary.

MLC Measurement Comparison and Reproducibility

The average dose gradient across the MLC leaf ends, defined as 50% of the open

field value, was 13.5% per mm. Figure 2-5 displays the MILO values measured with

both the array and EPID. The MILO results for the devices matched each other well,

with a mean difference of 0.11 mm 0.09 mm.

We conducted both short and long-term reproducibility measurements using the

RDRL method. For the short-term measurements, ten consecutive MILO and MALO

measurements were made with the array over a two hour period. The maximum

observed difference between individual values was 0.22 mm with a mean standard

deviation of 0.07 mm. The long term reproducibility was studied by obtaining five sets of

MILO and MALO measurements over a period of 12 weeks. The maximum difference

found was 0.51 mm with a mean standard deviation of 0.09 mm; this is in reasonable

comparison to Elekta's stated reproducibility of ~ 0.3 mm. Both the short and long term

reproducibility was comparable to previously published values.37 Reproducibility was not

evaluated using the EPID. However, the MILO results for leaves 16-25 of the EPID

images (the leaves common to both sets of images for one leaf bank) provide some









quantification of its short term reproducibility. The largest difference observed between

MILO values for these leaves was 0.12 mm.

MLC Service Issues

Mechanical alterations may affect the optically-based MLC control system, which

in turn could invalidate the MLC calibration. Two examples are illustrated.

Example 1 Replacement of the primary Mylar mirror: EPID and array MILO

values were determined using the RDRL technique prior to replacement of the primary

Mylar mirror. Measurements were repeated using the same method after replacement

of the mirror. The post-replacement measurements indicated that the values had

changed by as much as 1 mm in the upper half of the leaf bank as shown in Fig. 2-6.

Example 2 CCD camera replacement: During LINAC maintenance, the CCD

camera that is used to control the MLC leaf positioning was replaced. After the

procedure, radiographic film was used to check the MLC leaf positions. No problems

were found with a simple visual check of the film. Afterward, however, unacceptable

passing rates were obtained during routine patient specific IMRT QA measurements. An

MLC calibration was therefore performed using the RDRL method with the array. Before

calibration the initial leaf spread of the X1 leaf bank was nearly 2 mm as shown in Fig.

2-7. The leaves were then calibrated, using the "MLC Calibration" method described

above, to within the manufacturer's specified tolerance. A second calibration iteration

was then performed to further tighten the spread.

Discussion

MLC Calibration Stability and QA

Our long-term reproducibility results indicated that the Elekta MLC was stable over

a period of 3 months. For an IMRT program that includes patient specific QA, monthly









MLC checks should be sufficient to ensure the quality of patient treatment. However, if

patient-specific IMRT QA is not performed, we feel that MLC QA should be done

weekly.

Our experience after the replacement of both the primary Mylar mirror and the

CCD camera indicates that Elekta leaf positions must be checked after any type of

maintenance that could affect the optical control system of the MLC.

Device Comparisons

The PROFILER 2TM and EPID each have unique advantages and disadvantages

when used with the RDRL method.

The primary advantage of the PROFILER 2TM approach is time efficiency.

Measuring the minor leaf offsets for both leaf banks takes an average of 30 minutes

with an additional 10 minutes required for the major leaf offsets. Entering the leaf

adjustments into the Elekta software and re-measuring the calibrated positions takes an

additional 30 minutes. However, the time requirement could be dramatically decreased

if the entire procedure were incorporated into the PROFILER 2TM software using a

series of automatically delivered step-and-shoot fields. This has been requested of the

manufacturer.

The PROFILER 2TM is a very stable measurement platform, as indicated by the

detector offset reproducibility. Therefore, deviations in the measured leaf positions are

the result of leaf positions rather than device instability. Other benefits of the array

include the ease of data analysis with compatible-user-friendly formats such as Excel.

The array can also be used to measure an entire MLC bank without repositioning

whereas the EPID requires a shift in position to perform the same function. It is also









flexible in the sense that it can be used to calibrate any MLC on the market by simply

setting it up at the correct SDD and correct in-plane axis position.

A disadvantage of the PROFILER 2TM is the need to measure fields at 1 mm

increments for the determination of the relative detector offsets and MLC leaf positions,

which increases the required measurement time. A second disadvantage is the inability

to measure an entire leaf bank at off axis positions; this could be overcome by moving

the array off axis. The manufacturer is incorporating a motorized table that is capable of

shifting the array up to () 20 cm off axis. This will allow leaf bank measurements over

the entire field.

The EPID based approach is also a viable means of calibrating an MLC with the

RDRL method. Two major advantages of the EPID are its integration with the LINAC

and its larger cross-plane field of view. The larger cross-plane field of view means that

only one field is required for the reference line and the MLC measurements.

Disadvantages exist for the EPID. A scarcity of commercial software necessitates

user-based code and associated software for data analysis, which makes it inefficient

with regard to time and resources. For rigorous EPID measurements, many time

consuming corrections are required to account for the rotation, tilt, and sag of the

amorphous silicon panel.33

Considering the advantages and disadvantages, the efficiency with which routine

MLC QA and calibration can be performed is comparable for the two devices. The major

advantage of the PROFILER 2TM lies in its compatibility with Excel for data analysis,

whereas the EPID requires user-based code that is generally beyond Excel import

capabilities.









Conclusion

The purpose of this study was to develop an efficient and quantitative method for

calibrating an Elekta MLC. The RDRL method measures leaf-end positions relative to a

radiation reference line defined by the MLC backup jaw. The PROFILER 2TM detector

array was used to implement the method and the results were verified with an EPID.

The results obtained with the two devices agree well with each other and reproducibility

values agree with previously published values. The Elekta leaf positioning accuracy was

found to be vulnerable to alterations of the MLC optical control environment. Therefore,

MLC QA should be performed after any component of the MLC optical control system is

disturbed. The RDRL calibration procedure with the PROFILER 2TM is efficient for a

number of reasons. Primary among these reasons is the ease of data handling using

widely available software such as Excel. Our method can be easily utilized by any

facility with access to a PROFILER 2TM










Table 2-1. List of collimator settings for PROFILER 2TM measurements of the X1 MLC
leaf bank at CAX for minor leaf offsets. Settings are based on the IEC 1217
convention.


MLC X1 (mm)
100
150
150
150
150
-1
0
1


MLC X2 (mm)
-100
-150
-150
-150
-150
-150
-150
-150


Collimator
Backup Jaw
X1 (mm)
100
0
-1
0
1
100
100
100


Table 2-2. List of additional collimator settings for PROFILER 2TM measurements of the
major leaf offsets. The fields listed are for measurement of the X1 reference
leaf at retracted, SYM fields, and extended, X1 fields; the CAX fields were
measured during the minor leaf offsets measurements. Settings are based on
the IEC 1217 convention.


MLC X1 (mm)
74
75
76
-76
-75
-74


MLC X2 (mm)
-74
-75
-76
-150
-150
-150


Collimator
Backup Jaw
X1 (mm)
100
100
100
100
100
100


Backup Jaw
X2 (mm)
-100
-100
-100
-100
-100
-100


Field name
OF
Alignment
RL 30%
RL 50%
RL 70%
MLC 30%
MLC 50%
MLC 70%


Backup Jaw
X2 (mm)
-100
-100
-100
-100
-100
-100
-100
-100


Jaw Y1
(mm)
200
200
200
200
200
200
200
200


Jaw Y2
(mm)
200
200
200
200
200
200
200
200


Field name
SYM 30%
SYM 50%
SYM 70%
X1 30%
X1 50%
X1 70%


Jaw Y1
(mm)
200
200
200
200
200
200


Jaw Y2
(mm)
200
200
200
200
200
200










Leafl MILO,





20 MALO -20



Leaf 40 40


Figure 2-1. Minor leaf offsets are defined as the spatial offset of each leaf in a leaf bank
relative to the reference leaf, leaf 20. Major leaf offsets are defined as the
distance of the reference leaf to the radiation center of the beam.




RDRL i






--.. reference x reference x
detector reference detector
RDO-1- leaf RDRL
RDRL
0-----


SA B

Figure 2-2. (A) The radiation defined reference line (RDRL) method requires precisely
known detector positions; small positional errors in the array's detectors can
disrupt this method by introducing false offsets into the leaf positions. The
detector offsets are corrected for, which effectively creates uniform detector
positions. (B) Two detectors are located in the projection of each leaf and
measure the position of the leaf end.






44














80




S40


20 -Open Field
... Initial Alignment
-Re-alignment
-15 -10 -5 0 5 10 15
Profiler 2 y-axis (cm)
Figure 2-3. The measurements required to radiographically align the array. The thick
solid line shows an open field measurement used for comparing the
alignment profiles; the dashed line shows the initial setup of the array, the thin
solid line shows a correct radiographic alignment with the backup jaw after
repositioning the array.





!-Before diode detector replacement
After dde detector replacement







S2-Open Field
X1 Backup Jaw
-X2 Backup Jaw
15 -10 -5 0 5 10 15 -15 -10 -5 0 5 10 15
Profler 2 y-axis (cm) A Profer 2 y-axis (cm) B
Figure 2-4. (A) Detector offsets relative to a reference detector. The reference detector
is chosen based on which detector lies in the projection of the reference leaf.
Detectors at -0.4 cm, 4.4 cm, and 10 cm displayed a larger than average
offset value as shown by the solid line. After detector replacement, the
dashed line shows nominal offsets. (B) Measurements with each backup jaw
located at the CAX position. The mirrored behavior of the three large spikes
on the profiles indicates they are due to positional errors.

















0







5 10 15 20 25 30 35 40
Leaf
Figure 2-5. MLC leaf offsets relative to leaf number 20, the reference leaf. Offsets were
measured using the RDRL method with two separate devices, the PROFILER
2TM and an EPID.






-Pre-Profiler 2
--Pre-EPID
-- Post-Profiler 2 ,, .
05 PostEPID

o '. f i





-.

5 10 15 20 25 30 35 40
Leaf
Figure 2-6. Change in relative MLC offsets after replacement of the primary MylarTM
mirror.















Fg r1e -a X l ,a .-a .""T"aL-r
7 %







--A rt-Ni ; nteral in
S5 10 1 20 .5 30 2! 40
Leaf
Figure 2-7. Calibration of the X1 MLC leaf bank. Measurements show relative MLC
positions before calibration and after two iterations of the RDRL method with
the PROFILER 2TM.









CHAPTER 3
CHARACTERIZATION OF A MULTI-AXIS IONIZATION CHAMBER ARRAY

Introduction

Radiation therapy is continually increasing in complexity as new treatment

modalities are adopted in the clinic. While these advances benefit tumor dose

localization, they also increase pressure on departmental resources as the new

modality is implemented. A prime example of this cause and effect is the adoption of

intensity modulated radiation therapy (IMRT) and the complications that it added to

treatment planning and quality assurance (QA) programs.1 The adoption of new

modalities (e.g., CyberKnifeTM, single-arc IMRT, and RenaissanceTM) will likely continue

this trend.

New measurement technologies promise to ease these complications in two ways.

They can decrease burdens that were introduced by the new treatment modality. They

can also improve the efficiency of established (but time consuming) measurements,

such as those taken with a scanning water tank.

The scanning water tank has been an integral part of the radiation therapy clinic

for decades, providing beam profiles, fractional depth doses (FDDs), and point

measurements. While advances in computer technology have improved their efficiency,

they still require large amounts of preparation time (~ 1 12 hours for setup and break

down) and actual scanning time [several days to measure linear accelerator (LINAC)

beam commissioning data].28 Infrequent use by the clinical physicist adds to these

inefficiencies and decreases the likelihood of obtaining quality data.

Just as detector arrays streamlined the IMRT QA process, they also promise to

increase the efficiency of collecting LINAC beam data. The time required to measure









beam profiles at multiple depths can be decreased from days to several hours. The skill

required to obtain quality data is also reduced. This is due to the fact that detector

panels are mechanically simpler than water tanks and are more familiar to clinical

physicists.

The purpose of this work was to extensively evaluate a detector array that has the

potential to simplify the acquisition of LINAC beam data. The device chosen for this

research was the IC PROFILERTM (Sun Nuclear Corporation, Melbourne, FL 32940),

which is a multi-axis ion chamber array. The potential of the IC PROFILERTM to be used

as an alternative to a scanning water tank necessitates a study of its behavior in a

radiation environment.

Materials and Methods

Materials

The detector array used in this study was the IC PROFILERTM [Sun Nuclear

Corporation (SNC), Melbourne, FL USA] and will henceforth be referred to as the panel.

The panel contains 251 parallel plate ion chambers, each with a volume of 0.05 cm3.

The active area of the panel is 32 x 32 cm2 (Fig. 3-1). A detector array is located on

each coordinate axis (x and y) and on the negative and positive diagonal axes (nd and

pd, respectively). The detector spacings are 0.5 cm and 0.71 cm on the coordinate and

diagonal axes, respectively. All axes share a common center detector, with the x- and

diagonal axes missing detectors immediately adjacent to the center detector. The

inherent thicknesses of buildup and backscatter are 0.9 g/cm2 and 2.3 g/cm2,

respectively.

Included with the panel is a personal computer user interface (UI) that provides

control over data acquisition in the form of adjustable real time updates (125 800 ms)









and amplifier gains for low repetition rates. Data can be acquired in pulsed or

continuous mode. Measurements in pulsed mode are synced with radiation pulses,

which are detected by a system of trigger diode detectors. Hence, this mode is only

intended for use with a pulsed radiation source. Measurements in continuous mode are

clock-synced and acquire data at a set update interval, making it ideal for a continuous

radiation source. Continuous mode can measure pulsed radiation but with the possibility

of measurement updates occurring during a pulse.

Characterization measurements were carried out on an Elekta SynergyTM (Elekta

Limited, Crawley, SXW UK) LINAC that was operated at X-ray energies of 6 and 18 MV.

Two additional radiation sources were used that provided differentiating characteristics.

These were a 60Co teletherapy unit (Eldorado 6TM, Atomic Energy of Canada Limited,

Mississauga, ON CA) and a Varian TrilogyTM LINAC (Varian Medical Systems, Palo

Alto, CA USA); their use will be acknowledged accordingly.

Methods

The UI provides graphical views of radiation profiles from all four detector axes

and profile analyses of the usual beam features (flatness, symmetry, beam center, etc.).

This investigation of the panels relative response to various conditions was carried out

in MATLAB (The MathWorksTM, Natick, MA USA), not the panels UI, which is

designed primarily for absolute measurements of beam features. This is possible since

measurements are saved in a raw ASCII format that allows post processing in the form

of background subtraction and detector sensitivity adjustments.

The raw data for each detector was processed to remove background counts, and

a temperature-pressure correction and detector specific calibration factors were applied.

The resulting value [corrected counts (CC)] is expressed with









C (RawCount -TimeTic bias ) (3-1)
gain

To save space, all subscripts, variables, and equations are defined in Table 3-1.

The data produced by Eq. 3-1 is identical to that from the panels UI.

Reproducibility

The panel was characterized in areas that have traditionally been of interest for

radiation detectors and arrays. The first item tested was measurement reproducibility.

The short and long term reproducibility of the panel was determined on the 60Co

teletherapy unit. This radiation source was chosen because of its reproducible nature.

The panel was irradiated with a field size larger than its active area. No additional

buildup was used.

The short term reproducibility was evaluated with 10 consecutive 1 minute

deliveries; long term reproducibility included 1 minute deliveries over the course of nine

months. Only the center 45 s of saved data from each delivery was used in an effort to

reduce the effects of source travel that occurs in 60Co teletherapy units.41 The

measurements were normalized to the central axis (CAX) detector and the

reproducibility was quantified by determining each detector's a) coefficient of variation

[(CV), relative standard deviation] and b) maximum deviation from the mean.

Dose and instantaneous dose rate dependence

The dose dependence of the panel was evaluated over LINAC deliveries ranging

from 1 to 1000 monitor units (MU). The measurements were evaluated at a source to

detector distance (SDD) of 100 cm; buildup was equivalent to the energy specific d-max

(1.5 and 3 g/cm2 for 6 and 18 MV X-rays, respectively). Simultaneous measurements

were made with an independent system (Farmer-type chamber model N30001, PTW,









Freiberg, DE; electrometer model 1010, SNC) in a 4 x 30 x 30 cm3 slab of water

equivalent material that was located below the panel. The dose definitions for this

machine were such that one MU (100 cm from the source, 10 g/cm2 buildup, 10 x 10

cm2 field) delivered 0.78 and 0.88 cGy for 6 and 18 MV X-rays, respectively.

The measured signals from both systems were normalized to the 200 MU

deliveries. The panel's response was then normalized to the Farmer-type chamber's

response using,

SMEAS (MU)
[MEAS (200MU)
Response (MU) = panel (3-2)
MEAS (200MU)
MEAS (200MU) Farmer
chamber

The panel's instantaneous dose rate (D) dependence was evaluated at a

constant LINAC pulse rate frequency (PRF), exposure, and buildup (400 Hz, 200 MU,

and 10 g/cm2, respectively). The dose rate was altered by changing the SDD from 90

cm to 150 cm, in 10 cm increments. This effectively changed the D in accordance with

the inverse square law. All measurements were repeated using a Farmer-type chamber

under similar conditions.

The methodology of Eq. 3-2 was used to evaluate the panel's D dependence,

Response (D). The variable D was used in place of MU and the detector systems were

normalized to the 100 and 110 cm SDD dose rates for 6 and 18 MV X-rays,

respectively.









PRF dependence

The panel's PRF dependence was evaluated at a constant field size, SDD, and

buildup (35 x 35 cm2, 100 cm, and energy specific d-max, respectively). The LINAC's

control system was used to vary the PRF from 50 to 400 Hz. Measurements were made

with two panel orientations that differed by a 1800 rotation. Simultaneous measurements

were made with an independent system (Farmer-type chamber model N30001; 1010

electrometer) located below the panel in a slab of water equivalent material.

Once again, the same methodology (Eq. 3-2) was used to evaluate the PRF

dependence [Respone (PRF)] of the panel's center detector. The variable used was PRF

instead of MU and the detector systems were normalized to the 400 Hz PRF.

The PRF dependence of the panel's off-axis detectors was determined by relating

the PRF response of each detector to that of the center detector. This is expressed as

CC (PRF)
CC(400Hz) (3-3)
OA_ Response, (PRF) = CC. (3-3)

CC (400Hz))

A detector PRF response that differs from the center detector will manifest itself as

a deviation from unity. This analysis is useful since detector-specific variables such as

calibration factors and air density corrections cancel, leaving hidden variables that may

exist in the electronics or as extra camera volume.

An actual LINAC PRF variation might be interpreted incorrectly as panel PRF

dependence. This erroneous interpretation can be avoided by taking the difference

between the OA _Response values for the 180 and 0 panel orientations. This is

expressed as









DOA Responsepo (PRF, po)
(3-4)
(OA Response (1800)-OA Response (0))

Reflective symmetry in this value would indicate PRF dependence in the mirror

detector pair, i.e. the detectors that mirror each other about the array center.

Energy dependence

The panel's dependence on X-ray energy was evaluated using two methods. First,

a Farmer-type chamber and electrometer (model FC65-G, IBA, Schwarzenbruck, DE;

model K602, CNMC Company, Nashville, TN USA) with National Institute of Standards

and Technology traceable calibrations were used to measure 200 MU deliveries at X-

ray energies of 6 and 18 MV. The delivered dose was determined in accordance with

the American Association of Physicists in Medicine report by task group (TG) 51,42 the

source-to-axis distance setup was used. Equivalent measurements were made with the

panel and an appropriate amount of water equivalent material. The panel data was not

processed beyond the extent of Eq. 3-1; e.g. no Pion was applied.

The methodology of Eq. 3-2 was also used to evaluate the panel's energy

dependence, Response(E).The variable was E instead of MU and the detector systems

were normalized to the 6 MV X-ray deliveries.

Spectral differences that occur for different field sizes and with increasing buildup

could result in a panel energy response. The second method compared FDD curves for

6 and 18 MV X-rays and various square field sizes (5, 10 and 20 cm). The reference

FDDs were acquired with a Blue PhantomTM (IBA) scanning water tank equipped with

CC13TM (IBA) ion chambers for the reference and field detectors. The panel's FDD was









acquired with water equivalent material ranging from 0.9 to 30 cm. A 90 cm SSD was

used for both systems.

The FDDs were normalized to the 10 cm buildup measurement. The energy

response of the panel was then quantified by taking the ratio of the panel's FDD

measurement to that of the scanning water tank. This relationship is expressed as


Response(buildup) = FDDC t (3-5)
FDDc1 )bd

It was assumed that the CC13TM had a flat energy response.

Response to power being applied to the electronics

The time required for the temperature of the panel's electronics to reach

equilibrium upon power up was determined using the 60Co teletherapy unit. The panel

was allowed to equilibrate to room temperature overnight. In the morning, power was

applied to the panel and a background measurement was immediately taken. The

source was then extended into position and left for three hours, during which time

periodic measurements were made.

Since the source was not moved between measurements, any change in

measured signal was attributable to drift after power was applied. This change was

quantified by normalizing the mean detector value for each measurement to that of the

measurement at t = 180 minutes. This is expressed as


Response(t) = (t) (3-6)
CC(180min)

The 180 minute measurement was used as a normalization point since the panel's

electronics were assumed to be in equilibrium by this time.









Calibration constancy

The relative sensitivities of the panel's detectors were determined using a

proprietary calibration method that has been previously described and evaluated.25' 4345

While the calibration method for the panel is the same, the device is different, which

could affect the calibration quality. Therefore, calibration factors were evaluated for

reproducibility, accuracy, energy dependence, and buildup dependency.

Letourneau et al. described this calibration method's sensitivity to small changes in

the LINAC dose distribution. They reported that when the symmetry stability was

degraded from 0.1% to 0.3%, the calibration reproducibility degraded from () 0.8% to

() 1.2%. The symmetry stability for the Varian TrilogyTM was () 0.05% which was

slightly better than the Elekta SynergyTM at () 0.15%. Therefore, we used the Varian

unit to calibrate the panel.

The short term reproducibility was evaluated by calibrating the panel five

consecutive times over 2 hours using both 6 and 18 MV X-rays. The long term

reproducibility was evaluated over the course of six months. The calibration

reproducibility was quantified by determining each detector's a) CV and b) maximum

deviation from its mean.

The accuracy of the calibration was evaluated by taking two full-field

measurements with a 1800 panel rotation between measurements. The 180-

measurement data was rotated about the array center to maintain the LINAC coordinate

system of the 0 measurement. The calibration accuracy was quantified by taking the

ratio of the measurements. This is expressed as

SNMEAS (180)
CF Accuracy (po) =, (00 (3-7)
pos NiAAS4 (00)
pos\








An accurate calibration would result in minimal differences between the two

measurements since a measurement at a point in space should be detector invariant if

the detector's response is known by calibration.

The calibration factors' energy dependencies were evaluated by calculating the

ratio of the 18 MV calibration factors relative to the 6 MV calibration factors. This is

expressed as

CF, (18MV)
CF Response, (E) = F 8MV. (3-8)
CF, (6MV)

The calibration factors' buildup dependencies were evaluated by taking the ratio of

calibration factors determined with different amounts of buildup (0.9, 2.9, and 4.9 cm)

relative to the calibration done with 4.9 cm of buildup. This is expressed as

CF Response, (buildup) = C (buildu (3-9)
CF,(4.9 cm)

Backscatter dependence

The panel's dependence on backscatter was evaluated for a variety of square field

sizes, buildup amounts, and backscatter amounts [5, 10, 20, and 30 cm; an energy

specific d-max and 10 g/cm2; 2.3 (the inherent backscatter) to 19.1 g/cm2]. The effect of

increasing backscatter on profile shape was quantified by relating the backscatter

response of each detector to that of the center detector. This is expressed as

CC (bs)
CC (inherent)
OA Response, (bs) = CC(iere (3-10)
CC (bs)nt
CC (inherent)









Beam profile measurements and output factors

Beam profiles and FDDs were measured for field sizes ranging from 5 x 5 cm2 to

30 x 30 cm2. The SSD of the panel plus buildup was kept constant at 90 cm. These

measurements were compared to scanning water tank data (CC13TM) that were

collected under a similar geometry.

The profile accuracy was quantified by taking the ratio of the panel's measurement

to that of the CC13TM for two buildup values (the energy specific d-max and 10 g/cm2).

This is expressed as


PA (.fs) = MEASpane (3-11)
o .NMVEASCC13 pos,fs


The agreement between the FDDs produced by the two systems was previously

quantified [see Eq. (3-5)].

Output factors were measured for symmetric fields ranging in size from 1 x 1 cm2

to 30 x 30 cm2. The measurements were done with the panel at an SDD of 100 cm with

10 cm of buildup. These were checked using a one dimensional scanning water tank

and three different detectors: a Farmer-type chamber (model N30001, PTW), a CC13TM

scanning ion chamber, and the EdgeTM diode detector (SNC).

These detectors are applicable over a wide range of field sizes so as to provide

overlapping data and to accurately extend the results to both small and large field sizes.

Specifically, the Edge detector was used only for field sizes of up to 10 x 10 cm2 due to

the energy dependence of diode detectors.46 Due to the Farmer-type chamber's size,

it's use was limited to field sizes from 5 x 5 cm2 to 30 x 30 cm2. The CC13 was used

over the entire range of field sizes.








The accuracy of the panel's output factors was quantified through normalization to

the independent system's output factors. This is expressed as

= K Output Factor
NOF (fs) Output Facto (3-12)
Output Factor

Results and Discussion

Reproducibility

The panel's short and long term reproducibility are presented in Table 3-2. For 10

consecutive measurements in which the 60Co source was left in the extended position,

the maximum deviation from the mean was 0.15%. For long term measurements over

the space of nine months, the maximum deviation from the mean was 0.84%.

Dose and Instantaneous Dose Rate Dependence
In Fig. 3-2 we see the dose response of the panel normalized to that of the

Farmer-type chamber. A non-linear relationship was observed for the panel in pulsed

mode. During the first MU of delivery, the panel under responded by nearly 50% relative

to the Farmer-type chamber. As the delivered dose was increased, the response of the

panel converged to that of the Farmer-type chamber.

In contrast, a linear relationship was observed when the panel was operated in

continuous mode. The responses of the panel and Farmer-type chamber were within ()

0.6% of each other for lower MU deliveries (< 20 MU). For larger deliveries, the

agreement was significantly tighter.

The likely cause of the non-linear response observed in pulsed mode is the

panel's measurement trigger system. This system of trigger detectors initiates a

measurement only when a detected signal crosses a preset threshold. If the initial beam









intensity is low enough, then the measurement will not be triggered. Only pulses that

caused a response below the trigger threshold would be missed.

This theory was tested by creating a hypothetical panel response in which only a

fraction (f) of the first MU was missed. After n MU, the ratio of pulsed to Farmer-type

chamber measurements,

',nr f-/}/( nn
200 f 200

was plotted with a value f = 0.45, which was close to the under response observed with

the panel. As shown in Fig. 3-2, the hypothetical panel response closely matches the

actual panel response in pulsed mode. This information was provided to the

manufacturer and a solution to the trigger logic has been implemented.

An important question to ask is: why use pulsed mode? Pulse syncing has

advantages in machine diagnostics where only integral LINAC pulses have been

accumulated for each data frame. For example, a PRF of 60 Hz and a measurement

interval of 150 ms will result in 9 LINAC pulses being accumulated in each frame. This

allows studies of pulse stability, dose per pulse, etc.

The LINAC's nominal D (100 cm SDD, 10 cm of buildup, 10 x 10cm2 field) were

6.5 and 8.3 cGy/s for 6 and 18 MV X-ray beams, respectively. The dose rate was

altered by changing the SDD from 90 to 150 cm in 10 cm increments. This correlated to

instantaneous dose rate ranges of 8.0 to 3.0 and 10.3 to 3.7 cGy/s for 6 and 18 MV X-

ray beams, respectively. The ranges were determined by applying an inverse square

correction to the nominal D.









In Fig. 3-3 we see the instantaneous dose rate response of the panel normalized

to the response of the Farmer-type chamber. The panel over responded, relative to the

Farmer-type chamber, as the dose rate decreased. This effect reached a maximum of

0.9% at the lowest 6 MV dose rate.

The cause of the panel's over response is not entirely clear. A possible source of

the over response is a higher rate of volume recombination in the panel's ion chambers

at the higher irradiation rates.47' 48 In ref 48, Boag expresses the collection efficiency 'f'

in pulsed radiation beams as

f = -(e -1),

where v is proportional to p, the charge collected per pulse. In this equation, we can

vary p to see the effects on f. Differentiation of f and reordering with p results in the

equation

df =dp (-f ev).
f P

In regions of small changes, we can substitute df ~ f2 fl and dp ~ p2 p1 to get:


f i P 2- v ))

From our measurement of f by varying the panel's chamber bias voltage at a given p,42

we found f to be approximately 0.995, which estimates v at 0.01. Therefore, a change in

the dose per pulse by a factor of 2.8 (by varying the SSD from 150 to 90 cm) results in

an expected change in efficiency of 1 + (2.8 1)*(-0.005) = 0.991. From Fig. 3-3 and

relative to the Farmer-type chamber, we see f2/fl = 0.998/1.01 = 0.988. This is in

reasonable agreement with the theory estimate of 0.991, considering the measurement

uncertainty.









Other possible contributors to the response in this particular setup could be slight

changes in the scatter energy as well as different re-combination in small extra-cameral

volume regions. Regardless, the observed response is small and can probably be

ignored for the irradiation rates investigated. Measurements outside of these dose rates

(an un-flattened LINAC beam) would require further investigations.

PRF Dependence

In Fig. 3-4 we see the PRF response of the panel relative to the Farmer-type

chamber. The panel's response increased with a decrease in PRF. This was true for

continuous and pulsed measurement modes. For the 18 MV X-ray beam, the 50 Hz

PRF was unavailable due to a LINAC symmetry interlock that would terminate the

beam.

The increased response in continuous mode was the result of fewer radiation

pulses occurring while the device was performing a measurement update. As a result,

the number of missed pulses decreased and the signal increased. It is important to note

that this is not a fault with continuous mode since this mode is not intended for use with

a pulsed radiation source.

The panel's over-response in continuous mode can be lowered by using a

collection interval longer than 125 ms (the default). This effectively reduces the number

of measurement updates, and hence the number of pulses that occur, during

measurement updates. We chose a collection interval of 500 ms to demonstrate this

effect. In Fig. 3-4 we see that the PRF responses obtained with the longer collection

intervals closely matched those obtained with the Farmer-type chamber.

The signal increase observed in pulsed mode was related to the measurement

trigger error previously discussed. As the PRF was decreased, the total delivery time









increased which gave the measurement-trigger-system more time to detect a pulse and

initiate a measurement. The result was fewer missed pulses and an increase in signal.

The PRF response of the x-axis detectors relative to the center detector's

response is shown in Fig. 3-5A. Three distinct features are apparent. First, there was a

singular spike at the (-) 4 cm detector. This detector had an over response that

increased to 4% as the PRF was decreased to 50 Hz. The second feature was a global

PRF response that resembled a sine function. The magnitude of this response had a

maximum value of () 1%. The final feature was a global response in which curve

separation increased at lower PRF.

It is possible for the LINAC to exhibit a PRF variation. For the analysis method that

was used in Fig. 3-5A it was impossible to distinguish between a LINAC and a panel

PRF response. Rotating the panel by 1800 and repeating the experiment helped isolate

the source.

In Fig. 3-5B, we see the results for the 1800 measurement. The data was rotated

to maintain the LINAC coordinate system. In comparing Figs. 3-5A and B, it is apparent

that the singular spike stayed with the (-) 4 cm detector [now in the (+) 4 cm position

due to the data rotation]. This indicates the detector response was dependent on PRF.

We also see that the sine shaped response stayed true to the LINAC coordinate

system, which indicates it is due to the LINAC.

The detector PRF responses were visually amplified by taking the difference

between the OA_Response(PRF) values for the two panel orientations, Fig. 3-6A. This

analysis method eliminated any response due to the LINAC (the sine feature) while

preserving the other two PRF features. The singular detector response is present at the









() 4 cm positions because each is a function of both mirror detectors. Singular

responses (spikes) were also observed for the (+) 0.5 cm y-axis and (-) 1.4 cm nd-axis

detectors. These singular detector responses have been reported to the manufacturer

and are under further investigation. The global PRF response was relatively small,

typically less than () 1% for the tested PRFs.

In Fig. 3-6B we see the 6 MV DOA_Response(PRF,po) values for the x-axis

detectors when the panel was operated in continuous mode. The global PRF response

is similar to what was observed in pulsed mode. Interestingly, the PRF response at the

() 4 cm detector pair is not present. The cause for this discrepancy between

measurement modes has been reported to the manufacturer and is under further

investigation. The responses for the remaining axes and energies were similar to those

values presented in Figs. 3-6A and B.

Energy Dependence

Increasing the X-ray energy from 6 to 18 MV decreased the panel's response by

0.47 0.02% relative to that of the Farmer-type chamber. This demonstrates that the

panel has a low level of energy dependence relative to a Farmer-type chamber that is

used for clinical reference dosimetry.

The energy response was also quantified by taking the ratio of panel FDD

measurements relative to those of a CC13TM ion chamber. The results for the 6 and 18

MV photon beams are shown in Fig. 3-7. The dashed-vertical-bold-lines represent the

energy specific d-max values.

The maximum deviation relative to the CC13TM occurred for a 20 x 20 cm2 6 MV

X-ray beam. For this beam and field size, the panel under responded by approximately

2% at d-max. Typically, the two systems were within () 1% of each other. While the









agreement in the buildup region was poor, this area is unpredictable due to the lack of

electronic equilibrium and the difference in volume averaging that occurs between the

panel and CC13TM.

Larger field sizes (30 x 30 cm2) were not included because of increased

measurement noise due to electronic cross talk. This was a result of increased scatter

entering the panel's electronics due to the larger field size (at increased SDDs) and

increased secondary scatter from the buildup. One should be aware of this possibility

when measuring larger field sizes in conjunction with a substantial amount of buildup.

Response to Power Applied to the Electronics

While the panel never completely reached equilibrium, it did quickly reach a state

of minimal change. Two minutes after initial power up, the response was within 1% of

the 180 minute measurement; after ten minutes, the response was within 0.3%. This

effect can be minimized by powering up the panel at the beginning of a measurement

setup or by leaving it powered up at all times.

Calibration Constancy

The short and long term reproducibility of the detector's calibration factors are

shown in Table 3-3. The maximum deviations from the mean were 0.58 and 0.68% for 6

and 18 MV X-rays, respectively.

The accuracy of the calibration factors was evaluated by comparing

measurements from two panel orientations (0 and 1800). Correcting the data order for

the rotation in the 1800 data and assuming a reproducible LINAC beam, perfect

calibration factors should provide a ratio of unity for all positions. In Fig. 3-8 we see that

all of the calibration factors were within () 0.8% of unity. The symmetric error that

occurred about the y- and pd-axes indicates a calibration error of less than 0.8%.









The energy response of the panel's calibration factors is shown in Fig. 3-9. The 18

MV calibration factors were reproducibly larger (up to 1.5%) than the 6 MV calibration

factors. To obtain the most accurate profile measurements, separate calibrations at

each energy are recommended.

The response of the panel's calibration factors to additional buildup is presented in

Fig. 3-10. While the range of buildup used was relatively small (0.9 to 4.9 cm), it

represented a significant range in beam quality and electronic equilibrium. The majority

of the detectors (>97%) were within the reported calibration reproducibility. This

indicates that a single calibration file can be used for a range of buildup values.

However, calibrating with at least a buildup amount that is equivalent to d-max is

advisable to avoid including a response from collimator electron contamination.

Backscatter Dependence

The effect that increasing backscatter has on profile shape is presented in Fig. 3-

11. Only the 20 x 20 cm2 field is shown. A limited number of backscatter amounts were

chosen that reflect the behavior of the panel.

In Fig. 3-11A we see the 6 MV backscatter response for the negative x-axis; the

positive x-axis behaved similarly. The response within the beam appears relatively

uniform. Outside the beam, the increase in measured signal is more pronounced, rising

by up to 50%. This represents a small quantitative increase due to the lower initial value

of the signal. An expanded view of the panel within the beam is presented as an inset in

Fig. 3-11A. The addition of backscatter decreased the panel's measured signal by a

maximum of 0.8% over 80% of the field width.

The 18 MV results are presented in Fig. 3-11B. The out-of-beam increase is

smaller than was observed with the 6 MV beam. In the inset of Fig. 3-11B we see an









expanded view of the in-beam portion of the negative x-axis. Again, the change in

profile shape was reduced relative to the 6 MV beam.

For all energy, backscatter, field size, buildup, and array combinations, the profile

shape changed by less than 1% within 80% of the field width. Two exceptions occurred.

The 6 MV 30 x 30 cm2 beams resulted in backscatter responses of up to 1.6%. The 18

MV 5 x 5 cm2 beam resulted in a backscatter response of up to 1.5%.

Beam Profile Measurements and Output Factors

In Fig. 3-12 we see a sample of cross-plane profiles measured by the panel and

the CC13TM for a 6 MV beam. The buildup for these measurements was 10 g/cm2; no

additional backscatter was used for beam profile measurements because of the panel's

minimal backscatter response.

Taking the ratio between the two detectors' systems provided a clearer indication

as to their agreement, Figs. 3-13A D. The ratios covered 80% of the field width. The

total spread in error between the two systems was typically on the order of 1.5% and

was biased towards positive errors.

The cause of the positive bias was a combination of error in the center detector's

calibration factor and the measurements. Since there is only one normalization point,

error in it can skew the entire measurement. If instead the panel and CC13TM

measurements were normalized to the mean value over 80% of the field width, a much

tighter agreement was observed between the panel and the water tank results.

In Fig. 3-14 we see a visual representation of the 6 MV FDD for a 10 x 10 cm2

field. The agreement between the two detector systems was previously presented (see

Fig. 3-7). Their agreement was generally within () 1%. The exception occurred for

shallow regions of the 6 MV 20 x 20 cm2 field. For this field size and energy combination









the panel under responded relative to the CC13TM by approximately 2% at d-max. By 5

cm of buildup, the two systems are within () 1%. The data were collected for field sizes

of up to 20 x 20 cm2. Radiation contributions to the panel's electronics prohibited FDD

measurements for larger field sizes.

Output factors for the 6 and 18 MV X-ray beams are presented in Figs. 3-15A and

B, respectively. There were some deviations between the panel and the secondary

detectors for field sizes below 5 x 5 cm2. The ratio of the panel's output factors to those

of the independent detectors quantified the agreement between the two systems, Figs.

3-15C and D. The agreement between the panel and the secondary detectors was

within () 1% for the specified field sizes, with the exception of the 1 x 1 cm2 and the 2 x

2 cm2 field sizes, which exceeded 1%.

Conclusion

The purpose of this study was to extensively evaluate a detector array (IC

PROFILERTM) that has the potential to simplify the acquisition of LINAC beam data. We

have shown that the IC PROFILERTM generally behaved within () 1% of an

independent (or nominal) response in all areas tested. One primary exception occurred.

The response of the IC PROFILERTM to dose was non-linear when the device was

operated in pulsed measurement mode. The problem occurred because the

measurement trigger was not initially activated and pulses were missed. However, after

the measurement was triggered all radiation was acquired. A solution to the trigger logic

has been implemented by the manufacturer.

The ability of the IC PROFILERTM to accurately reproduce water tank profiles,

FDDs, and output factors is an indication of its abilities as a water tank alternative. The

primary benefit of using the IC PROFILERTM versus a scanning water tank is time









reduction. The same measurements (including device setup and breakdown) for both

systems took 180 minutes with the water tank versus 30 minutes with the IC

PROFILERTM. The time savings increases as the measurement load is increased. This

work evaluated the IC PROFILERTM and demonstrated that it is capable of efficiently

providing water tank equivalent scans.










Table 3-1. List of subscripts,
Subscript
ctr
i


pos

Variable
bs
buildup
CC
CF

CV
D
E
FDD
fs
gain
bias
MEAS
MU
NMEAS
PTP
po
PRF
RawCount
t
Time Tic

Eq. Name
3-1 CC
3-2 Response(MU)

Response(D)

Response(PRF)

3-3 OA_ Response(PRF)

3-4 DOA_ Response(PRF,po)

Response(E)


3-5
3-6
3-7
3-8
3-9
3-10


Response(buildup)
Response(t)
CF_ Accuracy(po)
CF_ Response(E)
CF_ Response(buildup)
OA_ Response(bs)


3-11 PA(fs)
3-12 NOF(fs)


variables, and equations
Description
the panel's center detector
panel detector (i: x-, y-, pd-, and nd-axes)
radiation detector (j: panel, Farmer-type chamber, CC13TM,
EdgeTM diode detector)
position on the LINAC coordinate system

Description
total backscatter (cm)
total buildup (cm)
mean corrected counts
calibration factor that normalizes the relative sensitivities of the
panel's detectors
coefficient of variation (relative standard deviation)
instantaneous dose rate (cGy/s)
X-ray energy
fractional depth dose
side of a square field
ion chamber amplifier gain
detector leakage rate in RawCounts per TimeTic
measured value of the detector system (nC or CC)
monitor units
profile measurement normalized to the central axis (CAX)
pressure and temperature correction
panel orientation (0 or 180)
pulse rate frequency (Hz)
raw counts measured by the panel detectors
elapsed time since power was applied to the panel
elapsed time since starting the panel measurement

Description
corrected counts of the panel
the panel's dose response relative to a Farmer-type chamber's
dose response
the panel's instantaneous dose rate response relative to a Farmer-
type chamber's instantaneous dose rate response
the panel's PRF response relative to a Farmer-type chamber's
PRF response
PRF response of off-axis detectors relative to the center detector's
PRF response
Difference between OA_Response(PRF,180) and
OA_Response(PRF,0O)
response of the panel relative to TG-51 based Farmer-type
chamber measurements for 6 and 18 MV X-rays
FDD agreement between the panel and CC13TM
response of the panel to power being applied to the electronics
accuracy of the detector's calibration factors
energy response of the detector's calibration factors
buildup response of the detector's calibration factors
backscatter response of off-axis detectors relative to the center
detector's backscatter response
profile agreement between the panel and CC13TM
the panel's output factors normalized to independent detectors











Table 3-2. The panel's short and long-term reproducibility were evaluated on a 60Co
teletherapy unit
Short term Long term
Mean CV(%) 0.05 0.24
Maximum CV(%) 0.11 0.50
Maximum deviation (%) from the mean 0.15 0.84

Table 3-3. The short and long term reproducibility of the relative detector calibration
factors.


Mean CV (%)
Maximum CV (%)
Maximum deviation from the mean (%)


6 MV
Short term Long term
0.12 0.09
0.35 0.54


0.58


0.57


18 MV
Short term Long term
0.08 0.20
0.18 0.59
0.20 0.68



























Figure 3-1. Overlay of the IC PROFILERTM (panel) showing the multiple detector axes.


o.A

0
0.


100


1000


MU
Figure 3-2. The panel's dose response relative to the Farmer-type chamber's dose
response.


-6 MV pulsed
-6 MV continuous
-18 MV pulsed
--.18 MV continuous


111 11 II II II II II II II II II II II II II II II II II II II II II II II II 11 111 11 II 11 11111 111






uarr In
E~7f~(~d~,~,u~?;f~'


0.5L
1










1.01v


*Q

q)
0u



o:


,


-6 MV
--18 MV


1.01


1.005


0.995t


2 4. 6
D (cGy/s)


Figure 3-3. The panel's instantaneous
chamber's.


8 10 12


dose rate response relative to the Farmer-type


G ****500 ms, 6 MV, continuous

QI)
( 1 1.005






o.995
0 100 200 300 400
PRF (Hz)
Figure 3-4. The panel's PRF response relative to the Farmer-type chamber's PRF
response. The measurement update interval was the default (125 ms) for all
measurements expect one, which was set to 500 ms.













C. 1.(

v,




0'


-10 0 10 -10 0 10
x-axis (cm) A x-axis (cm) B
Figure 3-5. The off-axis PRF response for the x-axis detectors relative to the center
detector for panel orientations (A) 0 and (B) 1800. The panel was operated in
pulsed mode for each panel orientation. The data for panel orientation 1800
was rotated to maintain the 0 LINAC coordinate system.



0.05 0.05
****50 Hz 50 Hz
S0.04 -100 Hz 0 0.04 -100 Hz
.3 --200 Hz C --200 Hz
0 002 002



1-0,0 1-,001 ...... ------------........ ..... ........
IXI




Figure 3-6. The difference between the 1800 and Oo OAResponse(PRF) values for the

panel operated in (A) pulsed mode and (B) continuous mode. The cause of
the spike at the () 4 cm positions in pulsed mode (or its absence in
continuous mode) is not entirely understood and is under continued
investigation by the manufacturer.



1.12
1.1 5 x 5 cm2
0. 1.08 -10 x 10 cm2

-o i i.E >
S1.06 x20
1 .



1 02 6 MV
----------------------- --------
0 98--
0.09 10 20 30
buildup (cm)
Figure 3-7. The energy response of the panel's center detector presented as a ratio of
the panel's FDD to the CC13'sTM FDD.














1.015






O -6 MV
18 MV
0.99
x-axis y-axis pd-axis nd-axis

Figure 3-8. The accuracy of the calibration factors.



1.02

2 1.015-
C/-
S1.01

& 1 005




0.995-
x-axis y-axis pd-axis nd-axis

Figure 3-9. The energy response of the calibration factors.



1.005 .


S0.995
S0.99 6 MV
-2 0.9 cm

S1.00 =5 18 MV


S0995 ii

x-axis y-axis pd-axis nd-axis

Figure 3-10. The buildup response of the calibration factors for 6 and 18 MV X-rays.




























Figure 3-11. Off axis backscatter response of the x-axis detectors for (A) a 6 MV 20 x
20 cm2 field and (B) an 18 MV 20 x 20 cm2 field. Buildup was equivalent to d-
max.





12*0 *CC13
-Panel
100


40




20 .......... ..............

-10 0 10 20
cross-plane (cm)
Figure 3-12. Normalized cross-plane measurements with a CC13TM and the panel. The
beam energy was 6 MV and the buildup was 10 g/cm2.


x-axis (cm)


x-axis (cm)














1 02 10 g/cm2

1 01


099 -
09 5 x 5 cm2
-10 x 10 m2
.20 x 20 m2 1.5 g/cm2
102 K 30 r
r "


.15 -10 -5 0 5
cross-plane (cm)


10 15


1.02 10 g/cm2




099 5 x 5 cm2
-10 x10 m10 rm
-20 x20m 3 g/cm2
102- 3r.


1 ......-------
0.99 .


-15 -10 -S 0 5
cross-plane (cm)


10 15


1.02

1.01






102
0.99





A


0l






C


15 -10 -5 0 5
in-plane (cm)


I .u
1.02 10 g/cm2





10 x10 x 10 rem
-.20 x 20 cm2 3 g/cm2
102



099


-15 -10 -S 0 5
in-plane (cm)


Figure 3-13. Profile agreement (over 80% of the field width) between the panel and
CC13TM for 6 MV X-ray (A) cross-plane measurements and (B) in-plane
measurements and 18 MV X-ray (C) cross-plane measurements and (D) in-
plane measurements.


buildup (cm)
Figure 3-14. FDDs for a 6 MV 10 x 10 cm2 field.


10 g/cm2





..5 x5cm2
.0 x 10 x 10 m
-.20x 20 m2 1.5 g/cm2
'- 30.' m
'~O~On


10 15


10 15











1.2-



1.1


I L
S0.9-


O 0.8-

0.7
-


--Panel
-*Edge
-"CC13
--Farmer


0 5 10 15
fs (cm)


20 25 30


1. 1




o0.9

S0.


0.7
0


i ..Edge
--CC13
--Farmer


5 10 15
fs (cm)


20 25 30


1.0151


- Edge
-CC13
--Farmer


5 10 15 20 25 30
fs (cm)


-1.005

LL 1

0.995

0.99


n


-. Edge
-CC13
--Farmer


5 10 15 20 25 30
fs (cm)


0
0


Figure 3-15. Output factors measured with the panel's center chamber and three
independent detectors for a (A) 6 MV and (B) 18 MV X-ray beam. The panel
output factors were normalized to each independent detector for the (C) 6 MV
and (D) 18 MV X-ray beams.


1.015.

1.01

.1.005

1

0.995-


n


0
0


nKori


nKori


---'









CHAPTER 4
WIDE FIELD ARRAY CALIBRATION DEPENDENCE ON THE STABILITY OF
MEASURED DOSE DISTRIBUTIONS

Introduction

Detector arrays have become an integral tool in radiation oncology by efficiently

providing dosimetric information on which clinical decisions are often made. By

definition, an array contains multiple detectors, underlying the importance that each one

has an equivalent sensitivity to radiation. A violation of this requirement will

consequently bias the measurement and possibly the clinical decision.

Since some variations in the detectors' sensitivities are inevitable, calibration

techniques have been developed to compensate them.43'49, 50 One such method, known

as wide field (WF) calibration,43 has become widely used by the medical physics

community.25, 45, 51'52 The basis of the WF calibration protocol is determining the relative

detector sensitivities through a response normalization that is determined by detector

substitutions at given field locations.

Accompanying an array calibration is a responsibility to verify that the calibration

results do not significantly alter the appearance of the true beam shape. Verification of

the array calibration can be accomplished with a comparison to scans in water, but the

equipment setup is tedious and avoided in many circumstances. Instead, many users

rely on a visual examination of measured profiles under the premonition that if it looks

correct, it must be correct.

In 2004, Letourneau et al. described the use of a two-dimensional diode detector

array for patient specific quality assurance (QA) of intensity modulated radiation therapy

(IMRT).45 Wide field was used to calibrate the detector array. Letourneau reported that









during calibration, linear accelerator (LINAC) symmetry variations as small as 0.3% led

to calibration errors of up to () 1.2%.45

During recent work to characterize an ionization chamber array (under review by

Medical Physics),53 our group also experienced the WF calibration algorithms sensitivity

to symmetry instabilities. Calibration errors of up to 1.6% [Fig. 4-1A, see Eq. 4-11 for

p_error] were observed for a LINAC that had symmetry variations on the order of ()

0.15%. The calibration error reduced to less than () 0.5% [Fig. 4-1 B] for a LINAC with

symmetry variations of () 0.05%.

While symmetry variations of 0.15% are unlikely to be clinically relevant, the

resultant calibration errors have the potential to bias measurements by a clinically

significant amount. The purpose of this work was to reduce the effect that field shape

perturbations (e.g. symmetry variations) have on WF calibration reproducibility.

Materials and Methods

Materials

The detector array used in this study was the IC PROFILERTM [Sun Nuclear

Corporation (SNC), Melbourne, FL USA], which will henceforth be referred to as the

panel. The panel contains 251 semi-cylindrical ionization chambers and an active area

of 32 x 32 cm2, Fig. 4-2. A detector array is located on each coordinate axis (x and y)

and on the negative and positive diagonal axes (nd and pd, respectively). The detector

spacing is 0.5 and 0.7071 cm on the coordinate and diagonal axes, respectively. All

axes share a common center detector, with the x- and diagonal axes missing detectors

immediately adjacent to the center detector. The inherent thicknesses of buildup and

backscatter are 0.9 and 2.3 g/cm2, respectively.









The 6 MV X-ray beams used in this study were produced by either an Elekta

Synergy (Elekta Limited, Crawley, SXW UK) or a Varian TrilogyTM (Varian Medical

Systems, Palo Alto, CA USA). Each provided different beam characteristics that were

useful in evaluating the calibration algorithm; their use will be acknowledged

accordingly.

Methods

This inquiry into the calibration method was for a single detector array, in this case

the panel's y-axis. This was done to decrease the description length of the calibration

procedure and also to ease the reporting of results. The procedure for determining the

orthogonal axis (x-axis) calibration factors was not changed from the original method

and will,43 therefore, not be covered. Calibration of the pd- and nd-axes follows the

methodology of the y- and x-axes, respectively, and was not covered for the same

reason.

Wide field calibration theory

While the WF calibration was described in detail by Simon et al.,43 a brief overview

is required for this work. The algorithm corrects intra-array detector sensitivity variations

by applying detector specific calibration factors in the form of scalar multipliers. For a

linear detector array, the calibration factors are determined with three array

measurements (a, e and A).

For a, the array is centered on the LINAC crosshairs. For e, the array is rotated

1800 from its position in a. For A, the array is shifted by one detector's spacing from its

position in e. The beam parameters are constant during each measurement. The

primary beam requirement is that the field size needs to be larger than the active area









of the array (35 x 35 cm2 for the full panel). This is to avoid critical spatial substitution in

the high field gradients of the beam's penumbra.

The WF calibration algorithm operates under three postulates imposed during the

three array calibration measurements. The first is that the delivered dose distribution

does not change during a complete calibration. The second is that the relative

sensitivities of the detectors do not change. The third is that the movement of the

detector array from measurement to measurement does not change the scatter

conditions to the LINAC-space reference-frame, i.e. the array phantom heterogeneities

have substitution symmetry.

In theory, a detector array can be calibrated with only e and A. This is

accomplished by determining the relative detector sensitivities through ratios of the

detector readings at the same field locations. The sensitivity of any detector relative to

the first is then expressed as


ctn= -1 (4-1)


where cf' is the theoretical calibration factor for detector n {ne I :2,3,4,...,E} and E is

the last detector of the array; by definition cf', = 1.

Using Eq. 4-1 requires invariant dose deliveries and detector sensitivities between

e and A. Changes in either will introduce error that will be mistaken as differences in the

detectors' relative sensitivities. For this reason, the calibration procedure needs to

account for the changes in dose delivery and global detector sensitivity. This is

accomplished by introducing a correction term (D, S,,) into Eq. 4-1, the derivation of









which can be found in the WF calibration patent.43 The calibration factors are now

calculated using


cf,,== (DO, SA) (4-2)


where cf is the calibration factor for detector n {ne : 2,3,4,...,E}, D is the change in

delivered dose between measurements A and e, and S is the change in global detector

sensitivity between measurements A and e. Again, by definition cf = 1.

The changes in dose and/or detector sensitivity from A to e are difficult to quantify.

One method of doing so is to monitor the dose delivery with a reference detector. This is

difficult since an incomplete compensation of these errors would be propagated by the

product sequence term in Eq. 4-2. With the a measurement, a symmetry solution exists

by determining the mirror calibration factors, i.e. the sensitivity of a detector relative to

its mirror about the array's center.

Mirror calibration factors (cfm) can be determined using two methods. The first

involves calculating the relative calibration factors of mirror detector pairs with a

variation of Eq. 4-2. This is expressed as


cfin, = nH (D, SA)2n-(E+l) (4-3)
L=E-n+ A+1

where n = [(E+1)/2]+1 through E, when E is odd. The second method for calculating

mirror calibration factors involves measurements a and e. Using these measurements,

the mirror calibration factors can be calculated with


cinn E-n+1 C--n+1 (4-4)
V n









where n = [(E+1)/2]+1 through E, when E is odd.

Two solutions now exist for each mirror calibration factor which allows us to solve

for DA6eSA in Eq. 4-3. The mean DAeSAe value for all mirror pairs is used to reduce any

error in this term. The final expression for calibration factors is expressed as


cf = .- (. S,1 (4-5)


For convenience, cf is re-normalized to the center detector, i.e. cfcenter = 1.

The calibration algorithm was recreated in the MATLAB (The MathWorksTM,

Natick MA, USA) software environment, which allowed complete control over the

calibration procedure. Calibration factors created in MATLAB match those created in

the panel's software.

Limiting calibration error

Since the WF calibration algorithm can produce reproducible results [see Fig. 4-

1 B], the primary source of calibration error is related to the delivered beam and

violations of the first postulate. Limiting these postulate violations should reduce WF

calibration errors. For comprehensiveness, the other two postulates are also examined.

Effects of postulate failure

The effect of small postulate violations was quantified by applying a sine shaped

perturbation [pert(y)] to hypothetical calibration fields, effectively simulating a field

perturbation due to a postulate violation. The perturbation was created with

pert(y) =0.001- sin r- Y +1, (4-6)
\ 32cm)









where y {ye I :-16.5,-16,-15.5,...,16} is the in-line position (cm) on the LINAC

coordinate system. The perturbation reaches a maximum of () 0.1% at the array

edges.

A single perturbed field was paired with unperturbed fields in the calibration

algorithm to produce cfx, where X represents the perturbed field (a, e or A). For

example, applying pert(y) to a but not to e and A results in cfa. Using all unperturbed

fields creates a baseline calibration factor (cf).

The effect of perturbation was quantified by determining the percent error

[p_error (cfx, cf ) between cfx and cf for each detector n. This is expressed as


error, (cf, cf ) = fx 1 100. (4-7)


It will be shown in the results section that the WF calibration algorithm is sensitive to

postulate violations. Specifically, with the magnitude defined in Eq. 4-6, perturbations to

e or A result in calibration errors of up to 2%, while a is nearly immune.

Limiting violations of the first postulate

The first postulate assumes that the dose distribution does not change between

measurements. The validity of this assumption was evaluated by determining the in-line

beam reproducibility of the two LINACs (from Fig. 4-1) over 10 consecutive

measurements. Reproducibility was quantified by determining the percent error

[p_error (nmeas)] between each detector's measurements to their mean value. This is

expressed as

Lnmeas>,, 1 (4-8)
p error, (nmeas) -1 -100 (4-8)
nmeas ),









where nmeas is the measurement normalized to the center detector and nmeas is the

mean nmeas value for each detector n.

It will be shown in the results section that the dose distribution reproducibility is

improved while the beam is continuously running, instead of being cycled on and off.

Because of this and the sensitivity of e and A to postulate violations, the array was

calibrated on the Elekta with a continuous beam during e and A. Movement was

provided by a linear stage motor (model NLS4-2.5-16; Newmark Systems, Mission

Viego, CA USA) that was capable of sub-mm movement. Measurement a was a

standard cycling of the beam on and off. This process was repeated four times for a

total of five calibrations.

Limiting violations of the second postulate

The second postulate assumes that if there is a change in detector sensitivity

during the calibration, then all of the relative detector sensitivities change by the same

amount. Failures of this postulate are difficult to quantify since the detector sensitivities

are the values being pursued in the WF calibration. Actions can be taken to reduce the

likelihood of a postulate failure. These include storing the panel in the environment that

it will be measuring and also maintaining a constant power supply to the panel. If this is

impossible, power should be applied for several minutes prior to making

measurements.53

For example, if a panel was stored in an area whose temperature is significantly

different than the temperature of the treatment vault, then as thermal equilibrium takes

place, the temperature gradients across the array may differentially influence the

relative sensitivities from one region to another. Another and more difficult example is









the beam's off-axis energy distribution and its effect, if any, on the energy response of

the detectors. With the rotational detector substitution (a to 9), there will be no effect if

the energy distribution has radial symmetry. But with the linear array shift (0 to A), all the

detectors move into a new energy distribution, albeit ever so slightly different. Neither of

these effects is likely to be large; however, we do see significant calibration errors

induced with very small beam symmetry changes.

Limiting violations of the third postulate

The third postulate assumes that the detector array's movement between

measurements does not change the scatter conditions encountered by the array. While

the array is radially symmetric, that symmetry begins to break down near the distal ends

of the panel. This is well illustrated by the panel's y-axis. The positive y-axis has a

reduced amount of side-scatter relative to the negative y-axis, which is adjacent to the

panel's electronics (see Fig. 4-2). Furthermore, the recommended field size for

calibrating the panel is 35 x 35 cm2 at a source to panel surface distance of 100 cm; the

panel size is 35.3 x 35.3 cm2. The proximity of the field to the panel's edge creates a

dynamic scatter condition as the panel is moved through the steps of calibration (0 and

A).

The effect that additional side-scatter has on the measured dose distribution was

quantified by taking the ratio between the mean of five normalized measurements with

(ss) and without additional side-scatter. This is expressed as


response (nmeasss, nmeas) = nmeas (4-9)
Snmeas
n









where response, (nmeass,nmeas) is the effect that additional side-scatter has on the

measured beam shape. The side-scatter was created with a 4 x 4.2 cm (width x height)

acrylic border that completely surrounded the three open sides of the panel, and

therefore the array. The width of the additional side-scatter was chosen to mimic the

scatter conditions of the negative y-axis, while the height matched the thickness of the

panel and the additional buildup (1 cm) that was used during calibrations.

It will be shown that additional side-scatter at the positive y-axis perturbs the

measured dose distribution. The effect this has on the WF calibration was determined

by calibrating the array five times with and without the additional side-scatter. The

Varian LINAC was used for this portion in an effort to reduce the influence of symmetry

variations from the Elekta during beam on/off cycles. The agreement

[agreement^ (cf,,cf)] between the two sets of calibration factors was evaluated using


agreement cf)= f)- (4-10)


where cf and cf are the mean calibration factors determined with and without

additional side-scatter, respectively.

Evaluating calibration factors

The effect of modifications to the WF calibration methodology was evaluated in

two areas: reproducibility and accuracy. Reproducibility was quantified by determining

a) the coefficient of variation (CV, relative standard deviation) and b) the percent error

[p_errorn(cf )] of each detector's calibration factors (cf) relative to their mean value (cf).

This is expressed as









perror -1 100. (4-11)


Long term reproducibility was evaluated in the previous work on the IC PROFILERTM

using the standard WF calibration protocol and the Varian LINAC.53 Since the

calibration algorithm was unchanged and the results were obtained with the Varian

LINAC, which provided stable array calibrations, the long term reproducibility was not

re-evaluated in this study.

The WF calibration inter-LINAC reproducibility was evaluated with the Varian and

Elekta LINACs; five calibrations were done on each unit. The calibrations on the Elekta

used a continuous beam (during e and A) and additional side-scatter. Those done on

the Varian used additional side-scatter but not a continuous beam; thereby testing the

effects of using a continuous beam as well. Their agreement [agreement,,(c fE cf) was

quantified by determining the ratio between the mean of the two sets of calibrations.

This is expressed as

Cf E
agreement cfEV) f Ej (4-12)


where cfE and cf, are the mean calibration factors obtained with the Elekta and

Varian, respectively.

The accuracy of the array calibration was evaluated by taking two full field

measurements with a 1800 panel rotation between measurements. The 1800

measurement had its data rotated about the center detector to maintain the LINAC

coordinate system of the 0 measurement. The calibration accuracy









[accuracypo (nmeas1 o, nmeaso)] was then quantified by taking the ratio of the 1800 to the

0 measurement. This is expressed as


accuracy (nmeas1, nmeas nmea (4-13)
nmeas po

where nmeasoo and nmeasao80 are the normalized measurements for the 0 and 1800

panel orientations, respectively, and pos is the spatial position on the LINAC axis. An

accurate calibration would result in minimal differences between the two measurements

since measurements at a point in space should be detector invariant if the detector's

response is known by calibration. However, there may be a subtle effect across the

panel from the off axis energy distribution. It would probably not be detected in a mirror

symmetry test using a 1800 rotation measurement comparison. To address this, we

compared to a scan in water.

The calibration accuracy was also evaluated by comparing panel and scanning

water tank [Blue PhantomTM and CC13TM (IBA, Schwarzenbruck, DE)] profile

measurements of a 6 MV X-ray beam (Elekta). Four square field sizes were measured

(with sides of 5, 10, 20, and 30 cm) at two depths (1.5 and 10 g/cm2) and a source to

surface distance (SSD) of 90 cm. Two array calibrations were evaluated that used the

continuous beam (Elekta LINAC) in either the presence or absence of additional side-

scatter. The profile agreement [accuracypo (nmeas pane, nmeaswater)] was quantified by

taking the ratio of panel to water tank measurements. This is expressed as


accuracypo (nmeaspanel, nmeaswatr ) =mea pane (4-14)
Snmeaswater po









where nmeaswater and nmeaspanel are the normalized measurements for the water tank

and panel, respectively, and pos is the spatial position on the LINAC axis.

Results and Discussion

Effects of Postulate Failure

The perturbation that was applied to the hypothetical calibration fields is seen in

Fig. 4-3A; the inset offers an expanded view. This shape was chosen because it closely

resembles the actual symmetry variations in beam on and off cycling that were

observed with the Elekta LINAC (these results are presented in the next sub-section).

Within real measurement and data precision limits, it can be difficult to track the effects

of a field perturbation on the calibration. The value in this hypothetical analysis lies in its

precise ability to quantify the effects that small perturbations have on the calibration

algorithm.

The calibration error due to simulated perturbations is seen in Fig. 4-3B. When

applied to a, the perturbation resulted in minimal changes [() 0.07%] relative to the

baseline calibration. The relative immunity of this measurement to postulate violations is

due to its role in determining the D6eSAe correction (see Eq. 4-4), which has limited

potential for error propagation. Perturbation applied to e or A resulted in calibration

errors of up to () 2%. This strong response highlights the potential for error propagation

due to the product sequence term in the WF calibration algorithm (see Eq. 4-5).

While not covered in this work, the effect of perturbations to the orthogonal (x-axis)

calibration factors is minimal. Their method of calculation lacks a linear shift and hence

the same potential for error propagation.43 The x-axis cfs do, however, use the y-axis

cfs in their calculation and are hence susceptible to the errors associated with them.









Limiting Violations of the First Calibration Postulate

The potential of a 2% calibration error due to a 0.1% dose distribution error seems

to preclude the WF calibration algorithm from generating consistent results.

Measurement errors that exceed this are unavoidable. However, if we recall the results

from Fig. 4-1 B, we know that the WF calibration can produce reproducible calibration

factors. Therefore, the variations in calibration reproducibility that were observed in Fig.

4-1A must be related to the LINAC and the first calibration postulate.

This conclusion is supported by comparing the reproducibility of the Elekta and

Varian dose distributions. In Fig. 4-4A, we see the in-line dose distribution variations for

the Elekta LINAC. The overall variation was small, () 0.15%. However, some of the

variations closely resembled the perturbation from the previous subsection, making

them the likely source of the WF calibration errors.

Similarly, we would expect the Varian LINAC to have a tighter error spread for its

beam. In Fig. 4-4B, we see this to be the case. The first two measurements (1 and 2)

have a higher noise component compared to the subsequent measurements. The

decrease in noise from the first to the second measurement and the relative absence

thereafter indicates that this phenomenon is due to the pre-irradiation effect.54 While the

first two measurements have increased noise, the shape of their dose distributions was

consistent with the subsequent measurements. This indicates a remarkably small

amount of beam variation.

It must be emphasized that these two machines represent sample sizes of one.

Each is in clinical use and the observed variation of 0.15% is insignificant to the

intended use of clinical treatment.









The increased error observed with the center detector [see Fig. 4-4B] has a limited

effect on WF calibration reproducibility. A noise event at a single detector will only

contribute to error propagation once; all subsequent detectors will be off by that amount.

However, that level of error (0.15%) is minimal. The increased error in the other

detectors is biased both positively and negatively. These cancel each other and

therefore do not contribute significantly to calibration errors. While the pre-irradiation

effect is small, it should be considered when performing WF calibrations.

The likely cause of the beam variations in Fig. 4-4A is small fluctuations in the

LINAC's electron spot. As the beam is cycled on and off, the spot has micro variations

in its location and shape. This causes variations in the impingement of X-rays on the

flattening filter, which consequently alters the delivered dose distribution. Turning the

beam on and leaving it on produces less variation in the electron spot and ultimately a

more stable dose distribution. Evidence of this is seen in Fig. 4-4C. The cross-line

symmetry is also likely to be more stable than the in-line, making it a better candidate

for the WF calibration. However, this did not always hold true for other machines that

were tested.

The p_error of calibrations done with a continuous beam (during e and A) on the

Elekta are shown in Fig. 4-5. The maximum error decreased from () 1.6% to () 0.68%,

Table 4-1. This method did not entirely eliminate the error trends that were observed in

Fig 4-1A, which indicates a reduced but continued variation in the dose distribution

between e and A.

Limiting Violations of the Third Calibration Postulate

The effect that additional side-scatter has on the measured calibration beam is

shown in Fig. 4-6A. An increase in signal begins at approximately (+) 10 cm on the y-









axis and increases up to 0.35% at the array's end. Additional side-scatter had a minimal

effect on the negative y-axis, which is expected since no additional side-scatter was

added here due to the presence of the panel's electronics (see Fig. 4-2).

The effect that additional side-scatter had on calibration factors is shown in Fig. 4-

6B. An asymmetric response occurred in which the negative y-axis cfs under responded

by up to (-) 0.35% and the positive y-axis cfs under responded by a maximum of (-)

2.21%. These results indicate that the side-scatter conditions are not adhering to the

third postulate.

Evaluating Calibration Factors

The inter-LINAC agreement of calibrations is shown in Fig. 4-7. The two

calibrations agree within () 0.4% of each other. Since both calibrations used side-

scatter, this tested the viability of using a continuous beam (Elekta) to match the

calibration results obtained with a standard on and off beam (Varian). The general

agreement between the two sources shows that the algorithm produces consistent

results for different LINACs when the continuous beam approach is used.

The accuracy of the continuous calibration factors was determined with four

measurement (meas) and calibration (cal) combinations. These were created with

(w/ss) and without (n/ss) additional side-scatter. The calibration accuracy is shown in

Fig. 4-8; the symmetric error indicates calibration errors. The smallest deviations from

unity [() 0.4%] occurred when additional side-scatter was used during both the

calibration and the measurement. The worst combination [() 2%] involved

measurements with additional side-scatter paired with a calibration that did not use

additional side-scatter. These results indicate that additional side-scatter should be

used during calibration and for best results, during measurements. However, the









penalty for not using side-scatter during measurements is approximately 0.5% at the

extreme array ends.

In Fig. 4-9, we see the profile agreement between the panel and water tank

measurements. Two panel calibrations were used that were acquired with a continuous

beam (Elekta LINAC) in the presence and absence of additional side-scatter, Figs. 4-9A

and B, respectively. The additional side-scatter calibration improved the agreement

between the two detector systems for the shallow measurement (30 x 30 cm2), reducing

the maximum error from to 1.3% to 1.0%. However, for the 10 g/cm2 measurement (30 x

30 cm2), the side-scatter calibration reduced the agreement from a maximum error of

0.87% to 1.0%. For smaller fields (< 30 x 30 cm2) the panel measurements were

typically within () 0.5% of the water tank.

Judging by the absolute change in error, using additional side-scatter during

calibration improved the overall agreement. However, the relatively steep increase in

error for the 10 g/cm2 measurement beginning at approximately (+) 8 cm is suspect.

That position coincides with the beginning of the WF calibration's response to side-

scatter [see Fig. 4-6B].

The benefit of using additional side-scatter during calibration is not entirely clear. It

provided the most reproducible calibrations, the best accuracypos(nmeas80oo,nmeasoo)

values, and improvement in the shallow profile agreement. It did, however, worsen the

agreement for the 10 g/cm2 measurement. The use of additional side-scatter during WF

calibration is under continued investigation.

Other Factors Affecting the WF Calibration

The response of the array's detectors to external stimuli (e.g. beam energy and

additional buildup) may produce WF calibration dependence to that stimulus. In









previous work to characterize the IC PROFILERM, the dependence of the WF

calibration to these two variables was quantified.53 The Varian LINAC was used with the

standard calibration procedure since it produced stable calibrations. The methodology

of these experiments was covered in the text of that manuscript.

There was minimal calibration response to additional buildup (< 0.8% for 0.9 to 4.9

g/cm2) and a slight response to energy (< 1.5% for 18 to 6 MV X-rays).53 The

recommendations of that work were to calibrate the array with buildup that met or

exceeded the energy specific d-max and to use an energy specific calibration.

Other Arrays and IMRT

The majority of arrays using the WF calibration algorithm have a reduced number

of detectors in their translational arrays relative to the panel. For example, the

MapCHECK 2TM (SNC) has 33 detectors in its translational array, versus 65 for the

panel. This lower number reduces the amount of error propagation that can occur in Eq.

4-5. As a result, the MapCHECK 2TM calibration error would likely be half of the panel's

on the same LINAC.

An important question to ask is: how does this affect patient specific QA for IMRT?

For the case of a MapCHECK 2TM on a LINAC with symmetry variations of () 0.15%,

the error would likely be half of what was observed with the panel, making it roughly ()

0.8%. Using passing criteria of 3%/3mm, a calibration error of this size is unlikely to

artificially pass an unsuitable plan. However, certain conditions could lead to biased

clinical decisions if errors stack up in the same direction.

Conclusion

The aim of this work was to reduce the effect that minor measurement

perturbations have on the calibration reproducibility. These occur via violations of the









three WF calibration postulates, which state that a) the beam shape does not change

between measurements; b) the relative sensitivities of the detectors do not change; c)

the scatter does not change as the array is moved between measurements. During WF

calibration, perturbations to the beam shape of 0.1% can lead to calibration errors of ()

2%.

Postulate violations were limited by using a continuously on beam during portions

of the calibration procedure. This increased the beams stability and therefore limited the

error propagation that occurs in the algorithm. Additional side-scatter was also added to

the IC PPROFILERTM to increase the scatter uniformity.

The overall effect was to reduce the calibration error from approximately () 1.6%

to () 0.48%. The agreement between measurements done with the panel and a

scanning water tank were () 0.9%. The calibration error for arrays commonly used in

patient specific QA (IMRT) would likely be half of what was observed with the IC

PROFILERM, which would be unlikely to affect a clinical decision using a distance to

agreement of 3%/3mm. This work increased the reproducibility of the WF calibration

procedure for LINACs that have minor symmetry variations.









Table 4-1. The short term reproducibility of WF calibrations performed on the Elekta
using either the standard protocol or the continuous beam [with and without
additional side-scatter (ss)].
Standard Continuous without ss Continuous with ss
Mean CV(%) 0.56 0.25 0.15
Maximum CV(%) 1.4 0.57 0.33
Maximum p_error (%) 1.6 0.68 0.48














o 1 ""**-... ......o 1


0-1 .---- -1
1 .. -I .................. ....................................... ........... .......................................
-calibration 1 -calibration 1
-2 calibration 2 -2 calibration 2
--calibration 3 --calibration 3
-10 0 10 -10 0 10
y-axis (cm) A y-axis (cm) B
Figure 4-1. Wide field (WF) calibration reproducibility on LINACs with beam symmetry
variations of (A) () 0.15% (Elekta LINAC) and (B) () 0.05% (Varian LINAC).


Figure 4-2. Oblique view of the panel's arrays and electronics. The panel's y-axis is
parallel to the long axis of the device.











98

















1
0a
.0.99

0.98


-10 0 10 -10 0 10
y-axis (cm) A y-axis (cm) B
Figure 4-3. (A) The perturbation that was applied to the hypothetical calibration
measurements. (B) The effects of perturbations applied, separately, to each
calibration measurement.


-0.0!

6 -o.1,
0. -0.1!


-15 -10 -5 0 5


10 15


15 -10 -5 0 5 10 1
in-line (cm)


in-line (cm) C
Figure 4-4. The percentage error between ten consecutive measurements and their
means for (A) a LINAC that produces () 1.6% calibration reproducibility, (B)
a LINAC that produces () 0.5% calibration reproducibility, and (C) the same
machine as used in (A) but with the beam continuously on. The spikes in the
center detector of (B) were due to the pre-irradiation effect of ionization
chambers and do not contribute significantly to calibration errors.




99


0 9 10 0 10
y-axis (cm)


151------- i --[------ '---------r
o il I I I J ,




3.1 t...-


031












2


o O. -------------- ----1 ..
-calibration 1
CL -1calibration 2'
*calibration 3
--calibration 4
-- calibration 5
-10 0 10
y-axis (cm)
Figure 4-5. Calibration reproducibility using a continuous beam during measurements e
and A on an Elekta LINAC; no additional side-scatter was used.



1.o006o 1.03
S1.005 1

1 /,
1uI 1 /u
1.004
01.01
9 17
(D n: J 1_ -- -------- --------------



0I9 ....................... ------------ -----------------------------
0.999
-109 . a A.


Figure


0.99 -15 -10 -4 0 5 10 1.5 -10 0 10
y-axis (cm) A y-axis (cm) B
4-6. The effect of additional side-scatter on (A) beam measurements and (B)
array calibrations. The measurements and calibrations were performed on a
Varian LINAC.


1.03r


1.02
1.01


0.99.
0,9% ...........................................................................................................................................---
CU
0.97
-10 0 10
y-axis (cm)
Figure 4-7. The agreement between calibration factors obtained with side-scatter and
either a continuous beam (Elekta) or a standard on/off delivery (Varian).


100














S1.03
0
CD
1.02


| 1.01
O0

1


W0.99



C,


(U 0.97


Figure 4-8. The calibration accuracy was evaluated for four measurement
combinations. These included the presence (w/ss) or absence (n/ss) of
additional side-scatter in the measurement (meas) and/or calibration (cal).


1': 1~


S1 03
1 02
S01


0 99


S W02
01


8 .,1,.


1L 1i';


in-line (cm)


Figure 4-9. The calibration accuracy expressed as the ratio between water tank
measurements and panel measurements. The panel calibrations were
performed using a continuous radiation source in the (A) presence of and (B)
absence of additional side-scatter.























101


-10 -5 0 5 10 15
y-axis (cm)


-cal w/ss, meas w/ss
---cal w/ss, meas n/ss
-- cal n/ss, meas n/ss
.*** cal n/ss, meas w/ss




"..





--- -- -- -- -- -


I -


1 03
1 02
D 1 01


Q099


1C2
101


Vo.:


10 g/cm2




....*5x 5 cm2
-- 1- :: I ,
--v. 1 5 glcm


V \


10 g/cm2





...5x5cm2
- 1 :: I, .
-"-'..7 ::",: 1 5 gicmi

"-: -


1i 1%.


in-line (cm)


1,' 1


I I


. -15


1., 1 .


1C 1"-









CHAPTER 5
A QUALITY ASSURANCE PROGRAM FOR A DETECTOR ARRAY

Introduction

A vital, but oft overlooked, component of an effective quality assurance (QA)

program is evaluating and maintaining the measurement equipment. The importance of

this is immediately apparent for certain systems (a local standard ionization chamber)

and in general was addressed by the American Association of Physicists in Medicine

(AAPM) in a report by Task Group (TG) 40 of the Radiation Therapy Committee.5

However, since TG-40 was published in 1994 measurement equipment has changed

dramatically, tending toward more electronic and complex systems.

Detector arrays are a prime example of this trend. An argument can be made that

their role in the clinic is every bit as important as that of more traditional systems. Their

use as a benchmark dosimetry system for intensity modulated radiation therapy (IMRT)

plans helped to usher in that modalities widespread use.22' 45,55,56 Detector arrays play

various other important roles including the measurement of linear accelerator (LINAC)

symmetry and flatness levels and calibrating mechanical components.33'44

Since TG-40 was published, there have been few new QA guidelines suggested

for measurement systems. In contrast, there have been numerous characterization

papers that have accompanied new products.44, 51-53, 57, 58 While these are important

pieces of work that establish the behavior of the system in its intended environment,

they do not provide practical guidelines that a physicist can use to evaluate the devices

continued operation. Recommendations are also typically lacking in the device manuals.

The purpose of this work was to create a QA program for detector arrays.


102









First, a question needs to be asked: why is there a lack of guidance in this area?

The primary reason is that it is difficult to establish recommendations that are based on

hard science. Instead, most QA recommendations, including those for LINACs, are

based on experience and/or convention. This work is not, and cannot be, completely

different from this behavior. But it will, where practical, use sound scientific reasoning to

establish tests and tolerances.

Materials and Methods

Materials

This work is intended to be general and applicable to most detector arrays. A

specific array was used to illustrate certain portions of the QA program. That array is the

IC PROFILERTM [Sun Nuclear Corporation (SNC), Melbourne, FL USA] and will

henceforth be referred to as the panel.53 The panel contains 251 semi-cylindrical

ionization chambers and has an active area of 32 x 32 cm2. A detector array is located

on each coordinate axis (x and y) and on the negative and positive diagonal axes (nd

and pd, respectively).

The panel performs measurement updates at near integer multiples of a preset

collection interval. This effectively allows the panel to measure an indefinite amount of

radiation by not saturating the capacitors. The measurement file can either preserve

these updates (in the form of a running cumulative signal) or the final cumulative update

is kept. A file that preserves each update is referred to as a multi-frame measurement.

The option between multiple and single frames is available due to file size. For example,

a 30 s multi-frame measurement with 125 ms collection interval will produce a file that is

approximately 450 kb, while a single frame measurement is approximately 10 kb.


103









Panel measurements can also be operated in either pulsed or continuous mode.

Pulsed mode measurements are synced with radiation pulses. Continuous mode

measurements are clock synced. Both have intended uses and associated qualities.53

The 6 MV X-ray beams used in this study were produced by an Elekta Synergy

(Elekta Limited, Crawley, SXW UK).

Methods

The QA program is broken into four areas: physical, personal computer software

and device firmware, electronics, and array calibration. When a test is described, it will

generally refer to an array; portions of the program will then be applied to the panel in

the results and discussion. The tests, their frequencies, and action levels specific for the

IC PROFILERTM, at our institution, are consolidated in Table 5-1.

Physical

Detector arrays are relatively robust compared to typical dosimetry systems

(thimble ionization chambers, radio sensitive film, scanning water tanks, etc.). This is

primarily due to their packaging and an inherent lack of moving parts and water. Despite

this, they are still complex systems and need to be treated accordingly. Before each use

of the array, it should be examined for any physical damage. Relatively little can be

done on site if severe physical damage occurs. Two items that can be continually

checked and taken care of are the array's buildup and power/data cables.

Buildup: Extended use will accumulate grime on an array's top surface (including

additional buildup). While this poses little to no risk to the actual device, it does increase

the friction between the array and added buildup. Consequently, there is a higher

likelihood of the array being pushed out of position while adding or sliding buildup

across the top surface, thereby skewing the measurement geometry.


104









As accumulation occurs, the top surface (and additional buildup) is cleaned with a

micro-fiber cloth moistened with a small amount of water. All power to the array should

be disconnected during cleaning and extreme care should be taken so that water is not

allowed to flow or collect on any surface.

Cables: Power and data cables are a common failure point for devices and occur

almost entirely due to improper handling. Cables should always be disconnected by

grasping and pulling the connector rather than the cable. Pulling the cable will stress the

connection between the connector and the cord, which can expose conductors.

Failures also occur due to repetitive circular movement. This is caused by taking

an array (attached to a cable) out of a cabinet and turning clockwise to put it on the

treatment couch; after the measurement session, the array is then put back into the

cabinet using a second clockwise movement. The result is a cable that is continually

twisted and strained, and can quickly wear out through frequent repetitions. Cables that

are separating from the connector or that are severely twisted should be replaced

immediately to prevent personal injury, physical damage to the array, and loss of

measurement capability.

Firmware and software

It is important to stay current with software for reasons of defect correction and

improved performance. It is also important to report defects and desired improvements

that contribute to confidence in measurements. Firmware and software updates occur at

different frequencies for different manufacturers. Ideally, an array's software would

automatically inform a user when an update is available. A solution that SNC is

incorporating is predefined bi-annual release dates. Knowing these dates, the user can


105









efficiently schedule an update after the planned release date has passed. Otherwise,

updates are checked for based on the manufacturers release patterns.

Measurement consistency: Before applying an update, a benchmark setup

measurement measly ) is made twice and checked for consistency. In this situation a

benchmark setup measurement is defined as a set of invariant beam and measurement

parameters [e.g. for the panel the beam parameters could be a 6 MV X-ray beam, 33 x

33 cm2 field size, no added buildup, and a 100 cm source to surface distance (SSD)].

Without disturbing the array, the update is applied and a second benchmark

measurement (meas2) is made twice; these are also checked for consistency.

An inconsistency between the two measurements could be the result of a number

of different problems: instable beam delivery, a temperature gradient in the array, or the

pre-irradiation effect.54 59 If an inconsistency is identified and corrective actions are

taken, two measurements should be made again and checked for consistency. The goal

is to be certain that the array measurements are not influenced by source outside of the

software update, which could invalidate this test.

Using the first measurement from each set, the percent error (p_error) is

determined for detector n with


p _error (measl, meas2) = mea -1 100. (5-1)
meas2l)

Since the geometry was unchanged between measurements the difference in data

should be due to LINAC and measurement noise. Large differences are the result of the

software or the firmware update.

Over a short time frame, the panel's maximum percent error versus a mean was

reported as () 0.15%;53 similarly the LINAC's maximum error was also reported as


106









approximately () 0.15%.59 Accounting for other possible sources of error,54 59 we can

estimate a maximum amount of measured error at approximately () 0.5%, this value

can be looked on as a tolerance level. Similar analyses can be used to set tolerance

levels for other arrays. This test is only valid for detectors within the radiation field.

Detectors outside of the direct beam have a lower signal to noise ratio, which can result

in false positives.

A software update that results in a legitimate difference should be accompanied by

an explanation and also should be investigated for its impact on past and future

measurement applications. For example, a software bug that improperly calculates

symmetry may have produced a symmetry value that is used to justify IMRT patient

plan QA agreement.

Data consistency: Following an update, the consistency of previously measured

data is evaluated. A baseline measurement that was taken upon first receiving the array

is opened in the array's software and the data is copied (or exported) to a spreadsheet

program (Excel; Microsoft Corporation, Redmond, WA USA) with previous exports of

the same measurement; each export should be identical. Any difference is a deviation

and should be accompanied by an explanation from the vender. A software update that

results in a difference should be investigated for how it impacts past and future

measurements.

Software features: The proper function of software's analysis features (flatness,

symmetry, y index, etc.) is an important aspect of a QA program. Their QA is beyond

the scope of this work given their variety amongst all manufacturers and devices.


107









However, fundamental and essential QA can be accomplished by trending their output

of a benchmark measurement setup.

Electronics

The operation of the array's electronics and detectors is evaluated background

and beam measurements, both of which contain unique and separable information

about the system performance.

Visual examination of background: At the beginning of a measurement session, a

background measurement is performed and visually examined. The appearance and

amplitude of the signal should be consistent with previous background measurements.

Frequent use of the detector array provides the user a visual memory of prior

background appearance; however, if the array is used infrequently, a previous

background measurement can be opened for comparison. Despite its simplicity, this is a

powerful test since a systematic influence will result in a deviation from the array's

normal behavior. This test is done before each measurement session. Abnormal

behavior should be investigated before the array is used.

Test of background normality: A more rigorous test of the measured background is

to examine its' probability density function (PDF) for normality. A convenient method of

doing this is to make a background measurement with the array operating in continuous

multi-frame mode. The raw counts that occur during each update (accumulated charge

measurement) are then calculated by taking the difference between the successive

measurement updates. This is expressed as

UC, (1) = RawCounts, (1) and
(5-2)
UC, (ii) = Raw Countsn (ii) Raw Countsn (ii -1),


108









where UC is the raw counts per update ii {iie D :2,3,4,...,L}, L is the last measurement

update, and n is the detector. Since UC is a measure of background, its PDF should be

normal.

The normality of UC was evaluated using two goodness of fit (GOF) tests: Chi-

squared and Anderson-Darling.60'61 Chi-squared was chosen because it is a common

GOF test and serves as a benchmark; Anderson-Darling is better suited at assessing

the normality of an observed PDF.62 The null hypothesis (Ho) was that UC's PDF was

normal for each detector; the alternative hypothesis (Ha) was that each detector's PDF

was non-normal.

By convention, a significance level (a) of 0.05 was used; this level can easily be

increased, but with the risk of increasing Type I errors (i.e. rejecting Ho when it is true).

Both GOF tests are easily implemented in a spread sheet program such as Excel. A

rejection of Ho at the selected significance level is evaluated for the magnitude of

rejection. Rejected values that are close to a should be repeated. A second rejection

likely indicates a biased system.

Ideally, this test would be performed before each measurement session. However,

if a highly integrated version of the test is not available, then it can be used after an

anomaly is encountered during the visual inspection of the background measurement.

The test should also be performed at least annually.

Dose linearity: The dose linearity of the array is evaluated with LINAC deliveries of

100 and 200 monitor units (MU). Each is delivered fives times; these measurements will

also be used to evaluate the array's reproducibility. The measurements are performed

at an SSD of 100 cm to the array's surface and with a buildup that is at least as much


109








as the energy specific d-max. The deliveries are simultaneously measured with a local

standard instrument (LSI) in a 4 x 30 x 30 cm3 slab of water equivalent material that is

located below the array and aligned to the LINAC's cross hairs.

The protocol of TG-51 is used to convert the charge measured by the LSI to

dose.42 While this setup violates the protocol of TG-51, the measurement is proportional

dose and can be used as a normalization method to compensate any changes or

deviations in the delivery system and the air density.

The measured signal from the panel's center chamber and the independent

system are normalized to the 200 MU delivery. The array's response is then normalized

to the LSI's response using,

SMEAS (100MU)
MEAS(200MU) ) panel
response (MU)= M S (ane (5-3)
SMEAS (10OMU)
MEANS (200MU)

where response(MU) is the array's dose linearity relative to a Farmer-type chamber and

MEAS is the measured signal for each detector system (CC or cGy for the array or LSI,

respectively).

This test is done annually. The tolerance level is based on the users accepted

level of error. For example, TG-142 recommends a dose linearity tolerance level of ()

2% for exposures greater than or equal to 5 MU.17 Therefore the array should have a

dose linearity error less than 2%. For example, if the arrays dose linearity is 0.5%, then

the balance of the error budget for setup and LINAC is 1.5% which effectively places a

tighter limit on the recommendations of TG-142.


110









Measurement reproducibility: The measurements from the dose linearity section

are used to evaluate the array's short and long term reproducibility. The protocol from

TG-51 is once again used to convert the charge measured by the LSI to dose.42

The short term reproducibility is evaluated using five 200 MU dose linearity

measurements. Each panel measurement is normalized to the LSI measurement,

resulting in NORM. This is done to correct for any variations in the LINACs output and

will be of value for the long term reproducibility. The percent error [p_error(NORMii)]

between each normalized measurement ii and the mean of all the normalized

measurements is determined using


p_ error (NORMAI ) NOR=, 1 .100, (5-4)
NORM 20oMU

where NORMii is the normalized measurement and NORM is the mean of all

normalized measurements.

Long term reproducibility is evaluated for the mean of a measurement set kk

versus the mean of the baseline measurement set. This is expressed as

NORMkkk
p error (NORM kk\= -O 20_i -2 100 (5-5)


where NORM is the mean of measurement set kk or the baseline set. These tests are

done annually. The tolerance level is based on the users accepted level of error.

Array calibration

An important property of detector arrays is a uniform exhibition of detector

sensitivity to radiation. Otherwise, array measurements will be proportionally biased by

each detector's sensitivity. Many methods have been developed to normalize these


111









sensitivity variations;43 49, 50 among these, the wide field (WF) calibration is commonly

used due to the ease and efficiency with which it can be performed on site.45' 52,59 Since

many arrays use WF calibration, this work will focus on it, but in general the QA

program should still be applicable to other calibration techniques.

Reproducibility: Since many arrays are user calibrated, it is the user's

responsibility for verifying the array's calibration accuracy. An initial step is determining

the calibration reproducibility; this can be quickly evaluated for two successive WF

calibrations using


p _error(cf2,cf = cf] 1 100, (5-6)


where p_errorn(cf2,cfl) is the percent error between the two successive WF calibrations

(cfl and cf2, respectively) for detector n.

The user needs to determine the level of calibration error acceptable for the

application, as derived from requirements in task group reports, original equipment

maker (OEM) standards, acceptance testing, etc. For example, IMRT QA that uses tight

passing criteria will have less room to accommodate calibration error. A 2% calibration

error coupled with IMRT passing criteria of 3%/3mm will leave a window of 1% for all

other errors (TPS algorithm, measurement, setup, etc.). The array calibration error must

at least be less than the passing criteria or the tolerance of the item that is being

measured. In contrast, an array that is solely used for tracking profile deviations from a

baseline measurement has no real calibration reproducibility constraints.

The frequency for evaluating the calibration reproducibility is during an array

calibration, which should occur annually. During the annual calibration, the short and


112









long term reproducibility's are determined. For long term stability, Jursinic and Nelms

estimated that annual calibrations would be sufficient to maintain an accurate

calibration.57 If there is reason to suspect that the array calibration has drifted and

needs to be repeated, it's accuracy can be checked with the following test.

Calibration accuracy rotational substitution: The accuracy of the calibration

factors can be tested by performing two measurements with setups that differ by a 1800

array rotation. This locates each detector in the same field location as its mirror detector

about the array's center. Since the array calibration is intended to provide measurement

invariance with respect to detector, there should be minimal difference between the two

measurements. This is assessed by taking the ratio between the two measurements

using

(nmeas^'o
cf accuracylp e (80n,s1 nmeaso) = nmeaso (5-7)
nmeasso

where cf_accuracy is the accuracy of the calibration factor for axis position pos and

nmeas is the normalized measurement for array orientation 0 or 1800.

The tolerance level for this test of the calibration accuracy is dependent on the

tolerance level established for the calibration reproducibility assuming there are no

systematic influences that result from an array rotation. To determine the accepted level

of error, the tolerance from the reproducibility test should be doubled since the error is a

result of two detectors. This is not stochastic error, it is the maximum expected error,

and therefore should not be added via quadrature.

This test is conducted during an array calibration and if there is a specific reason

to believe the calibration is invalid (an increase in the failure rate of patient specific QA

or a deviation from a baseline measurement). Due to the nature of the calibration


113









procedure,59 small shifts in the off-axis X-ray spectrum may introduce a symmetric error

to the calibration factors. This rotational test cannot detect this or other types of

symmetric calibration errors; a direct comparison to a water scan is needed.

Calibration accuracy water scan: The accuracy of the calibration factors can be

checked against a water scan using

nmeas
cf _accuracy, (nmeasarr, nmeaswatr) nmeas arra (5-8)
nmeas
pos, water

where cf_accuracy is the calibration factor accuracy for detector n, axis position pos;

nmeaswater and nmeasarray are the normalized measurements acquired with the water

tank and array, respectively. The array measurements can be checked against annual

QA measurements using a similar geometry as the water tank. However, if the LINAC

beam varied significantly for any reason (a major component failure) since the water

scans, those water scans should not be used in comparison to the array measurements.

Once again the tolerance is dependent on the type of test that is being conducted

by the array and is user defined. This test is only recommended during the initial use (or

acceptance) of the array. Afterward, the reproducibility and the rotational substitution

tests will be sufficient to maintain calibration accuracy.

Results and Discussion

The QA program described in Materials and Methods was applied to the panel.

The results and experiences are described below.

Physical

During the course of one year, the panel's top surface was cleaned twice due to

buildup of grime. Cable damage due to improper handling is seen in Fig. 5-1. The cable


114









handling recommendations (discussed in the previous section) should be followed to

avoid these types of damage.

Firmware and Software

Measurement consistency: An example of the panel's measurement reproducibility

on the Elekta Synergy is shown in Fig. 5-2. The methodology of Eq. 5-1 was carried

out for five measurement sets. There were no software updates during these

measurements; these results purely represent the magnitude of measurement variation

that occurs with the panel on the Elekta Synergy over a short time period.

Two interesting effects are present in set 1. The first is a sine shaped symmetry

variation that is likely the result of micro-variations in the electron beam's spot size and

location. The second is the pre-irradiation effect on the panel's ionization chambers; this

results in increased noise throughout the measurement, specifically for the center

ionization chamber for this particular device.

Electronics

Visual examination of background: In Fig. 5-3A we seen a screen shot from the

panel's software (pd-axis) of a background measurement with a spike in one of the

detector's responses. A contaminant had entered the chamber and was bridging the

gap between the anode and the cathode, causing a high leakage rate. The panel was

serviced by the manufacturer, resolving the singular detector response.

In Fig. 5-3B we see a comparison of backgrounds that were taken seven months

apart on a panel that was operating correctly, the black line represents the older

measurement and the red line represents the newer measurement. While their

magnitudes are different, the overall shape is similar, which is a good indication that the

panel is working correctly.


115









Over time, the panel's sensitivity to background gradually changed with

accumulated dose, Fig. 5-3C. These values represent the mean bias value (leakage

rate) from the background measurement that is taken when the panel's software is

turned on at the beginning of a measurement session. This analysis was not conducted

as a formal study, which accounts for the use of days instead of accumulated dose and

the somewhat inconsistent conditions of measurement. The inconsistent measurement

conditions specifically refers to the time when power was applied to the panel before the

detectors' background bias was measured, this can affect the bias amplitude. While it is

not a formal study, it does give some indication that the background sensitivity changes

with accumulated dose. Whether this occurs for all arrays is unknown; but it should be

considered when one visually compares background measurements to previous

versions.

Test of background normality: Upon initially receiving the panel, five background

measurements were made, each approximately 23 s in duration and containing 180

updates (125 ms collection interval). For each measurement, the UC values (see Eq. 5-

2) were determined for each of the 251 individual ionization chambers. On these, the

GOF tests were used to evaluate the normality of UC's PDF. The percentage of

detectors that rejected Ho are shown in Table 5-2. under the row heading of 'Leaky.'

The normality of these UCs is poor, as can be seen in the percentage of detectors that

rejected Ho; all were in the upper 90's. The panel was discovered to have a contaminant

(the same as discussed in visual examination of the background) on several of the

electrodes that were bridging the gap between the anode and the cathode, thus

introducing a bias into the measurements.


116









The panel was returned to manufacturer for service. After which, the

measurements and GOF tests were repeated. The PDF of UC for the (-) 12.5 cm x-axis

detector (X8) is shown in Fig. 5-4 and appears normal. The percentage of detectors that

rejected Ho are shown in Table 5-2. under the row heading of 'Post.' The percentage of

detectors rejecting Ho dropped to 23 and 15% for the Chi-squared test with significance

values of 0.1 and 0.05, respectively. Using the Anderson-Darling test, no detectors

rejected Ho. The Anderson-Darling GOF test is better suited to assess the normality of a

PDF than the Chi-squared test due to a higher power score.62 Based on these results,

the panel's behavior was deemed acceptable for use.

After a year of heavy use, the measurements and GOF tests were repeated, Table

5-2. under the row heading of 'Post 1 year.' The results show that despite a change in

the global bias value (see Fig. 5-3), the behavior of the background measurements

remained normal.

The two GOF tests were tested against 251 randomly generated normal

distributions (generated in MATLAB), simulating background measurements for each

of the panel's 251 ionization chambers. The percentage of simulated detectors that

rejected Ho are shown in Table 5-2. under the row heading of RND. These values

illustrate the ability of each test to assess the normality of a PDF. Once again the

Anderson-Darling test proved to be a better test for assessing normality.

Dose linearity: The dose linearity of the panel was determined relative to the dose

linearity of a LSI, in this case a Farmer-type chamber and electrometer (model FC65-G,

IBA, Schwarzenbruck, DE; model K602, CNMC Company, Nashville, TN USA). In

previous work to characterize the IC PROFILERTM,53 a dose non-linearity was identified


117









and corrected with new firmware. Using this uncorrected version of the panel, the dose

linearity test identified a maximum panel deviation from the LSI of 1.1%. While this

value is less than the monitor chamber linearity tolerance [() 2%], and therefore could

theoretically pass, there was a cause for concern and it was investigated further.

Using the corrected firmware, the maximum deviation of the panel's dose linearity

relative to a LSI decreased from 1.1% to 0.48%. Once again using the tolerance set in

TG-142 as a guideline, this value left 1.52% error available to the machine QA test

before the tolerance was crossed or a machine with a dose-nonlinearity was not

identified.

Measurement reproducibility: The short term reproducibility of the panel relative to

the LSI resulted in a maximum percent error of 0.08% and a mean percent error of

0.04%. These values are consistent with previously determined values and small

enough to be a non-factor.53

Setting a tolerance value is dependent on the quantity being measured. Measuring

output with the recommended tolerances of TG-142 [3%, 2%, and () 1% (absolute), for

daily, monthly, and annual QA, respectively] would require a panel reproducibility that is

less than those tolerances. The short term panel reproducibility of 0.08% is well below

these values. The long term panel reproducibility established by Simon et al. was 0.84%

using a cobalt source that was not normalized to a LSI. Instead a set amount of time to

integrate the signal was used as the standard; the timer standard deviation was

insignificant. Using 0.84% gives an error play of 2.2% and 1.2% for measuring daily and

monthly output before the output is falsely failed or a failure is not identified.


118









Array Calibration

Reproducibility: The primary factor affecting WF calibration reproducibility is

variation in the delivered beam symmetry.59 A beam that has a higher degree of

variation may only be capable of producing array calibrations with error approaching 1

to 2%.45, 59 Simon et al. described steps to reduce this effect; on a machine with minor

symmetry variations [() 0.15%] the panel's calibration error was reduced from () 1.6%

to () 0.48% (short term reproducibility).

Simon et al. established the WF calibration reproducibility (for the panel) as ()

0.5%.53, 59 If the panel is used as an absolute comparison to water scans, then this

reproducibility would allow a combined error of 0.5% due to other sources before the

tolerance level of 1% (from a baseline value) is crossed.17 This level of calibration error

seems acceptable not only for profile measurements, but also for other common

detector array uses (IMRT QA).

Calibration accuracy rotational substitution: Using the calibration reproducibility

of () 0.5% established above, the panel's tolerance for calibration factor accuracy

(determined by rotational substitution) was set at () 1%. The cf_accuracy results for the

panel are seen in Fig. 5-5. The symmetric error that occurs on the y- and pd-axes is a

calibration error less than the tolerance; the maximum deviation from unity was 0.87%.

Since each position is a combination of the mirror detectors, the calibration error is half

of the cf_accuracy value or 0.44%.

Using the recommended tolerance of 1% from TG-142, this leaves a combined

error of 0.56% before that tolerance level is reached. Whether this is an acceptable

level of error depends on the institution.


119









Calibration accuracy water scan: Profile measurements (10 g/cm2) with the

panel and a scanning water tank [Blue Phantom and CC13 ionization chamber (IBA,

Schwarzenbruck, DE)] are seen in Fig. 5-6A. The calibration factor accuracy values are

shown in Fig. 5-6B. The maximum disagreement between the panel and the CC13 for

the 1.5 and 10 g/cm2 measurements was (+) 1.3% and (-) 0.71%, respectively. The

agreement for the 10 g/cm2 measurements is relatively uniform, with some noise

involved; there is a symmetric concave error associated with the 1.5 g/cm2

measurement. This is the type of error that the rotational substitution test cannot detect.

Contrary to the 1.5 g/cm2 measurement, the 10 g/cm2 results are relatively uniform

and possibly even slightly convex in shape. This indicates that the calibration factors or

the actual detectors have a slight off-axis dependence relative to the CC13 response at

these depths. A likely cause of this are the spectral changes that occur in an X-ray

beam as the off-axis distance is increased. This is highlighted by a slight energy

dependence of the panel's detectors to low energy scatter and/or contaminant

electrons.53

Using the tolerance level of 1%, from TG-142, the agreement values for the 10

g/cm2 comply while the 1.5 g/cm2 do not. This is taking the stance of one to one

comparison with a water tank. However, if we take the role of a baseline comparison,

then the primary purpose of this test was to explore the possibility of a large symmetric

calibration error, which is not present for the reasons already stated.

Conclusion

We have developed a QA program that is general to detector arrays. It focuses on

four test areas: physical, software and firmware, electronics, and array calibration.


120









Specific tests were described with guidelines for determining the tolerance values and

frequencies were also recommended.

Firm tolerance values were not established; instead guidelines and discussion

were given for physicists to set their own values. While specific tolerances would be

beneficial, the complication associated with their determination would be extraordinary

and not universally applicable. Logically, the acceptable error must be less than the test

parameter being measured; ultimately, the physicist is responsible for making an

informed and responsible decision in setting the tolerance levels.

The QA program was applied to the IC PROFILERTM. The test of background

normality effectively identified a background bias in the electronics that the

manufacturer repaired. This QA program is basic and can be expanded on by the

community. Ideally, this work will increase interest in this topic to the point of official

recommendations from the medical physics community and the vendors.


121










Table 5-1. Quality assurance program for the IC PROFILERTM; E, before every use of
the detector array; I, upon initial use of the array or following service; B, bi-
annual; A, annual; AN, as needed
Test Area Test Frequency Action
Physical Panel's top surface and E Clean
buildup
Cables E Replace


Check for updates

Before and after
measurements
File integrity

Background
Visual

Goodness of fit
Beam measurements
Dose response
Reproducibility

Reproducibility

Rotational substitution
Water


0.5%

0%


Visual difference in shape and/or
magnitude
Failure of Ho


() 0.5%


() 1%
() 1%


Table 5-2. Percent of detectors rejecting Ho (= the distribution of UC is normal) using
the chi-squared and Anderson-Darling goodness of fit tests. The hypothesis
was tested for two significance levels (0.1 and 0.05). Four data sets were
evaluated: measurements of the panel with leaky channels, post repair by
one day, post repair by one year; a random normal distribution (RND).
Chi-square test Anderson-Darling test
Measurements a = 0.1 a = 0.05 a = 0.1 a = 0.05
Leaky (%) 98 1.3 97 2.0 97 1.6 96 2.8
Post (%) 23 3.2 15 2.4 0 0 0 0
Post 1 year (%) 14 2.7 8.1 2.0 0 0 0 0
RND Test (%) 5.0 0.60 2.2 0.73 0 0 0 0


122


Firmware /
software


Electronics


Array
calibration






















Figure 5-1. Cable damage as a result of (A) disconnecting the cable by pulling the
cable instead of the connector and (B) repetitive circular movement.


,c .. .set 3
o 0.3 -0set4






0.
0.




1-0.
-0.3 -------------------~-------------


-0.5 -
-15 -10 -5 0 5 10 15
in-line (cm)
Figure 5-2. The measurement reproducibility for the panel on the Elekta Synergy@. The
methodology of Eq. 5-1 was used for five measurement sets; two unique
measurements were in each set.















123












*^- --






-. 5 ---------- ----
2.5 103
2
1.5


05
0 I


't) 100 200 300 400
days in use C
Figure 5-3. (A) An example of two background measurements that were separated by
seven months and have a similar shape and amplitude. (B) An example of an
anomaly in the measurement background. (C) The quantitative change in bias
(leakage rate) over one year.


3 O0


3 07
0
C OJt0
()
3 30 -'

P, 004
LL. 0..3


,I


I


0 02F


-sig -4sig -3sig -2sig -1sig mean 1sig 2sig 3sig 4sig 5sig
UC
X8
Figure 5-4. The PDF of UC for the (-) 12.5 cm x-axis detector (X8).


124


...,.
I


~













1.01
(. .



S1.005









S0.995


(0

x-axis y-axis pd-axis nd-axis

Figure 5-5. The accuracy of the panel's calibration factors were determined using
rotational substitution.


1 03
S1 02
101


('0
ci




U*<1 .-,


-20 -10 0 10 20 1 1' I 1
in-plane (cm) A in-line (cm) B
Figure 5-6. (A) Profile measurements with the panel and a scanning water tank. (B)
Calibration factor accuracy results, obtained by taking the ratio of the panel to
the CC13 profile measurements.



















125


10 g/cm2




.. 5x5cm2

1 5 grcmn

-V..'


I I I I I I I









CHAPTER 6
SUMMARY AND FUTURE WORK

Recently, the quality of radiation therapy (specifically instances of overdoses) have

been brought to the attention of the American public.6 Several instances of human error

that led to a lethal dose being delivered to patients were illustrated in the New York

Times. At the heart of these incidents is an inherent lack of time in this profession to

complete all aspects of QA and quality control. The purpose of this work was to illustrate

the increase in efficiency of LINAC QA and radiation data collection that can be

accomplished with the use of detector arrays. Two time consuming measurement

processes were used as a showcase: MLC calibration and water tank scans.

Specific Aim 1: Multi-leaf collimator calibration is traditionally done using time

consuming equipment and/or techniques. An MLC calibration method (RDRL) was

created that was capable of quantitative and efficient measurement of the MLC. The

method is based on measuring each leaf position with a detector in a detector array

(PROFILER 2TM) that was previously aligned to the LINAC's radiation coordinate

system. The PROFILER 2TM was chosen due to the high spatial resolution of its diode

detectors. Multi-leaf collimator calibration using the RDRL method takes on average 30

minutes to complete, which is a marked improvement over traditional methods that can

take several hours and are not as quantitative.

Specific Aim 2: Water scans, by definition, are acquired using a scanning water

tank. This detector system, while time tested, is accompanied by inefficiencies and

unfamiliarity amongst clinical medical physicists. Detector arrays in contrast, are much

more familiar to the clinical physicist due to regular use and are inherently more

efficient. Accordingly, an ionization chamber detector array was characterized in a


126









radiation environment to assess it's abilities as a system for acquiring beam profiles that

are water tank equivalent. The detector array chosen was the IC PROFILERM, which

uses ionization chambers for its detectors. The IC PROFILERTM was found to be a

suitable system for acquiring beam profiles with an average error relative to a water

tank acquired profile of () 0.8%; this error was positively offset for certain measurement

conditions. The benefit of using the IC PROFILERTM over a scanning water tank to

acquire beam profiles is the time savings. The required time to acquire a series of

profile measurements was 180 minutes for a scanning water tank versus 30 minutes for

the IC PROFIELRTM. The time savings increases with the number of measured profiles.

The next step in this research is to begin evaluating the ability of the IC

PROFIELRTM to perform current tasks that are assigned to scanning water tanks.

Specifically, the abilities of the IC PROFILERTM to perform monthly and annual QA are

going to be investigated. The ability of the IC PROFILERTM as a commissioning device

will also be evaluated. While the data density is too low for direct measurements to be

used for TPS commissioning, various compensatory methods will be investigated. One

promising method involves fitting previously measured golden beam data to a small

subset of measurements that were made with the IC PROFILERTM. This is possible due

to the dosimetric similarity of modern LINACs, amongst a particular make and model.63'
64

Specific Aim 3: During the course of characterizing the IC PROFILERTM, it was

noticed that the WF array calibration produced unstable calibration factors when the

calibration was acquired on a beam with minor symmetry instabilities. A machine with

clinically insignificant symmetry variations [() 0.1%] can readily cause calibration errors


127









of () 2%. In contrast, a machine with symmetry variations on the order of () 0.05%

produced WF calibrations that varied by less than () 0.5%. This occurs because the

WF calibration procedure requires an invariant beam during each calibration

measurement. The stability of WF calibrations was increased by combining the use of a

continuously on LINAC beam during portions of the calibration procedure with the

addition of side-scatter to the IC PROFIELRTM. A continually on LINAC beam has a

reduced amount of symmetry variation when compared to a beam that is cycled on and

off. The reproducibility of the WF calibration increased from () 1.6% to () 0.5% on a

LINAC with symmetry variations of () 0.15%.

Further work needs to be done on the WF calibration to fully automate the

calibration procedure, increasing its reproducibility. Additionally, small changes that

occur in the spectral energy of the beam as the off-axis distance is increased should be

investigated for their impact on calibration accuracy, particularly with different amounts

of buildup.

Specific Aim 4: With the increased use of detector arrays comes an increase in

their importance in the clinic and a need to QA their performance. A QA program was

developed that was general to detector arrays and included suggestions for establishing

tolerances and frequencies for testing. Four QA regions were designated: physical,

software and firmware, electronic, and array calibration. The QA program was applied to

the IC PROFILERM, identifying a leaky chamber that occurred due to a contaminant in

the ionization chamber.

Due to the acknowledged lack of time to perform all of the recommended QA tests

for LINACs (and other modalities),15 the entirety of this work needs to be continued.


128









Ideally, more efficient measurement systems and methods should be developed to

perform every test that is recommended in TG-40 and TG-142. Detector arrays have

shown an adaptability and proficiency for streamlining the QA and radiation data

collection processes. It is the aim of this candidate to continue research in this

endeavor.


129









LIST OF REFERENCES


1 S. Henderson and P. Biggs, "The rate of evolution of radiation therapy planning and
delivery systems is exceeding the evolution rate of quality assurance processes," Med
Phys 26, 1439-1441 (1999).

2 T. R. Mackie, T. Holmes, S. Swerdloff, P. Reckwerdt, J. O. Deasy, J. Yang, B. Paliwal
and T. Kinsella, "Tomotherapy: a new concept for the delivery of dynamic conformal
radiotherapy," Med Phys 20, 1709-1719 (1993).

3 J. F. Dempsey, B. Dionne, J. Fitzsimmons, A. Haghigat, J. G. Li, D. Low, S. Mutic, J.
R. Palta, H. E. Romeijn and G. Sjoden, "A Real-Time MRI Guided External Beam
Radiotherapy Delivery System," Med Phys 33, 2254 (2006).

4 P. L. Petti, C. F. Chuang, V. Smith and D. A. Larson, "Peripheral doses in CyberKnife
radiosurgery," Med Phys 33, 1770-1779 (2006).

5 G. J. Kutcher, L. Coia, M. Gillin, W. F. Hanson, S. Leibel, R. J. Morton, J. R. Palta, J.
A. Purdy, L. E. Reinstein, G. K. Svensson and et al., "Comprehensive QA for radiation
oncology: report of AAPM Radiation Therapy Committee Task Group 40," Med Phys 21,
581-618 (1994).

6 W. Bogdanich, "Radiation Offers New Cures, and Ways to Do Harm," in The New
York Times, (New York, 2010), pp. Al.

7 C. Burman, C. S. Chui, G. Kutcher, S. Leibel, M. Zelefsky, T. LoSasso, S. Spirou, Q.
Wu, J. Yang, J. Stein, R. Mohan, Z. Fuks and C. C. Ling, "Planning, delivery, and quality
assurance of intensity-modulated radiotherapy using dynamic multileaf collimator: a
strategy for large-scale implementation for the treatment of carcinoma of the prostate,"
Int J Radiat Oncol Biol Phys 39, 863-873 (1997).

8 X. Wang, S. Spirou, T. LoSasso, J. Stein, C. S. Chui and B. Mohan, "Dosimetric
verification of intensity-modulated fields," Med Phys 23, 317-327 (1996).

9 V. M. Systems,
"http://www.varian.com/us/oncology/treatments/treatmenttechniques/rapidarc/,"
(2008).

10 S. N. Corporation, "http://www.sunnuclear.com/documents/ProCatB06252010.pdf,"
(2008).

11 A. Van Esch, C. Clermont, M. Devillers, M. lori and D. P. Huyskens, "On-line quality
assurance of rotational radiotherapy treatment delivery by means of a 2D ion chamber
array and the Octavius phantom," Med Phys 34, 3825-3837 (2007).

12 I. B. A., "http://www.iba-dosimetry.com/MULTICube. 1362.0.html," (2008).


130









13 S. N. Corporation, "http://www.sunnuclear.com/documents/arccheck.pdf," (2009).

14 ScandiDos, "http://www.scandidos.com/Documents/Brochures.aspx," Delta 4
(2009).

15 J. R. Palta, presented at the QA & Dosimetry Symposium, Lake Buena Vista, FL,
2009 (unpublished).

16 S. Huq, "The Philosophy of TG100: What It Is and What It Is Not," Med Phys 35,
2932 (2008).

17 E. E. Klein, J. Hanley, J. Bayouth, F. F. Yin, W. Simon, S. Dresser, C. Serago, F.
Aguirre, L. Ma, B. Arjomandy, C. Liu, C. Sandin and T. Holmes, "Task Group 142 report:
quality assurance of medical accelerators," Med Phys 36, 4197-4212 (2009).

18 B. D. Lynch, J. Kozelka, M. K. Ranade, J. G. Li, W. E. Simon and J. F. Dempsey,
"Important considerations for radiochromic film dosimetry with flatbed CCD scanners
and EBT GAFCHROMIC film," Med Phys 33, 4551-4556 (2006).

19 0. H. Suleiman, P. C. Bunch, P. F. Butler, J. E. Gray, A. G. Haus, M. H. P. McKetty,
E. L. Nickoloff, M. V. Pina, R. J. Pizzutiello, S. J. Shepar, D. C. Spelic, K. J. Strauss, S.
R. Szeglin and R. A. Uzenoff, Report No. 94, 2006.

20 A. Niroomand-Rad, C. R. Blackwell, B. M. Coursey, K. P. Gall, J. M. Galvin, W. L.
McLaughlin, A. S. Meigooni, R. Nath, J. E. Rodgers and C. G. Soares, "Radiochromic
film dosimetry: recommendations of AAPM Radiation Therapy Committee Task Group
55. American Association of Physicists in Medicine," Med Phys 25, 2093-2115 (1998).

21 S. Pai, I. J. Das, J. F. Dempsey, K. L. Lam, T. J. Losasso, A. J. Olch, J. R. Palta, L.
E. Reinstein, D. Ritt and E. E. Wilcox, "TG-69: radiographic film for megavoltage beam
dosimetry," Med Phys 34, 2228-2258 (2007).

22 B. E. Nelms and J. A. Simon, "A survey on planar IMRT QA analysis," J Appl Clin
Med Phys 8, 2448 (2007).

23 D. D. Leavitt and L. Larsson, "Evaluation of a diode detector array for measurement
of dynamic wedge dose distributions," Med Phys 20, 381-382 (1993).

24 H. H. Liu, E. P. Lief and E. C. McCullough, "Measuring dose distributions for
enhanced dynamic wedges using a multichamber detector array," Med Phys 24, 1515-
1519 (1997).

25 T. C. Zhu, L. Ding, C. R. Liu, J. R. Palta, W. E. Simon and J. Shi, "Performance
evaluation of a diode array for enhanced dynamic wedge dosimetry," Med Phys 24,
1173-1180 (1997).


131









26 S. N. Rustgi and D. M. Frye, "Dosimetric characterization of radiosurgical beams with
a diamond detector," Med Phys 22, 2117-2121 (1995).

27 G. Yan, C. Fox, C. Liu and J. G. Li, "The extraction of true profiles for TPS
commissioning and its impact on IMRT patient-specific QA," Med Phys 35, 3661-3670
(2008).

28 I. J. Das, C. W. Cheng, R. J. Watts, A. Ahnesjo, J. Gibbons, X. A. Li, J. Lowenstein,
R. K. Mitra, W. E. Simon and T. C. Zhu, "Accelerator beam data commissioning
equipment and procedures: report of the TG-106 of the Therapy Physics Committee of
the AAPM," Med Phys 35, 4186-4215 (2008).

29 G. Mu, E. Ludlum and P. Xia, "Impact of MLC leaf position errors on simple and
complex IMRT plans for head and neck cancer," Phys Med Biol 53, 77-88 (2008).

30 ICRU, Report No. 24, 1976.

31 A. Boyer, P. Biggs, J. M. Galvin, E. Klein, T. LoSasso, D. Low, K. Mah and C. Yu,
Report No. 72, 2001.

32 M. Pasquino, V. C. Borca, P. Catuzzo, F. Ozzello and S. Tofani, "Transmission,
penumbra and leaf positional accuracy in commissioning and quality assurance
program of a multileaf collimator for step-and-shoot IMRT treatments," Tumori 92, 511-
516(2006).

33 L. Parent, J. Seco, P. M. Evans, D. R. Dance and A. Fielding, "Evaluation of two
methods of predicting MLC leaf positions using EPID measurements," Med Phys 33,
3174-3182 (2006).

34 M. Mamalui-Hunter, H. Li and D. A. Low, "MLC quality assurance using EPID: a
fitting technique with subpixel precision," Med Phys 35, 2347-2355 (2008).

35 M. Mohammadi and E. Bezak, "Evaluation of MLC leaf positioning using a scanning
liquid ionization chamber EPID," Phys Med Biol 52, N21-33 (2007).

36 M. C. Lopes, A. Chaves and M. Capela, "A dosimetric calibration method for a
double-focused multileaf collimator," Med Phys 34, 3473-3474 (2007).

37 T. J. Jordan and P. C. Williams, "The design and performance characteristics of a
multileaf collimator," Phys Med Biol 39, 231-251 (1994).

38 J. R. Palta, D. K. Yeung and V. Frouhar, "Dosimetric considerations for a multileaf
collimator system," Med Phys 23, 1219-1224 (1996).

39 I. E. Commission, Report No. IEC 1217, 1996.


132









40 J. F. Dempsey, H. E. Romeijn, J. G. Li, D. A. Low and J. R. Palta, "A fourier analysis
of the dose grid resolution required for accurate IMRT fluence map optimization," Med
Phys 32, 380-388 (2005).

41 H. E. Johns and J. R. Cunningham, The physics of radiology, 4th ed. (Charles C.
Thomas, Springfield, Ill., U.S.A., 1983).

42 P. R. Almond, P. J. Biggs, B. M. Coursey, W. F. Hanson, M. S. Huq, R. Nath and D.
W. Rogers, "AAPM's TG-51 protocol for clinical reference dosimetry of high-energy
photon and electron beams," Med Phys 26, 1847-1870 (1999).

43 W. E. Simon, J. Shi and C. A. lannello, U.S.A. Patent No. 6125335 (2000).

44 R. J. Watts, "Evaluation of a diode detector array for use as a linear accelerator QC
device," Med Phys 25, 247-250 (1998).

45 D. Letourneau, M. Gulam, D. Yan, M. Oldham and J. W. Wong, "Evaluation of a 2D
diode array for IMRT quality assurance," Radiother Oncol 70, 199-206 (2004).

46 E. Yorke and AAPM Radiation Therapy Committee. Task Group No. 62., Diode in
vivo dosimetry for patients receiving external beam radiation therapy: report of Task
Group No. 62, Radiation Therapy Committee, AAPM. (Published by Medical Physics
Publishing for the American Association of Physicists in Medicine, College Park, MD,
2005).

47 G. F. Knoll, Radiation detection and measurement, 3rd ed. (Wiley, New York, 2000).

48 K. R. Kase, B. E. Bjarngard and F. H. Attix, The Dosimetry of ionizing radiation.
(Academic Press, Orlando, 1985).

49 L. Parent, A. L. Fielding, D. R. Dance, J. Seco and P. M. Evans, "Amorphous silicon
EPID calibration for dosimetric applications: comparison of a method based on Monte
Carlo prediction of response with existing techniques," Phys Med Biol 52, 3351-3368
(2007).

50 G. Yan, B. Lu, J. Kozelka, C. Liu and J. G. Li, "Calibration of a novel four-
dimensional diode array," Med Phys 37, 108-115 (2010).

51 S. Amerio, A. Boriano, F. Bourhaleb, R. Cirio, M. Donetti, A. Fidanzio, E. Garelli, S.
Giordanengo, E. Madon, F. Marchetto, U. Nastasi, C. Peroni, A. Piermattei, C. J. Sanz
Freire, A. Sardo and E. Trevisiol, "Dosimetric characterization of a large area pixel-
segmented ionization chamber," Med Phys 31, 414-420 (2004).

52 R. Sadagopan, J. A. Bencomo, R. L. Martin, G. Nilsson, T. Matzen and P. A. Baiter,
"Characterization and clinical evaluation of a novel IMRT quality assurance system," J
Appl Clin Med Phys 10, 104-119 (2009).


133










53 T. A. Simon, W. E. Simon, D. Kahler, J. G. Li and C. Liu, "Characterization of a multi-
axis ion chamber array," Med Phys Under Review (2010).

54 J. P. McCaffrey, B. Downtown, H. Shen, D. Niven and M. McEwen, "Pre-irradiation
effects on ionization chambers used in radiation therapy," Phys Med Biol 50, N121-133
(2005).

55 B. Poppe, A. Blechschmidt, A. Djouguela, R. Kollhoff, A. Rubach, K. C. Willborn and
D. Harder, "Two-dimensional ionization chamber arrays for IMRT plan verification," Med
Phys 33, 1005-1015 (2006).

56 M. Stasi, S. Giordanengo, R. Cirio, A. Boriano, F. Bourhaleb, I. Cornelius, M. Donetti,
E. Garelli, I. Gomola, F. Marchetto, M. Porzio, C. J. Sanz Freire, A. Sardo and C.
Peroni, "D-IMRT verification with a 2D pixel ionization chamber: dosimetric and clinical
results in head and neck cancer," Phys Med Biol 50, 4681-4694 (2005).

57 P. A. Jursinic and B. E. Nelms, "A 2-D diode array and analysis software for
verification of intensity modulated radiation therapy delivery," Med Phys 30, 870-879
(2003).

58 E. Spezi, A. L. Angelini, F. Romani and A. Ferri, "Characterization of a 2D ion
chamber array for the verification of radiotherapy treatments," Phys Med Biol 50, 3361-
3373 (2005).

59 T. A. Simon, W. E. Simon, D. Kahler, J. G. Li and C. Liu, "Wide field array calibration
dependence on the stability of measured dose distributions," Med Phys Under Review
(2010).

60 J. L. Devore, Probability and statistics for engineering and the sciences, 7th ed.
(Thomson/Brooks/Cole, Belmont, CA, 2008).

61 T. W. Anderson and D. A. Darling, "A Test of Goodness of Fit," Journal of the
American Statistical Association 49, 765-769 (1954).

62 M. A. Stephens, "EDF Statistics for Goodness of Fit and Some Comparisons,"
Journal of the American Statistical Association 69, 730-737 (1974).

63 D. A. Jaffray, J. J. Battista, A. Fenster and P. Munro, "X-ray sources of medical linear
accelerators: focal and extra-focal radiation," Med Phys 20, 1417-1427 (1993).

64 R. J. Watts, "Comparative measurements on a series of accelerators by the same
vendor," Med Phys 26, 2581-2585 (1999).


134









BIOGRAPHICAL SKETCH

Thomas Allan Simon was born in Euclid, Ohio. In 1985, his family moved to

Satellite Beach, Florida. He graduated from Satellite High School in June of 1997 and

then enrolled at the Florida State University. After one year he transferred to the

University of Florida to pursue a degree in microbiology, graduating with a B.S. in 2002.

In 2002 he enrolled in the medical physics program of the Department of Nuclear

and Radiological Engineering at the University of Florida. His initial research involved

anthropomorphic phantom construction with Drs. David Hintenlang and Wesley Bolch.

In 2005 he passed the qualifying exam and began working for his doctoral advisor Dr.

Chihray Liu.

In 2006, he became involved in a small business innovative research grant that

was tasked with collecting golden beam data sets (also serving as benchmark data) for

each of the three primary linear accelerator (LINAC) manufacturers (Elekta, Siemens,

and Varian). During this time he was educated in and first began to appreciate

metrology.

Upon graduating, he plans on moving to Melbourne, FL to work at Sun Nuclear

Corporation on a calibration laboratory. His current research interests lie in array

calibration methodology and investigating the boundaries of replacing water tanks with

detector arrays.


135





PAGE 1

USING DETECTOR ARRAYS TO IMPRO VE THE EFFICIENCY OF LINEAR ACCELERATOR QUALITY ASSURANCE AND RADIATION DATA COLLECTION By THOMAS ALLAN SIMON A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2010 1

PAGE 2

2010 Thomas Allan Simon 2

PAGE 3

To my darling wife 3

PAGE 4

ACKNOWLEDGMENTS First, I would like to thank my advisor, Dr. Chihray Liu, for cont inually showing me how to be a good physicist and a good person th rough his direct demonstration. Next, I would also like to thank my committee memb ers, Drs. Jonathan Li, Sanjiv Samant, and William Mendenhall for their guidance and support. Additionally, I would like to thank Dr. Darren Kahler for help in preparing this manuscript and Phil Basset for useful discussions about linear accelerator operation. Then there are my fellow graduate student s and colleagues. Useful discussions and even more useful distractions were a necessity during eight long years of graduate school. I would particularly like to acknow ledge Kevin Segall, Ch ristopher Fox, Bart Lynch, Heeteak Chung, Guanghua Yan, Jean Peng, and Matthew Williams. Finally, I am indebted to my family fo r their continued support. My wife, Naomi, deserves more credit that words can prov ide for always encouraging and supporting me. My brother, Jeff, and his wife and my friend, Mariel, were always there to push me. This leaves my parents, Bill and Cathy. Bills guidance and encouragement was my guiding light. Cathys pride in me, despite t oo many years in graduate school, kept me from realizing that it was indeed too many years. 4

PAGE 5

TABLE OF CONTENTS page ACKNOWLEDG MENTS..................................................................................................4 LIST OF TABLES............................................................................................................8 LIST OF FI GURES..........................................................................................................9 LIST OF ABBR EVIATIONS...........................................................................................11 ABSTRACT ...................................................................................................................13 CHAPTER 1 INTRODUC TION....................................................................................................15 General Intr oduction ...............................................................................................15 New Treatment Modalities Increase a Fa cilitys Resource Requirements........16 How Do We Solve this Probl em?.....................................................................17 Streamlining Data Collec tion............................................................................19 Study Ai ms..............................................................................................................20 2 MLC CALIBRATION USING A DETECTOR ARRAY..............................................23 Introducti on.............................................................................................................23 Materials and Methods............................................................................................24 Materials...........................................................................................................24 Linear accelera tor and ML C.......................................................................24 Detector arrays..........................................................................................25 Methods ............................................................................................................26 Measuring minor l eaf offs ets......................................................................27 Measuring major l eaf offs ets......................................................................33 MLC calibra tion..........................................................................................36 Other MLC types........................................................................................37 Result s....................................................................................................................37 Detector Of fsets...............................................................................................37 MLC Measurement Comparis on and Reproduc ibility.......................................38 MLC Service Issues.........................................................................................39 Discussio n..............................................................................................................39 MLC Calibration St ability and QA .....................................................................39 Device Com parisons ........................................................................................40 Conclusi on..............................................................................................................42 3 CHARACTERIZATION OF A MULTI-AXI S IONIZATION CHAMBER ARRAY.......47 Introducti on.............................................................................................................47 5

PAGE 6

Materials and Methods............................................................................................48 Materials...........................................................................................................48 Methods ............................................................................................................49 Reproducib ility...........................................................................................50 Dose and instantaneous dose rate dependence ........................................50 PRF depen dence.......................................................................................52 Energy dep endence ...................................................................................53 Response to power being app lied to the el ectronics..................................54 Calibration c onstanc y.................................................................................55 Backscatter dependence ...........................................................................56 Beam profile measurement s and output fa ctors.........................................57 Results and Discussion...........................................................................................58 Reproducib ility..................................................................................................58 Dose and Instantaneous Do se Rate D ependence ...........................................58 PRF Depe ndence.............................................................................................61 Energy De pendence .........................................................................................63 Response to Power Appli ed to the El ectronics.................................................64 Calibration C onstanc y......................................................................................64 Backscatter D ependence .................................................................................65 Beam Profile Measurements and Output Factors.............................................66 Conclusi on..............................................................................................................67 4 WIDE FIELD ARRAY CALIBRATION DEPENDENCE ON THE STABILITY OF MEASURED DOSE DIST RIBUTION S....................................................................78 Introducti on.............................................................................................................78 Materials and Methods............................................................................................79 Materials...........................................................................................................79 Methods ............................................................................................................80 Wide field calibra tion t heory.......................................................................80 Limiting calibrat ion erro r.............................................................................83 Effects of postul ate fail ure..........................................................................83 Limiting violations of the first postulate ......................................................84 Limiting violations of the second pos tulate.................................................85 Limiting violations of the third pos tulate .....................................................86 Evaluating calibrati on factors.....................................................................87 Results and Discussion...........................................................................................90 Effects of Post ulate Failure...............................................................................90 Limiting Violations of the Fi rst Calibration Postulate.........................................91 Limiting Violations of the Th ird Calibration Postulate.......................................92 Evaluating Calibrat ion Fact ors..........................................................................93 Other Factors Affecting the WF Calib ration......................................................94 Other Arrays and IMRT....................................................................................95 Conclusi on..............................................................................................................95 5 A QUALITY ASSURANCE PROGRAM FOR A DETECTOR ARRAY...................102 6

PAGE 7

Introducti on...........................................................................................................102 Materials and Methods..........................................................................................103 Materials.........................................................................................................103 Methods ..........................................................................................................104 Physical...................................................................................................104 Firmware and so ftware............................................................................105 Electroni cs...............................................................................................108 Array calibra tion.......................................................................................111 Results and Di scussion.........................................................................................114 Physical ..........................................................................................................114 Firmware and So ftware..................................................................................115 Electroni cs......................................................................................................115 Array Calibra tion.............................................................................................119 Conclusi on............................................................................................................120 6 SUMMARY AND FU TURE WORK.......................................................................126 LIST OF REFE RENCES.............................................................................................130 BIOGRAPHICAL SKETCH ..........................................................................................135 7

PAGE 8

LIST OF TABLES Table page 2-1 List of collimator sett ings for PROF ILER 2......................................................43 2-2 List of additional collimator setting s for PROFILER 2 measurements of the major leaf offsets................................................................................................43 3-1 List of subscripts, variables, and equati ons........................................................69 3-2 The panels short and long-term r eproducibility were evaluated on a 60Co teletherapy unit...................................................................................................70 3-3 The short and long term reproducibility of the relative detector calibration factor s.................................................................................................................70 4-1 The short term reproducibility of WF calibrations performed on the Elekta........97 5-1 Quality assurance program for the IC PR OFILER........................................122 5-2 Percent of detectors rejecting H0......................................................................122 8

PAGE 9

LIST OF FIGURES Figure page 2-1 Minor leaf offsets are def ined as the spatial offset of each leaf in a leaf.............44 2-2 (A) The radiation defined refer ence line (RDRL) me thod requires......................44 2-3 The measurements required to r adiographically ali gn the a rray.........................45 2-4 (A) Detector offsets relati ve to a reference detec tor...........................................45 2-5 MLC leaf offsets relative to l eaf number 20, the re ference l eaf..........................46 2-6 Change in relative MLC offsets af ter replacement of the prim ary.......................46 2-7 Calibration of the X1 MLC leaf bank...................................................................46 3-1 Overlay of the IC PROFILER ( panel) showing the mult iple detector...............71 3-2 The panels dose response relative to the Farmer-type chambers dose...........71 3-3 The panels instantaneous dose rate response relative to the Farmer-..............72 3-4 The panels PRF response relative to the Farmer-type chambers PRF.............72 3-5 The off-axis PRF response for the x-axis detectors relative to the center..........73 3-6 The difference between the 180 and 0 OA_Response ( PRF ) values for..........73 3-7 The energy response of the panels cent er detector presented as a ratio..........73 3-8 The accuracy of the calibration fa ctors...............................................................74 3-9 The energy response of the calibration factors...................................................74 3-10 The buildup response of the ca libration factors for 6 and 18 MV........................74 3-11 Off axis backscatter response of the x-axis detectors for (A) a 6 MV.................75 3-12 Normalized cross-plane measurem ents with a CC13 and the panel...............75 3-13 Profile agreement (over 80% of th e field width) betw een the panel and.............76 3-14 FDDs for a 6 MV 10 x 10 cm2 fiel d.....................................................................76 3-15 Output factors meas ured with the panels center chamber and three.................77 4-1 Wide field (WF) calibration r eproducibility on LI NACs wit h beam .......................98 9

PAGE 10

4-2 Oblique view of the panels arrays and electronics. The panels y-axis is..........98 4-3 (A) The perturbation that was appli ed to the hypothetical calibra tion.................99 4-4 The percentage error between ten cons ecutive measurements and their..........99 4-5 Calibration reproducibility us ing a continuous beam during..............................100 4-6 The effect of additional side-scatter on (A) beam measurements and (B)........100 4-7 The agreement between calibration factors obtai ned with side-scatter............100 4-8 The calibration accuracy was ev aluated for four measurem ent........................101 4-9 The calibration accuracy expressed as the ratio between water tank...............101 5-1 Cable damage as a result of (A) disc onnecting the cable by pulling the...........123 5-2 The measurement reproducibility for the panel on the Elekta Synergy..........123 5-3 (A) An example of two background m easurements that were separated.........124 5-4 The PDF of UC for the (-) 12.5 cm x-axis detector (X8)....................................124 5-5 The accuracy of the panels calibration factors were dete rmined using............125 5-6 (A) Profile measurements with the panel and a scanning water tank...............125 10

PAGE 11

LIST OF ABBREVIATIONS AAPM American Association of Physicists in Medicine CAX Central axis of the LINAC coordinate system CCD Charge coupled device camera us ed in the control of the MLC EPID Electronic portal imaging device EDW Enhanced dynamic wedge FDD Fractional depth dose FMEA Failure mode and effects analysis IBA Ion Beam Applications (equipment manufacturer) IMRT Intensity modulat ed radiation therapy LINAC Linear accelerator LSI Local standard instrument MALO Major leaf offset MILO Minor leaf offset MLC Multi-leaf collimator NIST Nation Institute of Standards and Technology panel IC PROFILER PTW Physikalisch-Technische Werkst tten (equipment manufacturer) QA Quality assurance RDRL Radiation defined reference line method SA1 Specific aim 1 SA2 Specific aim 2 SA3 Specific aim 3 SA4 Specific aim 4 SNC Sun Nuclear Corporati on (equipment manufacturer) 11

PAGE 12

SDD Source to detector distance SSD Source to surface distance TG Task group TPS Treatment planning system 12

PAGE 13

Abstract of Dissertation Pr esented to the Graduate School of the University of Florida in Partial Fulf illment of the Requirements for t he Degree of Doctor of Philosophy USING DETECTOR ARRAYS TO IMPRO VE THE EFFICIENCY OF LINEAR ACCELERATOR QUALITY ASSURANCE AND RADIATION DATA COLLECTION By Thomas Allan Simon August 2010 Chair: Chihray Liu Major: Nuclear Engineering Sciences The complexity of radiation therapy is continually increasing as new treatment modalities are implemented in the clinic. While these advances often benefit tumor dose localization, they also increase pressu re on departmental resources as the new modality is adopted. This driving force comes at a time of increased pressure to perform quality assurance (QA) of the entire treatment process. The effect is a work force with too many measurements to do and not enough time in which to do them. The purpose of this work is to establish the use of detector arrays to improve the automation and efficiency of linear accelerator (LINAC) quality assurance and radiation data collection. Two traditionally time consuming measurem ent processes were evaluated for the potential for increased efficiency and automation: multi-leaf collimator (MLC) calibration and scanning water tank measurements. Usi ng traditional measurement techniques, MLC calibration can take hours to accomplish with mixed results or require a significant investment of time to write in-house softw are. We developed a quantitative and efficient (less than 30 minutes for both leaf banks) MLC calibration method that we termed the radiation defined reference line (RDRL) me thod. The method uses a detector array 13

PAGE 14

14 [PROFILER 2; Sun Nuclear Corporation (S NC), Melbourne, FL USA] to measure the penumbral position of each leaf relative to a known reference point (or line). Profile measurements are typically obtained with a scanning water tank. While time tested, the sys tem requires above average skill an d time to properly setup and acquire data. We extensively characteriz ed and assessed the potential of a multi-axis ionization chamber array (IC PROFILER; S NC) to measure water tank equivalent profiles. The IC PROFILER had an error spr ead of approximately () 0.75% relative to a water scan, with the potentia l of a positive offset in that error. During the characterization, the array calibration method was found to be susceptible to the LINACs symmetry stability. Symmetry variations of () 0.1% can cause calibration errors of () 2%. The cause was investigat ed and corrective measures were developed. Finally, a time efficient QA program was developed to determine the operation of the detector arrays.

PAGE 15

CHAPTER 1 INTRODUCTION General Introduction The complexity of radiation therapy is continually increasing as new treatment modalities are implemented in the clinic. While these advances often benefit tumor dose localization, they also increase pressu re on departmental resources as the new modality is adopted.1 An example of this cause and effe ct is the use of dynamic wedges over the more traditional physical wedge. The enhanced dynamic wedge (EDW) incr eased the efficiency of radiation treatment by eliminating the need to enter the treatment vault to install a wedge. Instead of a physical beam attenuator, the wedge intens ity pattern is created by dynamically sweeping the collimator across the field while delivering r adiation. The EDW increased treatment efficiency; but it also complic ated the radiation data collection process. Measuring an EDW with a scanning water tank requires each profile point to be integrated through an entire de livery. Measuring a range of field sizes, depths, and energies could extend to several days. To cope with the increased measurement requi rements, the clinical physicist relied on a new measurement technology detector arrays. The use of detector arrays greatly increased the efficiency of EDW measurem ents due to their simultaneous measurement of multiple points. What previously took scanning water tanks several days to measure was reduced to a few hours with detector arrays. Just as dynamic wedges increased the complexity of radiation therapy, an abundance of new treatment modalit ies are poised to do the same.2-4 This comes at a time when the demands on the physi cs department are already high.5, 6 To adapt and 15

PAGE 16

evolve with the shifting clinical envi ronment, we need to reevaluate how quality assurance (QA) and radiation da ta collection is performed. The purpose of this work is to establish the use of detector arrays to improve the automation and efficiency of LINAC QA and radiation data collection. New Treatment Modalities Increase a Fa cilitys Resource Requirements New treatment modalities in radiation therapy are becoming increasingly complex through the incorporation of various imaging, treatment planning sy stem (TPS), patient localization, and delivery options.2-4 While the level of patient care is improved, the overall complexity and verification requireme nts for those treatment s increase as well. This cause and effect is usually accompani ed with a corresponding lag in the evolution of QA technology.1 An example lies in the introducti on of intensity modulated radiation therapy (IMRT). The radiation therapy process tree grew in complexity with the adoption of IMRT. It affected all aspects of the clinic, but it gr eatly increased the time demand on physicists in the form of treatment verification.7 During its initial introduc tion, the primary method for verifying two dimensional dose distributions was film dosimetry.7, 8 While this verification process provided high-spatialdensity-dosimetric-in formation, it was inefficient considering the multi-field, multifilm nature of IMRT QA A simple seven field prostate plan could take up to two hours to verify. An emerging modality that threatens to further increase complexity is single arc IMRT. This combines the principles of dynamic IMRT with the added complexity of continuously moving gantry components and a varying dose rate.9 Just as film was initially used to verify IMRT plans, the t ools and methods that we re developed for IMRT are now being used to verify arc therapies.10-12 While these technologies may prove to 16

PAGE 17

be just as arduous as film dosimetry was to IMRT, new dosimetry systems are already emerging that promise to efficiently handle the QA requirements of single arc IMRT.13, 14 How Do We Solve this Problem? With each new treatment modality, new verification requirements are added to the existing regime. So many tests already exist that it would be nearly impossible to do them all.15 This begs the question is all of this QA needed? For example, if a physicist performs annual QA that takes 10 hours and no problems were discovered, then were those 10 hours wasted? Could they have been spent on an item that is more likely to fail? Does this mean we should not perfo rm annual QA? No; it means we need to approach certain aspects of QA and data collection differently. Looking at QA in a different light is already being addressed. The American Association of Physicists in Medicine (AAPM) is taking two separate approaches, led by Task Groups (TGs) 100 and 142.16, 17 The approach of TG-100 is to apply Failure Mode and Effects Analysis (FMEA) to the radiatio n therapy process. This process is a powerful tool that aids in the allocation of resources to a system in an effort to detect modes of failure before they occur. The goal of TG-100 is for eac h radiation oncology clinic to compose a personalized FMEA analysi s for each treatment process tree in its clinic. While a completed FMEA analysis woul d help to focus departmental resources, the creation of the FMEA analysis will greatly increase the time requirements on the clinic and its physicists. The approach of TG-142 is somewhat diffe rent. Similar to TG-40, it provides updated machine item tolerances. Unlike TG-40, it recommends different tolerances for different modalities. While addressing that conventional radiotherapy LINACs do not need the same mechanical tolerances that st ereotactic-radio-surgery LINACs do, TG17

PAGE 18

142 fails to address the associated time load that already exists. As new modalities are introduced, the time load will only increase. Both TG-100 and 142 provide useful QA recommendations and insights. However, the overall result is likely to be an increas e in time and measurem ent requirements. We feel that measurement automat ion and increased efficiency is the key to aid the clinical physicist. Again, dosimetric verification of IMRT illu strates an example. A typical seven field prostate plan requires nearly 2 hours for verifi cation using film. There is also a high potential for error in the film dosimetry proc ess that can be a source of inconsistency throughout the community.18-21 The introduction of two dimensional detector arrays streamlined the IMRT QA proc ess. The effect was a reducti on in the time requirements from nearly two hours to 45 minutes for a typical seven field prostate exam. The acquisition of the results was also standardiz ed to a certain extent with the 2D arrays.22 This provided the community with a much more consistent reporting of results and also an increase in care. A second example exists with the measurem ent of EDWs. Initially, scanning water tanks were used to measure EDWs wit h a series of point measurements.23 Each measurement point represent ed the duration of an EDW delivery. This was a time intensive undertaking ev en without factoring in the time of tank setup and break-down. The whole process could easily take seve ral days to complete. The introduction of detector arrays that are mounted in a scanning water tank decreased the time requirements by providing multiple integration points per measurement.23, 24 However, 18

PAGE 19

the process still required a scanning water tank The introduction of a detector array that lay on the treatment reduced t he required time to minutes.25 Streamlining Data Collection Automating and streamlining the QA and data collection regimen is a daunting task due to the variety of measurements and de livery systems. It is nearly impossible for one agency or group to solve all of these pr oblems. However, a published group of tests that the agencies could recommend would help to solve this problem. This would allow for a much faster response time as QA and data collection demands shifted. As an example, two methods that have traditionally been resource intensive were investigated for the possibilit y of streamlining. They ar e the calibration of MLCs and scanning water tank measurements. The current methods for calibrating MLCs include the use of graph paper, film, electronic portal imaging devic es (EPIDs), and scanning wa ter tanks. Each of these measurement methods has certain advantages and disadvantages. Graph paper and film are both time tested and intuitive. However, they are resource intensive and may only provide qualitative result s. Water tanks are well under stood with highly accurate and repeatable mechanics, but require a lar ge amount of resources and suffer from detector volume averaging unless a pinpoint (diamond,26 diode detector,27 etc.) detector is used. The calibration of MLCs with any of these methods takes many hours to complete. A dosimeter that is more efficient is the EPID. Unfortunately, they require user written code due to a relative lack of commercial software. Detector arrays have the potential to reduce that time requirement from hours to minutes. Scanning water tanks are t he gold standard of radiatio n therapy measurements. They are time tested, precise, and repr oducible. They do however require large 19

PAGE 20

amounts of time for se tup and data collection.28 They also require a higher degree of skill to accurately use.28 The clinical physicist therefore rarely uses the scanning water tank. Most are only used during t he LINAC beam commissioning and annual QA. Advances in computer technology have improved their efficiency. However, they still require large amounts of preparation time (~ 1 hour s to setup and break down) and actual scanning time [several days to measure linear accelerator (LINAC) beam commissioning data].28 Infrequent use by the clinical physicist adds to these inefficiencies and decreases the likelihood of obtaining quality data. The skill required to obtain quality beam profile m easurements with a detector array is less. This is due to physicists being more familiar with detecto r panels and also due to their lack of mechanical parts and liquid water. Study Aims For this dissertation, two time consum ing measurement processes were chosen as a showcase for the potential of detecto r arrays to increase measurement automation and efficiency. The first specific aim covers the calibration of multi-leaf-collimators (MLC). The remaining specific aims deal with using the IC PROFILER as a water tank alternative. Specific Aim 1 (SA1) MLC calibration with detector arra ys: Each leaf end creates a penumbral position that corresponds to its actual position. An array that uses a detector with minimal volume averaging can accu rately measure these leaf positions for QA or calibration purposes. The purpose of this aim is to create an efficient and quantitative MLC calibration method that uses a commercially available detector array (e.g. the PROFILER 2 or an EPID). 20

PAGE 21

Specific Aim 2 (SA2) Characterize t he IC PROFILER: The IC PROFILER is a multi-axis ion chamber array and ther efore does not suffer from the undesirable detector characteristics that diode detectors possess. However, its potential as a water tank alternative underlies the importance of fully understandi ng the device and how it reacts in a radiation environment. The purpose of this aim is to do just that; extensively characterize the IC PROFILER in a radi ation environment and est ablish its ability to measure LINAC beam parameters. Specific Aim 3 (SA3) In crease the reproducibility of the wide field calibration theory for use in unstable beams: The wi de field calibration theory has become a prominent fixture in the radi ation oncology environment. It is used to correct the intradetector-sensitivity-variation in a wide va riety of detector arrays, including the MapCHECK, PROFILER 2, and IC PROF ILER. Accurate measurements with these systems require confidenc e in the individual detectors calibrations. However, the calibration theory requires a perfect ly reproducible LINAC beam; otherwise unacceptable error levels are encountered. The purpose of this specific aim is to minimize the effects of beam instability in the wide field calibration theory. Specific Aim 4 (SA4) Establish a quality assurance program for the IC PROFILER: The potential im portance of the dat a provided by the IC PROFILER (i.e. annual QA and LINAC commissioning) requi res the highest level of confidence in the array and its measured data. The purpose of this specific aim is to develop a series of field tests that indicate proper function of the IC PROFILER and detector arrays in general. In summary, this dissertation is or ganized into four specific aims: 21

PAGE 22

22 SA1: Create an efficient MLC calibration method using detector arrays (Chapter 2). SA2: Characterize the IC PROFILER in the radiation environment (Chapter 3). SA3: Increase the reproducibility of the Wide Field Calibration theory when operated in beams with micro in stabilities (Chapter 4). SA4: Establish a QA program for the IC PROFILER and detector arrays in general (Chapter 5).

PAGE 23

CHAPTER 2 MLC CALIBRATION USING A DETECTOR ARRAY Introduction Intensity modulated radiation therapy (IMRT) is a treatment modality that is used to deliver a dose prescription to a tumor site while minimizing the exposure to surrounding healthy tissues. A popular method of implementing IMRT is to superpose a series of irregular fields that are shaped with a multi leaf collimator (MLC) to create a complex radiation fluence map. The principles of radiation transport then govern the conversion of the fluence map to a dose map. The correct placement of high-gradient dose regions in and near the target volume is dependent on an accurate positioning of t he MLC leaves during the delivery of the IMRT fields. In a recent study, Mu et al. demonstrated that a systematic leaf positioning error of 1 mm in IMRT plans can result in dose errors of up to 7.6 % and 12.2 % for the target and critical structures, respectively.29 Errors of this size can have a biologically significant effect on the outcome of the therapy.30 For this reason the MLC must be accurately calibrated and periodically tested. MLC calibration requires the ability to prec isely measure individual leaf positions. Traditional methods of calibration are time consuming and/or non-reproducible in nature. These methods in clude the use of graph paper,31 radiosensitive film,31, 32 scanning water tanks, electronic portal imaging devices (EPID)s,33-35 detector arrays,36 and manufacturers proprietary methods. While each of these techniques has advantages and disadvantages, the current trend is toward more efficient and reproducible methods. 23

PAGE 24

Recent publications have shown a refi nement in measurement and calibration techniques. In 2006 Parent et al. used an EPID to measure and predict the positions of individual leaves on an Elekta MLC.33 In 2007 Lopes et al. used an ion-chamber array mounted in a water tank to calibrate the i ndividual leaf positions for a Siemens MLC.36 While both of these techniques are an improv ement on the traditional methods, they still require a significant investment of time Using an EPID to meas ure leaf positions requires mechanical and/or software corrections as well as user-written code. The ion chamber array-based approach is susceptible to the volume averaging of the ion chambers and requires the setup of a scanning water tank along with ancillary equipment. An integrated technique for MLC calibrati on exists as a proprietary method for Elekta MLCs. The AutoCAL (Elekta Oncology Systems, Crawley, UK) software suite uses EPID measurements to calibrate various machine items. This software has only recently become available and is used exclus ively with the Elekta EPID. While it represents an important st ep toward more efficient and quantitative calibration techniques, our initial uses have shown that it is prone to delays and calibrations with unacceptable leaf positions. The purpose of our research was to dev elop a more efficient and reproducible method for calibrating an MLC. Materials and Methods Materials Linear accelerator and MLC All tests were performed with an Elekta Synergy (Elekta Oncology Systems, Crawley, UK) linear accele rator (LINAC) using the 6 MV photon beam. The LINACs 24

PAGE 25

MLC is a 40 leaf-pair device that has been described in detail.37, 38 Each leaf projects to a width of 1 cm at the isocentric plane (100 cm from source). Each MLC leaf bank is located above a backup jaw that aids in beam collimation and reduces MLC radiation transmission. An MLC leaf bank and its a ssociated backup jaw have parallel leading edges that travel in the cross-plane directio n when the collimator is set to 0 degrees (IEC 1217 convention).39 Elekta leaf positions are controlled through an optical system that uses field light reflected from a marker on top of each lea f. Reference reflectors are located in the machine head outside of the largest obtainable field and are used to define the MLC coordinate system. The reflected light rays trac e through a series of mirrors to a charge coupled device (CCD) camera that is interfaced to a control computer. Detector arrays The PROFILER 2 is a two-axis detec tor array (Sun Nuclear Corporation, Melbourne, FL) that consists of 139 diode detec tors. The y-axis of the device contains 83 detectors over a length of 32.8 cm and the xaxis contains 57 detectors over a length of 22.4 cm. Both axes have a detector spacing of 4 mm and share a central detector. The inherent buildup of the device is 1 g/cm2 of water-equivalent material. The device was chosen due to the detector spacing and t he spatial measurement resolution of each detector (0.8 0.8 mm2). Data collected using the PROFILER 2 software can be transferred (using copy and paste) to a spread sheet program such as Excel (Microsoft, Redmond, Washington) for analysis. The EPID used in this study is an Elekta iView GT. It has a pixe l dimension of 0.4 0.4 mm2 and a sensitive area of 41 41 cm2. It operates at a fixed source to surface distance (SSD) of 160 cm. The iView softwar e automatically projects collected images 25

PAGE 26

to the isocentric plane by scaling the pixe l and field dimensions to 0.25 0.25 mm2 and 25.6 25.6 cm2, respectively. Since an MLC leaf bank projects to a maximum length of 40 cm at isocenter, it is necessary to sh ift the EPID in order to fully image one MLC bank. The method described herein is specific to the Elekta MLC. However, the principle is general and can be applied to other manufacturers MLCs provided that appropriate conditions for measurements are met. Methods We have termed this measurement techni que the radiation defined reference line (RDRL) method. Application of the method oper ates under three assumptions. First, the leading edge of an MLC leaf bank is parallel to its back up jaws leading edge. Second, the backup jaw can provide a reproducible and uniform radiation field edge. This field edge defines the RDRL. Third, the measured radiation field edge created by each leaf end is representative of that leafs position. The third assumption of the RDRL met hod requires detectors with a spatial measurement resolution that does not suffer from signal averaging in the high spatial frequency of the penumbra. Dempsey has s hown that measurements with a detector size of 2 mm or smaller is suffici ent for IMRT fields shaped with MLCs.40 The PROFILER 2s detector size satisfies this requirement, but the detector location must also be known with a precision better t han the desired leaf position accuracy. Elekta MLC leaf banks have traditionally been calibrated using standard measurement tools, e.g. f ilm and scanning water tanks, to determine what are termed major and minor leaf offsets, as illustrated in Fig. 2-1 A reference leaf pair, leaf pair 20, is used in the control of the MLC. The majo r leaf offset (MALO) is a calibration value 26

PAGE 27

that defines the field size created by the re ference leaf pair. Minor leaf offsets (MILO) are the position alignment errors of the other leaves in relation to the reference leaf and are the first focus of this method. Measuring minor leaf offsets The method described below uses the jaw edge to precisely locate all of the y-axis detectors offsets relative to the refe rence detector, as illustrated in Fig. 2-2 A; the reference detector is located in the reference leafs direction of travel. These relative detector positions are termed RDOj, where j is the y-axis diode number 1 j 83; they effectively create a uniform RDRL. Once t hese detector positions are known, the detector array is used to measure the position of each leaf that results in a field edge at detector j as seen in Fig. 2-2 B. PROFILER 2 The procedure that follows describes measuring the minor leaf offsets for the X1 leaf bank; the procedure is repeated for the X2 leaf bank but with appr opriate collimator configurations. Three main steps were requi red to measure the minor leaf offsets with the PROFILER 2 and the RDRL method: devic e setup, detector offset correction, and MLC measurement. Step 1 Device setup: The collimator is ro tated to 180 and the PROFILER 2 is set on the treatment table at a source to surface dist ance, SSD, of 79 cm and a corresponding source to detector distance (S DD) of 80 cm. The PROFILER 2s x and y axes are then aligned with the collimator crosshair shadow such that the positive yaxis of the PROFILER 2 points toward the gantry. This orientation places the y-axis co lumn of 83 detectors perpendicular to the direction of leaf movement. The orientati on also directionally matches the ascending 27

PAGE 28

numerical order of the detectors (1-83) with t hat of the leaves (1-40). This directional match makes the on-screen evaluation easier during data collection with the PROFILER 2 software. The same effect could have been achieved by rotating the PROFILER 2 instead of the collimator. Next, the PROFILER 2 is shifted by 2 mm in the direction of its negative y-axis. This centers the crosshairs between the central y-axis dete ctor and its immediate upper neighbor. Since the projected l eaf width at 80 cm from the source is 8 mm, this shift locates two detectors in the projection of each leaf, as seen in Fig. 2-2 B, and positions the x-axis of the PROFILER 2 in the projection of the reference leaf. Small rotational errors between the cro sshair and the field edge of the backup jaw are tested by moving one of the MLCs backup jaws to 3 mm before the central axis. A small consistent gap between the detector markers and the backup-jaws-field-edge indicates a proper alignment, while a dive rgence between the two is corrected by rotating the PROFILER 2. Since a visual alignment of the ar ray was inadequate for a precise MLC calibration, an alignment to the LINACs radi ation coordinate system is also performed. This is accomplished by measuring an ali gnment profile with the column of y-axis detectors. Before measuring the alignment pr ofile, a y-axis profile is measured with both jaws opened. This open field profile is used to reference the slope of the alignment profile. The jaw and leaf bank settings that we re used for the open field are shown in Table 2-1 The alignment profile m easurement is taken with the leaves retracted and with the backup jaw positioned at the central axis (CAX) so that the measured profile 28

PAGE 29

lies within the jaws penumbra. The jaw and leaf bank settings that were used to measure the alignment prof ile for the X1 backup jaw are also shown in Table 2-1 A tilt in the initial alig nment profile indicates that the PROFILER 2 was not aligned with the backup jaw, as seen in Fig. 2-3 For this case, a slight manual rotation of the PROFILER 2 is made and the measur ement is repeated, which minimizes the misalignment. Corrective rotations and re-measur ement of the alignment profile can be performed until the profile tilt is minimized. Step 2Determining detector offset corrections: Although the PROFILER 2s diode detectors are precisely attached to the ci rcuit board, small inherent errors in their fixed positions cause spikes and dips in the measured profiles when the measurements are taken in a region of high dose gradient. This is apparent in the alignment profiles of Fig. 2-3 A correction for these detector posit ions is necessary before accurate measurements of the leaf edge positions can be performed. Detector position corrections are determined as follows. Immediately following the PROFILER 2 setup, three y-axis profiles are measured with the X1 backup jaws positioned at -1, 0, and +1 mm relative to the CAX. To avoid mechanical hysteresis, the backup jaw is retracted in the same direction before each measurement. The MLC and back up jaw positions that are used to measure these three profiles are shown in Table 2-1 where reference line (RL) 30%, RL 50%, and RL 70% correspond to the meas urements taken with the X1 backup jaw positioned at -1 mm, 0 mm, and +1 mm of the CAX, respecti vely. The three profiles lie in the high dose gradient region at approxim ately 30%, 50% and 70% of the values 29

PAGE 30

measured at the center of the open field, OF, profile of Table 2-1 These three measurement positions were chosen because the penumbra within this region is linear. The data from these meas urements is transferred from the PROFILER 2 software to an Excel works heet for analysis. The data obtained with each detector is then normalized to the OF measurement using j jn jnOF MRL NRL, ,=, (2-1) where NRLn,j is the normalized reference line meas urement for backup jaw positions n (-1, 0, and +1 mm) and yaxis detectors j (1 j 83); MRLn,j is the reference line measurement for a si ngle detector j; OFj is the open field value for corresponding detector j. Linear regre ssion on the three data pairs for each detector j (NRL-1 j, NRL0 j, and NRL+1 j) produces slopes, mj, and intercepts, bj, for each detector. A linear interpolation is then used to obtain the 50% dos e position, DPj, for each detector using j j jm b DP = 5.0. (2-2) This value is a linear measure of the jaw po sition that results in a field edge at each detector. The detector offset correction for each detec tor relative to a reference detector, ref, is termed the relative detector offset, RDOj, and is calculated using ref jjDPDPRDO =. (2-3) The calculation of the RDO values sets a common origin for each detector that allows for a spatially unbiased m easurement of MLC positions Note that although two detectors are in the projection of the reference leaf, either one can be used as the reference detector. 30

PAGE 31

Step 3 Determining relative leaf positions: To measure the leaf positions for an MLC leaf bank, the procedure followed in t he previous step is repeated using the MLC leaf bank instead of the backup jaw. The colli mator configurations used for the fields MLC 30%, MLC 50% and MLC 70% are shown in Table 2-1 which correspond to measurements taken with the X1 leaf bank positioned at -1 mm, 0 mm and +1 mm of the CAX, respectively. The resulting profile s in the high dose gradi ent region are again approximately 30%, 50% and 70% of the value measured at the center OF profile. Since the array is aligned with the backup jaw in step 1, a tilt in t hese profiles indicates that the MLC bank may be in need of calibration. It is possible to calculate the degree of tilt in the reference line measurement and subsequently co rrect further measurements; however, this increases the complexity of the algorithm and is not used. The detector data is copied from the PROFILER 2 software to the Excel spreadsheet and is normalized to an open field measurement using j jn jnOF MMLC NMLC, ,=, (2-4) where NMLCn,j is the normalized MLC measurements for MLC leaf bank positions n and y-axis detector j; MMLCn,j is the leaf bank measurement for a single detector. As in the DPj calculations, the 50% dose-position for each leaf and detector combination, MLCPj, is linearly interpolated using j j jm b MLCP =5.0, (2-5) where the y-intercept, bj, and slope, mj, for each detector is determined by using linear regression on the three data pairs for each detector j. This value is a linear measure of the leaf position that would result in a field edge at the corresponding detector. 31

PAGE 32

Next, the leaf offset corre ction for each leaf and detector combination relative to the reference leaf, ref, and reference detector is termed the relative leaf offset, RLOi,j, and is calculated using refref ji jiMLCP MLCP RLO, ,=. (2-6) The relative detector offsets are then removed from the relative leaf offsets to provide the minor leaf offsets, MILO, using j ji jiRDO RLO MILO =, ,. (2-7) The MILOi,j values are averaged for the two detectors lying in the projection of each leaf to provide MILOi. The MILOi values are then linearly scaled to the isocentric plane at a 100 cm distance from the source. EPID The PROFILER 2 MILO results were checked using an EPID based version of the RDRL method. With this approach, the EPI D was used to image the fields RL 50% and MLC 50%, shown in Table 2-1 The data density of the EPID made it unnecessary to image the 30 and 70% RL and MLC fields. Since the active field size and fixed SDD of the EPID limited its field of view to only 25 leaves, it was necessary to shift the EPI D panel to fully measure a leaf bank. The EPID was first positioned so that the projec ted edge of the backup jaw for the RL 50% field falls at the center of the EPID in the in-plane direction. It was then shifted in the inplane direction to a position that allows the RL 50% and MLC 50% fields to be imaged for leaves 1-25. After these images were tak en, the EPID was shifted in the in-plane to a position that allowed for the RL 50% and MLC 50% fields to be imaged for leaves 1640. 32

PAGE 33

The RL 50% and MLC 50% fields were export ed from the EPID, in tiff format, into the MATLAB environment (The MathWorks, Natick, Massachusetts) for analysis. An in-house edge detection algorithm was then us ed to locate the field edges for the reference line, DPv, and each leaf, MLCPi,v, where i is the leaf number and v is the pixel number. Once these are known, the pixel distance of each leaf ends position to the reference line was calculated and c onverted to a spatial distance, SDi,v, using () = pixel mm DP MLCP SDvvi vi25.0, ,, (2-8) where 0.25 mm/pixel was the inherent pixel gain. The refe rence leafs spatial distance was then subtracted from the spatial distance for each of the other leaves to determine the minor leaf offsets using vref viviSDSD MILO, ,= (2-9) The MILO values from the tw o EPID positions were accepted based on the difference in leaf positions for leaves, 16 25, and the published leaf reproducibility. These leaves were chosen since they fall within the RL 50% and MLC 50% images taken for both of the EPID positions described above. The EPID was not used to measure major leaf offsets; its purpose was to check the results obtained with the PROFILER 2. Measuring major leaf offsets The method described below determines the position of the reference leaf pair, and therefore the leaf bank, rela tive to previously measured baseline positions at three MLC locationsretracted, C AX, and extended. The setup geom etry for the array is the same as was used during the minor leaf offs et measurements; the field configurations are similar. Note the projecti on of the center of the refer ence leaf for the measured leaf bank, X1 or X2, lies along the x-axis of the arra y. Determination of the major leaf offsets 33

PAGE 34

with the array involves three steps in additi on to those required for determining the minor leaf offsets, they are: off axis MLC measurement, array offset correction, and baseline comparison. Step 1Off axis MLC measurement: The proc edure for measuring the MLC at offaxis positions is similar to Step 3 of measuring the minor leaf offsets. For the minor leaf offsets, the MLC is measured at the CAX; therefore no additio nal measurements are required at this position. Fo r the major leaf offsets, re tracted and extended reference leaf positions are measured at 7.5 cm on the x-axis. These positions were chosen because they correspond to the largest IMRT fi elds our clinic regularly uses (~20 cm). Since the SDD of the array is 80 cm, these positions corre sponded to the 6 cm x-axis detectors on the array. The retracted and extended reference leaf positions are measured with three MLC geometries, three fields per geometry. The retracted reference leaf positions are measured with symmetric MLC fields. The co llimator configurations used for the symmetric fields SYM 30%, SYM 50% and SYM 70% are shown in Table 2-2 these correspond to measurements taken with the reference leaf pair incrementally being retracted over the -6 cm and +6 cm detecto rs. The resulting measurements in the high dose gradient region are approx imately 30%, 50%, and 70% of the value measured with the OF field at the same detector. The extended leaf positions are measured, in turn, by extending the X1 or X2 leaf bank over the center of the arra y to an off-axis position of -7 .5 or +7.5 cm respectively. The collimator configurations used to meas ure the extended X1 MLC fields X1 30%, X1 50%, and X1 70% are show in Table 2-2 these correspond to measurements taken with 34

PAGE 35

the X1 leaf bank respectively positioned at -7.6 cm, -7.5 cm, and -7.4 cm. Once again, the resulting measurements in the high dos e gradient region are approximately 30%, 50%, and 70% of the value measured with the OF field at the same det ector. Collimator configurations for the extend ed X2 MLC fields follow the same pattern but for the +7.5 cm position. The measured data are then copied from the PROFILER 2 software and pasted into the Excel worksheet. The worksheet cont ains the same linear-interpolation method, as previously described in Step 3 of measur ing the minor leaf offs ets, for determining the MLC position that results in a 50% open field value for the specific detector. These positions are labeled MLCk,m where k represents the -6 cm, 0 cm, or +6 cm x-axis detector position and m is the X1 or X2 reference leaf. Depending on the setup precision of the array relative to the CAX, all of the normalized values from the three measurements may have been either above or below the 50% open field value, if this was the case then shifted collimator configurations were necessary. Step 2Array offset correction: The offset of the array relative to the radiation CAX is determined using 2 ) (2,42 1,42 X XDP DP AO + =, (2-10) where DP42,X1 and DP42,X2 are the respective locations of the backup jaws X1 and X2 as determined by the detector array, which has an offset AO, the calculated midpoint between jaws. Each time the array was set up to measur e the MLC, its position relative to the CAX is slightly different fr om the previous measurement setup. This uncertainty does not affect the MILO values since they are relative values. However, the setup 35

PAGE 36

uncertainty does affect the major leaf offs ets and is removed from the reference leaf positions using AO MLC MALOmk mk=, ,, (2-11) where MALOk,m are the major leaf offset values for the X1 and X2 reference leaf. Step 3Baseline comparison: Before the baseline measurem ents are made, proper calibration of the MLC should be verified. Therefore, the ideal time to establish baseline values is during t he LINAC acceptance-testing / commissioning phase and / or following the annual LINAC QA. The initial se t of major leaf offset measurements establishes baseline values; subsequent measurements provide a comparison to the baseline values using mk mk mkbaseline MALO BO, = (2-12) where BOk is the baseline offset for x-axis detec tor position k and reference leaf m and baselinek,m is the initial set of MALO values. Ma jor leaf offsets that differ from the baseline by more than a preset toler ance are then in need of calibration. MLC calibration Elekta MLCs have a leaf-offset gain of 14 units/mm (0.071 mm/unit). Multiplying this gain by the minor leaf o ffsets and/or baseline offsets gi ves the adjustment values to bring each quantity into tolerance. These val ues are then entered into the Elekta LINAC software to complete the calib ration for the minor and major l eaf offsets. The calibration process is repeated if the leaf positions are not within our target tolerance of +/0.3 mm (+/4 units), the published Elekta MLC reproducibility.37 36

PAGE 37

Other MLC types Multi leaf collimat ors with leaves of mixed width require a variation of the described method. For example, an MLC that has leaf widths of 5 and 10 mm requires the same setup as described for the Elekta MLC used in this work. However, instead of two detectors being in the projection of each leaf, there are either two (for the 10 mm leaves) or one (for the 5 mm leaves). The backup jaw and MLC movements need to be altered due to sharper penumbras produced by the jaw and MLC. Multi leaf collimators with leaf widths other than 5 or 10 mm r equire a different SDD for the PROFILER 2. For example, an MLC with a 4 mm leaf width at isocenter would require a 100 cm SDD and a 2 mm shift. This would locate one detector in the projection of each leaf. Results Detector Offsets The linear dose gradient measured across the field edge (defined as 50% of the open field value) for each backup jaw was 15.2% per mm. The RDO values measured using the X1 backup jaw are shown in Fig. 2-4 A as a solid black line. This quantity was measured 10 consecutive times to determi ne the short-term reproducibility. The maximum difference between the measur ed values was 0.10 mm with an average standard deviation of 0.01 mm. This level of reproducib ility indicates that the PROFILER 2 is stable for measurements over a short period of time. There were three outliers in the RDO results in Fig. 2-4 A, at -0.4 cm, 4.4 cm, and 10 cm on the y-axis, whose positions were approximately 1 mm o ff the nominal axis. We speculated that the cause was a detector placement error in the off axis direction. While placement errors of this magnitude ar e not a factor in the intended use of the 37

PAGE 38

product, i.e. profile measurement s, they do play a factor in measuring leaf positions. Figure 2-4 B shows measurements of the X1 ba ckup jaw, the X2 backup jaw, and the OF field. The mirrored behavior of the spik es at these three pos itions verified the placement error speculation. The detectors at these three locations were replaced by the manufacturer, which brought the RDO values closer to the population average. The RDO values determined for the new detectors are shown in Fig. 2-4 A as the dashed line. In practice, the detector placement error is compensated by the correction techniques described herein; ongoing replacement of outliers is not necessary. MLC Measurement Comparison and Reproducibility The average dose gradient across the MLC leaf ends, defined as 50% of the open field value, was 13.5% per mm. Figure 2-5 displays the MILO values measured with both the array and EPID. The MILO results for the devices matched each other well, with a mean difference of 0.11 mm 0.09 mm. We conducted both short and long-term r eproducibility measurements using the RDRL method. For the short-term measurem ents, ten consecutiv e MILO and MALO measurements were made with the array over a two hour period. The maximum observed difference between individual va lues was 0.22 mm with a mean standard deviation of 0.07 mm. The long term reproducibi lity was studied by obtaining five sets of MILO and MALO measurements over a period of 12 weeks. The maximum difference found was 0.51 mm with a mean standard deviati on of 0.09 mm; this is in reasonable comparison to Elektas stated reproducibility of ~ 0.3 mm. Both the short and long term reproducibility was comparable to previously published values.37 Reproducibility was not evaluated using the EPID. However, the MILO results for leaves 16-25 of the EPID images (the leaves common to both sets of images for one leaf bank) provide some 38

PAGE 39

quantification of its short te rm reproducibility. The largest difference observed between MILO values for these leaves was 0.12 mm. MLC Service Issues Mechanical alterations may affect the optically-based MLC control system, which in turn could invalidate the MLC calib ration. Two examples are illustrated. Example 1 Replacement of the primary Mylar mirro r: EPID and array MILO values were determined using the RDRL techni que prior to replacem ent of the primary Mylar mirror. Measurements we re repeated using the same method after replacement of the mirror. The post-replacement meas urements indicated t hat the values had changed by as much as 1 mm in the upper hal f of the leaf bank as shown in Fig. 2-6 Example 2 CCD camera replacemen t: During LINAC maintenance, the CCD camera that is used to control the MLC leaf positioning was replaced. After the procedure, radiographic film was used to check the MLC leaf positions. No problems were found with a simple visual check of the film. Afterward, however, unacceptable passing rates were obtained during routine pati ent specific IMRT QA measurements. An MLC calibration was therefore performed usi ng the RDRL method with the array. Before calibration the initial leaf sp read of the X1 leaf bank was nearly 2 mm as shown in Fig. 2-7 The leaves were then calibrated, usi ng the MLC Calibration method described above, to within the manufacturers specified tolerance. A second calibration iteration was then performed to furt her tighten the spread. Discussion MLC Calibration Stability and QA Our long-term reproducibility re sults indicated that the El ekta MLC was stable over a period of 3 months. For an IM RT program that includes pat ient specific QA, monthly 39

PAGE 40

MLC checks should be sufficient to ensure the quality of patient treatment. However, if patient-specific IMRT QA is not perform ed, we feel that MLC QA should be done weekly. Our experience after the replacement of both the primary Mylar mirror and the CCD camera indicates that Elekta leaf posit ions must be checked after any type of maintenance that could affect the optical control system of the MLC. Device Comparisons The PROFILER 2 and EPID each have unique advantages and disadvantages when used with the RDRL method. The primary advantage of the PROFILER 2 approach is time efficiency. Measuring the minor leaf offsets for bot h leaf banks takes an average of 30 minutes with an additional 10 minutes r equired for the major leaf o ffsets. Entering the leaf adjustments into the Elekta software and re-measuring the calibrated positions takes an additional 30 minutes. Howeve r, the time requirement could be dramatically decreased if the entire procedure were incorporated into the PROFILER 2 software using a series of automatically deliv ered step-and-shoot fields. This has been requested of the manufacturer. The PROFILER 2 is a very stable meas urement platform, as indicated by the detector offset reproducibility. Therefore, deviations in t he measured leaf positions are the result of leaf positions rather than dev ice instability. Other benefits of the array include the ease of data analysis with compatible-user-friendly formats such as Excel. The array can also be used to measure an entire MLC bank wit hout repositioning whereas the EPID requires a shift in position to perform the same fu nction. It is also 40

PAGE 41

flexible in the sense that it can be used to calibrate any MLC on the market by simply setting it up at the correct SDD an d correct in-plane axis position. A disadvantage of the PROFILER 2 is the need to measure fields at 1 mm increments for the determination of the relative detector offsets and MLC leaf positions, which increases the required measurement ti me. A second disadvantage is the inability to measure an entire leaf bank at off axis positions; this could be overcome by moving the array off axis. The manufacturer is incor porating a motorized table that is capable of shifting the array up to () 20 cm off axis. This will allow leaf bank measurements over the entire field. The EPID based approach is also a viable m eans of calibrating an MLC with the RDRL method. Two major advantages of the EPID are its integrat ion with the LINAC and its larger cross-plane field of view. The la rger cross-plane field of view means that only one field is required for the refe rence line and the MLC measurements. Disadvantages exist for the EPID. A scarc ity of commercial software necessitates user-based code and associated software for data analysis, which makes it inefficient with regard to time and resources. For rigorous EPID measurements, many time consuming corrections are required to account for the rotation, tilt, and sag of the amorphous silicon panel.33 Considering the advantages and disadvantages, the effici ency with which routine MLC QA and calibration can be performed is comparable for the two devices. The major advantage of the PROFILER 2 lies in its co mpatibility with Excel for data analysis, whereas the EPID requires user-based code that is generally beyond Excel import capabilities. 41

PAGE 42

Conclusion The purpose of this study was to devel op an efficient and quantitative method for calibrating an Elekta MLC. The RDRL method m easures leaf-end positions relative to a radiation reference line defi ned by the MLC backup jaw. The PROFILER 2 detector array was used to implement the method and the results were verified with an EPID. The results obtained with the two devices agr ee well with each other and reproducibility values agree with previously published values The Elekta leaf positioning accuracy was found to be vulnerable to alterations of the MLC optical control env ironment. Therefore, MLC QA should be performed after any component of the MLC optical control system is disturbed. The RDRL calibration procedure wit h the PROFILER 2 is efficient for a number of reasons. Primary among these r easons is the ease of data handling using widely available software such as Excel. Our method can be easily utilized by any facility with access to a PROFILER 2. 42

PAGE 43

Table 2-1. List of collimator settings fo r PROFILER 2 measurements of the X1 MLC leaf bank at CAX for minor leaf offset s. Settings are based on the IEC 1217 convention. Collimator Field name MLC X1 (mm) MLC X2 (mm) Backup Jaw X1 (mm) Backup Jaw X2 (mm) Jaw Y1 (mm) Jaw Y2 (mm) OF 100 -100 100 -100 200 200 Alignment 150 -150 0 -100 200 200 RL 30% 150 -150 -1 -100 200 200 RL 50% 150 -150 0 -100 200 200 RL 70% 150 -150 1 -100 200 200 MLC 30% -1 -150 100 -100 200 200 MLC 50% 0 -150 100 -100 200 200 MLC 70% 1 -150 100 -100 200 200 Table 2-2. List of additiona l collimator settings for PROF ILER 2 measurements of the major leaf offsets. The fields listed ar e for measurement of the X1 reference leaf at retracted, SYM fields, and ext ended, X1 fields; the CAX fields were measured during the minor leaf offset s measurements. Settings are based on the IEC 1217 convention. Collimator Field name MLC X1 (mm) MLC X2 (mm) Backup Jaw X1 (mm) Backup Jaw X2 (mm) Jaw Y1 (mm) Jaw Y2 (mm) SYM 30% 74 -74 100 -100 200 200 SYM 50% 75 -75 100 -100 200 200 SYM 70% 76 -76 100 -100 200 200 X1 30% -76 -150 100 -100 200 200 X1 50% -75 -150 100 -100 200 200 X1 70% -74 -150 100 -100 200 200 43

PAGE 44

Figure 2-1. Minor leaf offsets are defined as the spatial offset of each leaf in a leaf bank relative to the reference leaf, leaf 20. Major leaf offsets are defined as the distance of the reference leaf to the radiation cent er of the beam. A B Figure 2-2. (A) The radiat ion defined reference line (RDRL) method requires precisely known detector positions; small positional errors in the arrays detectors can disrupt this method by introducing false offsets into the leaf positions. The detector offsets are correct ed for, which effectivel y creates uniform detector positions. (B) Two detectors are locat ed in the projection of each leaf and measure the position of the leaf end. 44

PAGE 45

Figure 2-3. The measurements required to r adiographically align the array. The thick solid line shows an open field meas urement used for comparing the alignment profiles; the dashed line shows the initial setup of the array, the thin solid line shows a correct radiographic alignment with the backup jaw after repositioning the array. A B Figure 2-4. (A) Detector offsets relative to a reference detector. The reference detector is chosen based on which detector lies in the projection of the reference leaf. Detectors at -0.4 cm, 4.4 cm, and 10 cm displayed a larger than average offset value as shown by the solid line. After detecto r replacement, the dashed line shows nominal offsets. (B ) Measurements with each backup jaw located at the CAX position. The mirro red behavior of the three large spikes on the profiles indicates they are due to positional errors. 45

PAGE 46

46 Figure 2-5. MLC leaf offsets relative to leaf number 20, the reference leaf. Offsets were measured using the RDRL method with two separate devices, the PROFILER 2 and an EPID. Figure 2-6. Change in relative MLC offsets after replacement of the primary Mylar mirror. Figure 2-7. Calibration of the X1 MLC l eaf bank. Measurements show relative MLC positions before calibration and after tw o iterations of the RDRL method with the PROFILER 2.

PAGE 47

CHAPTER 3 CHARACTERIZATION OF A MULTI-AXIS IONIZATION CHAMBER ARRAY Introduction Radiation therapy is continually increas ing in complexity as new treatment modalities are adopted in the clinic. While these advances benefit tumor dose localization, they also increase pressu re on departmental resources as the new modality is implemented. A prime example of this cause and effect is the adoption of intensity modulated radiation therapy (IMR T) and the complications that it added to treatment planning and qualit y assurance (QA) programs.1 The adoption of new modalities (e.g., CyberKnife, single-arc IMRT, and Renaissance) will likely continue this trend. New measurement technologies promise to ease these complications in two ways. They can decrease burdens that were intr oduced by the new treatment modality. They can also improve the efficiency of establis hed (but time consum ing) measurements, such as those taken with a scanning water tank. The scanning water tank has been an integral part of the radiatio n therapy clinic for decades, providing beam profiles, fractional depth doses (FDDs), and point measurements. While advances in computer technology have improved their efficiency, they still require large amount s of preparation time (~ 1 hours for setup and break down) and actual scanning time [several day s to measure linear accelerator (LINAC) beam commissioning data].28 Infrequent use by the clinical physicist adds to these inefficiencies and decreases the lik elihood of obtai ning quality data. Just as detector arrays streamlined the IM RT QA process, they also promise to increase the efficiency of collecting LINAC beam data. The time required to measure 47

PAGE 48

beam profiles at multiple depths can be decr eased from days to several hours. The skill required to obtain quality data is also reduced This is due to the fact that detector panels are mechanically simpler than water t anks and are more familiar to clinical physicists. The purpose of this work was to extensiv ely evaluate a detector array that has the potential to simplify the acquisition of LI NAC beam data. The device chosen for this research was the IC PROFILER (Sun Nuclear Corporation, Melbourne, FL 32940), which is a multi-axis ion chamber array. The potential of the IC PROFILER to be used as an alternative to a scanning water tank necessitates a study of its behavior in a radiation environment. Materials and Methods Materials The detector array used in this study was the IC PROFILER [Sun Nuclear Corporation (SNC), Melbourne, FL USA] and will henceforth be referred to as the panel. The panel contains 251 parallel plate ion chambers, each with a volume of 0.05 cm3. The active area of the panel is 32 x 32 cm2 (Fig. 3-1 ). A detector array is located on each coordinate axis (x and y) and on the negat ive and positive diagonal axes (nd and pd, respectively). The detector spacings ar e 0.5 cm and 0.71 cm on the coordinate and diagonal axes, respectively. All axes shar e a common center detec tor, with the xand diagonal axes missing detectors immediately adjacent to the c enter detector. The inherent thicknesses of buil dup and backscatter are 0.9 g/cm2 and 2.3 g/cm2, respectively. Included with the panel is a personal computer user interface (UI) that provides control over data acquisition in the form of adjustable real time updates (125 800 ms) 48

PAGE 49

and amplifier gains for low repetition rate s. Data can be acquired in pulsed or continuous mode. Measurement s in pulsed mode are sync ed with radiation pulses, which are detected by a system of trigger diode detectors. Hence, this mode is only intended for use with a pulsed radiation sour ce. Measurements in continuous mode are clock-synced and acquire data at a set update in terval, making it ideal for a continuous radiation source. Continuous mode can measur e pulsed radiation but with the possibility of measurement updates occurring during a pulse. Characterization measurements were ca rried out on an Elekta Synergy (Elekta Limited, Crawley, SXW UK) LINAC that wa s operated at X-ray energies of 6 and 18 MV. Two additional radiation sources were used that provided differentiating characteristics. These were a 60Co teletherapy unit (Eldorado 6, At omic Energy of Canada Limited, Mississauga, ON CA) and a Varian TrilogyTM LINAC (Varian Medi cal Systems, Palo Alto, CA USA); their use will be acknowledged accordingly. Methods The UI provides graphical vi ews of radiation profiles from all four detector axes and profile analyses of the usual beam features (flatness, sym metry, beam center, etc.). This investigation of the panels relative response to various conditions was carried out in MATLAB (The MathWorks, Natick, MA USA), not the panels UI, which is designed primarily for absolute measurements of beam featur es. This is possible since measurements are saved in a raw ASCII format that allows post processing in the form of background subtraction and detec tor sensitivity adjustments. The raw data for each detector was processed to remove background counts, and a temperature-pressure correction and detector specific calibration factors were applied. The resulting value [corrected counts (CC)] is expressed with 49

PAGE 50

() TP ii iRawCountTimeTicbias CC PCF gain =i (3-1) To save space, all subscripts, vari ables, and equations are defined in Table 3-1 The data produced by Eq. 3-1 is identical to that from the panels UI. Reproducibility The panel was characterized in areas that have traditionally been of interest for radiation detectors and arrays. The first item tested was measurement r eproducibility. The short and long term reproducibility of the panel was determined on the 60Co teletherapy unit. This radiation source wa s chosen because of its reproducible nature. The panel was irradiated with a field size la rger than its active area. No additional buildup was used. The short term reproducibility was evaluated with 10 consecutive 1 minute deliveries; long term reproducibility included 1 minute deliveries over the course of nine months. Only the center 45 s of saved data fr om each delivery was used in an effort to reduce the effects of source travel that occurs in 60Co teletherapy units.41 The measurements were normalized to the central axis (CAX) detector and the reproducibility was quant ified by determining each detector s a) coefficient of variation [(CV), relative standard deviation] and b) maximum deviation from the mean. Dose and instantaneous dose rate dependence The dose dependence of the panel was eval uated over LINAC deliveries ranging from 1 to 1000 monitor units (MU). The measurements were evaluated at a source to detector distance (SDD) of 100 cm; buildup was equivalent to the energy specific d-max (1.5 and 3 g/cm2 for 6 and 18 MV X-rays, respectively). Simultaneous measurements were made with an independent system (Farme r-type chamber m odel N30001, PTW, 50

PAGE 51

Freiberg, DE; electrometer model 1010, SNC) in a 4 x 30 x 30 cm3 slab of water equivalent material that was located below the panel. The dose definitions for this machine were such that one MU (100 cm from the source, 10 g/cm2 buildup, 10 x 10 cm2 field) delivered 0.78 and 0.88 cGy fo r 6 and 18 MV X-rays, respectively. The measured signals from both systems were normalized to the 200 MU deliveries. The panels response was then normalized to the Farmer-type chambers response using, () () () () () 200 200panel Farmer chamberMEASMU MEASMU ResponseMU MEASMU MEASMU = (3-2) The panels instantaneous dose rate ( ) dependence was evaluated at a constant LINAC pulse rate frequency (PRF), exposure, and buildup (400 Hz, 200 MU, and 10 g/cm2, respectively). The dose rate was altered by changing the SDD from 90 cm to 150 cm, in 10 cm incremen ts. This effectively changed the in accordance with the inverse square law. All measurements were repeated using a Farmer-type chamber under similar conditions. D D The methodology of Eq. 3-2 was used to evaluate the panels dependence, D () R esponseD The variable was used in place of MU and the detector systems were normalized to the 100 and 110 cm SDD dose rates for 6 and 18 MV X-rays, respectively. D 51

PAGE 52

PRF dependence The panels PRF dependence was evaluated at a constant field size, SDD, and buildup (35 x 35 cm2, 100 cm, and energy specific d-ma x, respectively). The LINACs control system was used to vary the PRF from 50 to 400 Hz. Measurements were made with two panel orientations that differed by a 180 rotation. Simultaneous measurements were made with an independent system (Far mer-type chamber model N30001; 1010 electrometer) located below the panel in a slab of water equivalent material. Once again, the same methodology (Eq. 3-2 ) was used to evaluate the PRF dependence [() R esponePRF] of the panels center detecto r. The variable used was PRF instead of MU and the detector systems were normalized to the 400 Hz PRF. The PRF dependence of the panels off-axis detectors was determined by relating the PRF response of each detector to that of the center detector. This is expressed as () () () () () 400Hz 400Hzi i ctrCCPRF CC OAResponsePRF CCPRF CC = (3-3) A detector PRF response that differs from the center detector will manifest itself as a deviation from unity. This analysis is useful since detector-specific variables such as calibration factors and air density corrections cancel, leaving hidden variables that may exist in the electronics or as extra cameral volume. An actual LINAC PRF variation might be interpreted incorrectly as panel PRF dependence. This erroneous interpretation can be avoided by taking the difference between the values for the 180 and 0 panel orientations. This is expressed as OAResponse 52

PAGE 53

() () ()(),_, _180_0pos posPRFDOAResponsePRFpo OAResponseOAResponse = (3-4) Reflective symmetry in this value woul d indicate PRF dependence in the mirror detector pair, i.e. the dete ctors that mirror each other about the array center. Energy dependence The panels dependence on X-ray energy was evaluated using two methods. First, a Farmer-type chamber and electrometer (m odel FC65-G, IBA, Schwarzenbruck, DE; model K602, CNMC Company, Nashville, TN USA) with National In stitute of Standards and Technology traceable calibrations were used to measure 200 MU deliveries at Xray energies of 6 and 18 MV. The delivered dose was determined in accordance with the American Association of Physicists in Medicine report by task group (TG) 51,42 the source-to-axis distance setup was used. E quivalent measurements were made with the panel and an appropriate amount of water equiva lent material. The panel data was not processed beyond the extent of Eq. 3-1 ; e.g. no Pion was applied. The methodology of Eq. 3-2 was also used to evaluate the panels energy dependence, () R esponseE.The variable was E instead of MU and the detector systems were normalized to the 6 MV X-ray deliveries. Spectral differences that o ccur for different field size s and with increasing buildup could result in a panel energy response. The second method compared FDD curves for 6 and 18 MV X-rays and various square field sizes (5, 10 and 20 cm). The reference FDDs were acquired with a Blue Phantom (IBA) scanning water tank equipped with CC13 (IBA) ion chambers for the reference and field detectors. The panels FDD was 53

PAGE 54

acquired with water equivalent material r anging from 0.9 to 30 cm. A 90 cm SSD was used for both systems. The FDDs were normalized to the 10 cm buildup measurement. The energy response of the panel was then quantified by taking the ratio of the panels FDD measurement to that of the scanning water tank. This relationship is expressed as () buildup CC ctrFDD FDD buildup Response =13. (3-5) It was assumed that the CC13 had a flat energy response. Response to power being applied to the electronics The time required for the temperatur e of the panels el ectronics to reach equilibrium upon power up wa s determined using the 60Co teletherapy unit. The panel was allowed to equilibrate to room temperat ure overnight. In the morning, power was applied to the panel and a background meas urement was immediately taken. The source was then extended into position and left for three hours, during which time periodic measurements were made. Since the source was not moved bet ween measurements, any change in measured signal was attributable to drift after power was applied. This change was quantified by normalizing the mean detector val ue for each measurement to that of the measurement at t = 180 minutes This is expressed as () () () min180CC tCC tResponse =. (3-6) The 180 minute measurement was used as a normalization point since the panels electronics were assumed to be in equilibrium by this time. 54

PAGE 55

Calibration constancy The relative sensitivities of the panel s detectors were determined using a proprietary calibration method that has been previously described and evaluated.25, 43-45 While the calibration method for the panel is the same, the device is different, which could affect the calibration quality. Therefor e, calibration factors were evaluated for reproducibility, accuracy, energy dependence, and bui ldup dependency. Letourneau et al. described this calibration methods sensitivity to small changes in the LINAC dose distribution. They reported that when the symmetry stability was degraded from 0.1% to 0.3%, t he calibration reproducibility degraded from () 0.8% to () 1.2%. The symmetry stability for the Varian Trilogy was () 0.05% which was slightly better than the Elekta Synergy at () 0.15%. Therefore, we used the Varian unit to calibrate the panel. The short term reproducibility was ev aluated by calibrating the panel five consecutive times over 2 hours using both 6 and 18 MV X-rays. The long term reproducibility was evaluated over the course of six months. The calibration reproducibility was quant ified by determining each detectors a) CV and b) maximum deviation from its mean. The accuracy of the calibration was evaluated by taking two full-field measurements with a 180 panel rotati on between measurem ents. The 180measurement data was rotated about the array center to maintain the LINAC coordinate system of the 0 measurement. The calibra tion accuracy was quantified by taking the ratio of the measurements. This is expressed as () () () =0 180 _pos pos posNMEAS NMEAS po Accuracy CF. (3-7) 55

PAGE 56

An accurate calibration would result in minimal differences between the two measurements since a measurement at a point in space should be detector invariant if the detectors response is known by calibration. The calibration factors energy dependencies were evaluated by calculating the ratio of the 18 MV calibration factors relative to the 6 MV calibration factors. This is expressed as () () () MV6 MV18 _i i iCF CF E Response CF =. (3-8) The calibration factors buildup dependencies were evaluated by taking the ratio of calibration factors determined with different amounts of buildup (0 .9, 2.9, and 4.9 cm) relative to the calibration done with 4.9 cm of buildup. This is expressed as () () () cm9.4 _i i iCF buildupCF buildup Response CF =. (3-9) Backscatter dependence The panels dependence on backscatter was evaluat ed for a variety of square field sizes, buildup amounts, and backscatter am ounts [5, 10, 20, and 30 cm; an energy specific d-max and 10 g/cm2; 2.3 (the inherent backscatter) to 19.1 g/cm2]. The effect of increasing backscatter on profile shape wa s quantified by relating the backscatter response of each detector to that of the center detector. This is expressed as () () () () () _i i ctrCCbs CCinherent OAResponsebs CCbs CCinherent = (3-10) 56

PAGE 57

Beam profile measurements and output factors Beam profiles and FDDs were measured for field sizes ranging from 5 x 5 cm2 to 30 x 30 cm2. The SSD of the panel pl us buildup was kept cons tant at 90 cm. These measurements were compared to scanning water tank data (CC13) that were collected under a similar geometry. The profile accuracy was quantified by taki ng the ratio of the panels measurement to that of the CC13 for two buildup val ues (the energy specific d-max and 10 g/cm2). This is expressed as () 13 ,Panel pos CC posfsNMEAS PAfs NMEAS= (3-11) The agreement between the FDDs produced by the two systems was previously quantified [see Eq. ( 3-5 )]. Output factors were measured for symmetric fields ranging in size from 1 x 1 cm2 to 30 x 30 cm2. The measurements were done with t he panel at an SDD of 100 cm with 10 cm of buildup. These were checked using a one dimensional scanning water tank and three different detectors: a Farmer-ty pe chamber (model N30 001, PTW), a CC13 scanning ion chamber, and the Edge diode detector (SNC). These detectors are applicable over a wide range of field sizes so as to provide overlapping data and to accurate ly extend the results to both small and large field sizes. Specifically, the Edge detector was used only for field sizes of up to 10 x 10 cm2 due to the energy dependence of diode detectors.46 Due to the Farmer-type chambers size, its use was limited to fi eld sizes from 5 x 5 cm2 to 30 x 30 cm2. The CC13 was used over the entire range of field sizes. 57

PAGE 58

The accuracy of the panels output factor s was quantified through normalization to the independent systems output factors. This is expressed as () panel j f sOutputFactor NOFfs OutputFactor= (3-12) Results and Discussion Reproducibility The panels short and long term repr oducibility are presented in Table 3-2 For 10 consecutive measurements in which the 60Co source was left in the extended position, the maximum deviation from the mean was 0.15%. For long term measurements over the space of nine months, the maximu m deviation from the mean was 0.84%. Dose and Instantaneous Dose Rate Dependence In Fig. 3-2 we see the dose response of t he panel normalized to that of the Farmer-type chamber. A non-linear relationshi p was observed for the panel in pulsed mode. During the first MU of delivery, the panel under responded by nearly 50% relative to the Farmer-type chamber. As the deliver ed dose was increased, the response of the panel converged to that of the Farmer-type chamber. In contrast, a linear relationship wa s observed when the panel was operated in continuous mode. The responses of the panel and Farmer-type chamber were within () 0.6% of each other for lower MU deliverie s (< 20 MU). For larger deliveries, the agreement was significantly tighter. The likely cause of the non-linear res ponse observed in pulsed mode is the panels measurement trigger syst em. This system of trigger detectors initiates a measurement only when a detected signal crosses a preset thre shold. If the initial beam 58

PAGE 59

intensity is low enough, then the measurement will not be tri ggered. Only pulses that caused a response below the trigger threshold would be missed. This theory was tested by creating a hypot hetical panel response in which only a fraction (f ) of the first MU was missed. After n MU, the ratio of pulsed to Farmer-type chamber measurements, 200200 nfn r f = was plotted with a value f = 0.45, which wa s close to the under response observed with the panel. As shown in Fig. 3-2 the hypothetical panel res ponse closely matches the actual panel response in pulsed mode. This information was provided to the manufacturer and a solution to the tr igger logic has been implemented. An important question to ask is: why use pulsed mode? Pulse syncing has advantages in machine diagnostics where only integral LINAC pulses have been accumulated for each data frame. For exam ple, a PRF of 60 Hz and a measurement interval of 150 ms will result in 9 LINAC pulses being accumulated in each frame. This allows studies of pulse st ability, dose per pulse, etc. The LINACs nominal (100 cm SDD, 10 cm of buildup, 10 x 10cm2 field) were 6.5 and 8.3 cGy/s for 6 and 18 MV X-ray beam s, respectively. The dose rate was altered by changing the SDD from 90 to 150 cm in 10 cm increments. This correlated to instantaneous dose rate ranges of 8.0 to 3.0 and 10.3 to 3.7 cGy/s for 6 and 18 MV Xray beams, respectively. The ranges were determined by applying an inverse square correction to the nominal DD 59

PAGE 60

In Fig. 3-3 we see the instantaneous dose rate response of the panel normalized to the response of the Farmer-type chamber. The panel over responded, relative to the Farmer-type chamber, as the dose rate decr eased. This effect reached a maximum of 0.9% at the lowest 6 MV dose rate. The cause of the panels over response is not entirely clear. A possible source of the over response is a higher rate of volu me recombination in the panels ion chambers at the higher irradiation rates.47, 48 In ref 48, Boag expresses t he collection efficiency f in pulsed radiation beams as () 1fe=, where is proportional to p, t he charge collected per pulse. In this equation, we can vary p to see the effects on f. Differentiati on of f and reordering wi th p results in the equation ()1dfdp f e fp =. In regions of small changes, we can substitute df ~ f2 f1 and dp ~ p2 p1 to get: ()22 11111 fp f e fp + From our measurement of f by varying the panels chamber bias voltage at a given p,42 we found f to be approximately 0.995, which estimates at 0.01. Therefore, a change in the dose per pulse by a factor of 2.8 (by va rying the SSD from 150 to 90 cm) results in an expected change in efficiency of 1 + (2 .8 1)*(-0.005) = 0.991. From Fig. 3-3 and relative to the Farmer-type chamber, we see f2/f1 = 0.998/1.01 = 0.988. This is in reasonable agreement with the theory estimate of 0.991, considering the measurement uncertainty. 60

PAGE 61

Other possible contributors to the response in this particular setup could be slight changes in the scatter energy as well as diffe rent re-combination in small extra-cameral volume regions. Regardless, the observ ed response is small and can probably be ignored for the irradiation rates investigated. Measurements outside of these dose rates (an un-flattened LINAC beam) would r equire further investigations. PRF Dependence In Fig. 3-4 we see the PRF response of t he panel relative to the Farmer-type chamber. The panels response increased with a decrease in PRF. This was true for continuous and pulsed measurement modes. Fo r the 18 MV X-ray beam, the 50 Hz PRF was unavailable due to a LINAC symmetr y interlock that would terminate the beam. The increased response in continuous m ode was the result of fewer radiation pulses occurring while the device was performi ng a measurement update. As a result, the number of missed pulses decreased and the si gnal increased. It is important to note that this is not a fault with continuous mode since this mode is not intended for use with a pulsed radiation source. The panels over-response in continuous mode can be lowered by using a collection interval longer than 125 ms (the de fault). This effectively reduces the number of measurement updates, and hence the num ber of pulses that occur, during measurement updates. We chose a collection interval of 500 ms to demonstrate this effect. In Fig. 3-4 we see that the PRF responses obtained with the longer collection intervals closely matched those obt ained with the Farmer-type chamber. The signal increase observed in pulsed mo de was related to the measurement trigger error previously discussed. As the PRF was decreased, the total delivery time 61

PAGE 62

increased which gave the measurement-triggersystem more time to detect a pulse and initiate a measurement. The result was fewer missed pulses and an increase in signal. The PRF response of the x-axis detectors relative to the center detectors response is shown in Fig. 3-5 A. Three distinct features are apparent. First, there was a singular spike at the (-) 4 cm detector. This detector had an over response that increased to 4% as the PRF was decreased to 50 Hz. The second feature was a global PRF response that resembled a sine functi on. The magnitude of this response had a maximum value of () 1%. The final featur e was a global response in which curve separation increased at lower PRF. It is possible for the LINAC to exhibit a PRF variation. For the analysis method that was used in Fig. 3-5 A it was impossible to disti nguish between a LINAC and a panel PRF response. Rotating the panel by 180 and repeating the experiment helped isolate the source. In Fig. 3-5 B, we see the results for the 180 measurement. The data was rotated to maintain the LINAC coordi nate system. In comparing Figs. 3-5 A and B, it is apparent that the singular spike stay ed with the (-) 4 cm detector [n ow in the (+) 4 cm position due to the data rotation]. This indicates the detector response was dependent on PRF. We also see that the sine shaped respons e stayed true to the LINAC coordinate system, which indicates it is due to the LINAC. The detector PRF responses were visually amplified by taking the difference between the OA_Response(PRF) values fo r the two panel orientations, Fig. 3-6 A. This analysis method eliminated any response due to the LINAC (the sine feature) while preserving the other two PRF fe atures. The singular detector response is present at the 62

PAGE 63

() 4 cm positions because each is a func tion of both mirror detectors. Singular responses (spikes) were also observed for t he (+) 0.5 cm y-axis and (-) 1.4 cm nd-axis detectors. These singular detector respons es have been reported to the manufacturer and are under further investigation. The glob al PRF response was relatively small, typically less than () 1% for the tested PRFs. In Fig. 3-6 B we see the 6 MV DOA_Response(PRF,po) values for the x-axis detectors when the panel was operated in continuous mode. The global PRF response is similar to what was observed in pulsed m ode. Interestingly, the PRF response at the () 4 cm detector pair is not present. The cause for this discrepancy between measurement modes has been reported to the manufacturer and is under further investigation. The responses for the remaini ng axes and energies were similar to those values presented in Figs. 3-6 A and B. Energy Dependence Increasing the X-ray energy from 6 to 18 MV decreased the panels response by 0.47 0.02% relative to that of the Farmer-type chamber. This demonstrates that the panel has a low level of energy dependence rela tive to a Farmer-type chamber that is used for clinical reference dosimetry. The energy response was also quantifi ed by taking the ratio of panel FDD measurements relative to those of a CC13 ion chamber. The results for the 6 and 18 MV photon beams are shown in Fig. 3-7 The dashed-vertical-bold-lines represent the energy specific d-max values. The maximum deviation relative to the CC13 occurred for a 20 x 20 cm2 6 MV X-ray beam. For this beam and field size, the panel under responded by approximately 2% at d-max. Typically, the two systems were within () 1% of each other. While the 63

PAGE 64

agreement in the buildup region was poor, this area is unpredictable due to the lack of electronic equilibrium and the difference in volume averaging that occurs between the panel and CC13. Larger field sizes (30 x 30 cm2) were not included because of increased measurement noise due to electronic cross ta lk. This was a result of increased scatter entering the panels electronics due to the larger field size (at increased SDDs) and increased secondary scatter from the buildup. One should be aware of this possibility when measuring larger field sizes in conjunc tion with a substantial amount of buildup. Response to Power Applied to the Electronics While the panel never completely reached equi librium, it did qui ckly reach a state of minimal change. Two minutes after initia l power up, the respons e was within 1% of the 180 minute measurement; after ten minutes, the response was within 0.3%. This effect can be minimized by powering up t he panel at the beginnin g of a measurement setup or by leaving it powered up at all times. Calibration Constancy The short and long term reproducibility of the detectors calibration factors are shown in Table 3-3 The maximum deviations from the mean were 0.58 and 0.68% for 6 and 18 MV X-rays, respectively. The accuracy of the calibration fact ors was evaluated by comparing measurements from two panel or ientations (0 and 180). Correcting the data order for the rotation in the 180 data and assuming a reproducible LINAC beam, perfect calibration factors should provide a rati o of unity for all positions. In Fig. 3-8 we see that all of the calibration factors were within ( ) 0.8% of unity. The symmetric error that occurred about the yand pd-axes indicates a calibration error of less than 0.8%. 64

PAGE 65

The energy response of the panels cali bration factors is shown in Fig. 3-9 The 18 MV calibration factors were reproducibly larger (up to 1.5%) than the 6 MV calibration factors. To obtain the most accurate prof ile measurements, separate calibrations at each energy are recommended. The response of the panels calibration fa ctors to additional buildup is presented in Fig. 3-10 While the range of buildup used was rela tively small (0.9 to 4.9 cm), it represented a significant range in beam quality and electronic equilibrium. The majority of the detectors (>97%) were within the reported calibration reproducibility. This indicates that a single calibration file can be used for a range of buildup values. However, calibrating with at least a build up amount that is equivalent to d-max is advisable to avoid including a response fr om collimator electron contamination. Backscatter Dependence The effect that increasing backscatter has on profile shape is presented in Fig. 311. Only the 20 x 20 cm2 field is shown. A limited number of backscatter amounts were chosen that reflect the behavior of the panel. In Fig. 3-11 A we see the 6 MV backscatter res ponse for the negative x-axis; the positive x-axis behaved similarly. The re sponse within the beam appears relatively uniform. Outside the beam, the increase in m easured signal is more pronounced, rising by up to 50%. This represents a small quantitative increase due to the lower initial value of the signal. An expanded view of the panel wit hin the beam is presented as an inset in Fig. 3-11 A. The addition of backscatter decreased the panels measured signal by a maximum of 0.8% over 80% of the field width. The 18 MV results are presented in Fig. 3-11 B. The out-of-beam increase is smaller than was observed with the 6 MV beam. In the inset of Fig. 3-11 B we see an 65

PAGE 66

expanded view of the in-beam portion of the negative x-axis. Again, the change in profile shape was reduced rela tive to the 6 MV beam. For all energy, backscatter, field size, buil dup, and array combinations, the profile shape changed by less than 1% within 80% of the field width Two exceptions occurred. The 6 MV 30 x 30 cm2 beams resulted in backscatter responses of up to 1.6%. The 18 MV 5 x 5 cm2 beam resulted in a backscatter response of up to 1.5%. Beam Profile Measurements and Output Factors In Fig. 3-12 we see a sample of cross-plane profiles measured by the panel and the CC13 for a 6 MV beam. The buildup for these measurements was 10 g/cm2; no additional backscatter was used for beam profile measurements because of the panels minimal backscatter response. Taking the ratio between the two detectors systems provided a clearer indication as to their agreement, Figs. 3-13 A D. The ratios covered 80% of the field width. The total spread in error between the two systems was typically on the order of 1.5% and was biased towards positive errors. The cause of the positive bias was a combination of error in the center detectors calibration factor and the measurements. Si nce there is only one normalization point, error in it can skew the entire meas urement. If instead the panel and CC13 measurements were normalized to the mean value over 80% of the field width, a much tighter agreement was observed between the panel and the water tank results. In Fig. 3-14 we see a visual representation of the 6 MV FDD for a 10 x 10 cm2 field. The agreement between the two detecto r systems was previous ly presented (see Fig. 3-7 ). Their agreement was generally within () 1%. The exception occurred for shallow regions of the 6 MV 20 x 20 cm2 field. For this field si ze and energy combination 66

PAGE 67

the panel under responded relative to the CC13 by approximately 2% at d-max. By 5 cm of buildup, the two systems are within () 1%. The data were collected for field sizes of up to 20 x 20 cm2. Radiation contributions to t he panels electronics prohibited FDD measurements for larger field sizes. Output factors for the 6 and 18 MV X-ray beams are presented in Figs. 3-15 A and B, respectively. There were some dev iations between the panel and the secondary detectors for field sizes below 5 x 5 cm2. The ratio of the panels output factors to those of the independent detectors qua ntified the agreement betwe en the two systems, Figs. 3-15 C and D. The agreement bet ween the panel and the secondary detectors was within () 1% for the specified field sizes, with the except ion of the 1 x 1 cm2 and the 2 x 2 cm2 field sizes, which exceeded 1%. Conclusion The purpose of this study was to ext ensively evaluate a detector array (IC PROFILER) that has the potential to simp lify the acquisition of LINAC beam data. We have shown that the IC PROFILER generally behaved within () 1% of an independent (or nominal) response in all area s tested. One primary exception occurred. The response of the IC PROFILER to dose was non-linear when the device was operated in pulsed measur ement mode. The problem occurred because the measurement trigger was not in itially activated and pulses we re missed. However, after the measurement was triggered al l radiation was acquired. A solution to the trigger logic has been implemented by the manufacturer. The ability of the IC PROFILER to a ccurately reproduce water tank profiles, FDDs, and output factors is an indication of its abilities as a water tank alternative. The primary benefit of using the IC PROFILER versus a scanning water tank is time 67

PAGE 68

reduction. The same measurements (including device setup and breakdown) for both systems took 180 minutes with the water tank versus 30 minutes with the IC PROFILER. The time savings increases as the measurement load is increased. This work evaluated the IC PROFILER and demonstr ated that it is capable of efficiently providing water tank equivalent scans. 68

PAGE 69

Table 3-1. List of subscripts, variables, and equations Subscript Description ctr the panels center detector i panel detector ( i: x -, y -, pd-, and nd-axes) j radiation detector ( j: panel, Farmer-type chamber, CC13, Edge diode detector) pos position on the LINAC coordinate system Variable Description bs total backscatter (cm) buildup total buildup (cm) CC mean corrected counts CF calibration factor that normalizes the relative sensitivities of the panels detectors CV coefficient of variation (relative standard deviation) D instantaneous dose rate (cGy/s) E X-ray energy FDD fractional depth dose fs side of a square field gain ion chamber amplifier gain bias detector leakage rate in RawCounts per TimeTic MEAS measured value of the detector system (nC or CC ) MU monitor units NMEAS profile measurement normalized to the central axis (CAX) PTP pressure and temperature correction po panel orientation (0 or 180) PRF pulse rate frequency (Hz) RawCount raw counts measured by the panel detectors t elapsed time since power was applied to the panel TimeTic elapsed time since starting the panel measurement Eq. Name Description 3-1 CC corrected counts of the panel 3-2 Response ( MU ) the panels dose response relative to a Farmer-type chambers dose response -Response () Dthe panels instantaneous dose rate response relative to a Farmertype chambers instantaneous dose rate response -Response ( PRF ) the panels PRF response relative to a Farmer-type chambers PRF response 3-3 OA_ Response ( PRF ) PRF response of off-axis detectors relative to the center detectors PRF response 3-4 DOA_ Response ( PRF,po ) Difference between OA_Response ( PRF ,180) and OA_Response ( PRF,0) -Response ( E ) response of the panel relative to TG-51 based Farmer-type chamber measurements for 6 and 18 MV X-rays 3-5 Response ( buildup ) FDD agreement between the panel and CC13 3-6 Response ( t ) response of the panel to power being applied to the electronics 3-7 CF_ Accuracy ( po ) accuracy of the detecto rs calibration factors 3-8 CF_ Response ( E ) energy response of the det ectors calibration factors 3-9 CF_ Response ( buildup ) buildup response of the detectors calibration factors 3-10 OA_ Response ( bs ) backscatter response of off-axis det ectors relative to the center detectors backscatter response 3-11 PA( fs ) profile agreement between the panel and CC13 3-12 NOF ( fs ) the panels output factors nor malized to independent detectors 69

PAGE 70

Table 3-2. The panels short and longterm reproducibility were evaluated on a 60Co teletherapy unit Short term Long term Mean CV (%) 0.05 0.24 Maximum CV (%) 0.11 0.50 Maximum deviation (%) from the mean 0.15 0.84 Table 3-3. The short and long term reproducibi lity of the relative detector calibration factors. 6 MV 18 MV Short term Long term Short term Long term Mean CV (%) 0.12 0.09 0.08 0.20 Maximum CV (%) 0.35 0.54 0.18 0.59 Maximum deviation from the mean (%) 0.58 0.57 0.20 0.68 70

PAGE 71

Figure 3-1. Overlay of the IC PROFILER ( panel) showing the multiple detector axes. Figure 3-2. The panels dose response rela tive to the Farmer-type chambers dose response. 71

PAGE 72

Figure 3-3. The panels instantaneous dose rate response relative to the Farmer-type chambers. Figure 3-4. The panels PRF response relative to the Farmer-type chambers PRF response. The measurement update interv al was the default (125 ms) for all measurements expect one, which was set to 500 ms. 72

PAGE 73

A B Figure 3-5. The off-axis PRF response for the x-axis detectors relative to the center detector for panel orientations (A) 0 and (B) 180. The panel was operated in pulsed mode for each panel orientation. The data for panel orientation 180 was rotated to maintain the 0 LINAC coordinate system. A B Figure 3-6. The difference between the 180 and 0 OA_Response ( PRF ) values for the panel operated in (A) pulsed mode and (B ) continuous mode. The cause of the spike at the () 4 cm positions in pulsed mode (or its absence in continuous mode) is not entirely understood and is under continued investigation by the manufacturer. Figure 3-7. The energy respons e of the panels center detec tor presented as a ratio of the panels FDD to the CC13s FDD. 73

PAGE 74

Figure 3-8. The accuracy of the calibration factors. Figure 3-9. The energy respons e of the calibra tion factors. Figure 3-10. The buildup response of the ca libration factors for 6 and 18 MV X-rays. 74

PAGE 75

A B Figure 3-11. Off axis backscatter response of the x -axis detectors for (A) a 6 MV 20 x 20 cm2 field and (B) an 18 MV 20 x 20 cm2 field. Buildup was equivalent to dmax. Figure 3-12. Normalized cross-plane meas urements with a CC13 and the panel. The beam energy was 6 MV and the buildup was 10 g/cm2. 75

PAGE 76

A B C D Figure 3-13. Profile agreement (over 80% of the field width) between the panel and CC13 for 6 MV X-ray (A) crossplane measurements and (B) in-plane measurements and 18 MV X-ray (C) cr oss-plane measurements and (D) inplane measurements. Figure 3-14. FDDs for a 6 MV 10 x 10 cm2 field. 76

PAGE 77

77 A B C D Figure 3-15. Output fact ors measured with the panels center chamber and three independent detectors for a (A) 6 MV and (B) 18 MV X-ray beam. The panel output factors were normalized to each independent detector for the (C) 6 MV and (D) 18 MV X-ray beams.

PAGE 78

CHAPTER 4 WIDE FIELD ARRAY CALIBRATION D EPENDENCE ON THE STABILITY OF MEASURED DOSE DI STRIBUTIONS Introduction Detector arrays have become an integral tool in radiation oncology by efficiently providing dosimetric information on which clinical decisions are often made. By definition, an array contains multiple detec tors, underlying the impor tance that each one has an equivalent sensitivity to radiati on. A violation of this requirement will consequently bias the measurement and possibly the clinical decision. Since some variations in the detectors sensitivities are inevitable, calibration techniques have been develo ped to compensate them.43, 49, 50 One such method, known as wide field (WF) calibration,43 has become widely used by the medical physics community.25, 45, 51, 52 The basis of the WF calibration protocol is determining the relative detector sensitivities through a response norma lization that is determined by detector substitutions at given field locations. Accompanying an array calibration is a respons ibility to verify that the calibration results do not significantly alter the appearance of the true beam shape. Verification of the array calibration can be accomplished with a comparison to scans in water, but the equipment setup is tedious and avoided in m any circumstances. Instead, many users rely on a visual examination of measured pr ofiles under the premoniti on that if it looks correct, it must be correct. In 2004, Letourneau et al. described the us e of a two-dimensional diode detector array for patient specific quality assurance (Q A) of intensity modul ated radiation therapy (IMRT).45 Wide field was used to calibrate the detector array. Letourneau reported that 78

PAGE 79

during calibration, linear acce lerator (LINAC) symmetry variations as small as 0.3% led to calibration errors of up to () 1.2%.45 During recent work to characterize an i onization chamber array (under review by Medical Physics),53 our group also experienced the WF calibration algorithms sensitivity to symmetry instabilities. Calibrati on errors of up to 1.6% [Fig. 4-1 A, see Eq. 4-11 for p_error] were observed for a LINAC that had symmetry variations on the order of () 0.15%. The calibration error reduced to less than () 0.5% [Fig. 4-1 B] for a LINAC with symmetry variations of () 0.05%. While symmetry variations of 0.15% are unlikely to be clinically relevant, the resultant calibration errors have the potentia l to bias measurements by a clinically significant amount. The purpose of this work was to reduce the effect that field shape perturbations (e.g. symmetry variations) have on WF calibration reproducibility. Materials and Methods Materials The detector array used in this study was the IC PROFILER [Sun Nuclear Corporation (SNC), Melbourne, FL USA], wh ich will henceforth be referred to as the panel. The panel contains 251 semi-cylindrical ionization chambers and an active area of 32 x 32 cm2, Fig. 4-2 A detector array is located on each coordinate axis (x and y) and on the negative and positive di agonal axes (nd and pd, respectively). The detector spacing is 0.5 and 0.7071 cm on the coordina te and diagonal axes, respectively. All axes share a common center detector, with the xand diagonal axes missing detectors immediately adjacent to the center detecto r. The inherent thicknesses of buildup and backscatter are 0.9 and 2.3 g/cm2, respectively. 79

PAGE 80

The 6 MV X-ray beams used in this study were produced by either an Elekta Synergy (Elekta Limited, Crawley, SXW UK) or a Varian Trilogy (Varian Medical Systems, Palo Alto, CA USA). Each provided different beam characteristics that were useful in evaluating the calibration al gorithm; their use wil l be acknowledged accordingly. Methods This inquiry into the calibration method was for a single detector array, in this case the panels y-axis. This was done to decrease the description length of the calibration procedure and also to ease the reporting of results. The procedure for determining the orthogonal axis (x-axis) calibration factor s was not changed from the original method and will,43 therefore, not be covered. Calibrati on of the pdand nd-axes follows the methodology of the yand x-axes, respecti vely, and was not covered for the same reason. Wide field calibration theory While the WF calibration was described in detail by Simon et al.,43 a brief overview is required for this work. The algorithm correc ts intra-array detector sensitivity variations by applying detector specific calibration fact ors in the form of scalar multipliers. For a linear detector array, the calibration factors are determined with three array measurements ( and ). For the array is centered on the LINAC crosshairs. For the array is rotated 180 from its position in For the array is shifted by one detectors spacing from its position in The beam parameters are const ant during each measurement. The primary beam requirement is th at the field size needs to be larger than the active area 80

PAGE 81

of the array (35 x 35 cm2 for the full panel). This is to av oid critical spatial substitution in the high field gradients of the beams penumbra. The WF calibration algorithm operates under three postulates imposed during the three array calibration measurements. The fi rst is that the deliv ered dose distribution does not change during a complete calibration. The second is that the relative sensitivities of the detectors do not change. The third is that the movement of the detector array from measurement to m easurement does not c hange the scatter conditions to the LINAC-space reference-fr ame, i.e. the array phantom heterogeneities have substitution symmetry. In theory, a detector array can be calibrated with only and This is accomplished by determining the relative det ector sensitivities through ratios of the detector readings at the same field locations. The sensitivity of any detector relative to the first is then expressed as = = + 1 1 1'n i i i ncf (4-1) where is the theoretical calibration factor for detector n 'cf { } :2,3,4,...,nEb and E is the last detector of the array; by definition = 1. 1'cfUsing Eq. 4-1 requires invariant dose deliveries and detector sensitivities between and Changes in either will introduce error that will be mistaken as differences in the detectors relative sensitivities. For this reason, the calibration procedure needs to account for the changes in dose delivery and global detector sensitivity. This is accomplished by introducing a correction term into Eq. (DS)4-1 the derivation of 81

PAGE 82

which can be found in the WF calibration patent.43 The calibration factors are now calculated using (1 1 1 1 = + =n n i i i nSD cf ) (4-2) where is the calibration factor for detector n cf { } :2,3,4,...,nb 1cfE, D is the change in delivered dose between measurements and and S is the change in global detector sensitivity between measurements and Again, by definition = 1. The changes in dose and/or detector sensitivity from to are difficult to quantify. One method of doing so is to monitor the dos e delivery with a reference detector. This is difficult since an incomplete compensation of these errors would be propagated by the product sequence term in Eq. 4-2 With the measurement, a symmetry solution exists by determining the mirror calibration factors, i.e. the sensitivity of a detector relative to its mirror about the arrays center. Mirror calibration factors (cfm) can be determined using two methods. The first involves calculating the relative calibrati on factors of mirror detector pairs with a variation of Eq. 4-2 This is expressed as ()(1 2 1 1 n nE i n iEn icfm DS + =+ += )1 (4-3) where n = [(E+1)/2]+1 thr ough E, when E is odd. The second method for calculating mirror calibration factors involves measurements and Using these measurements, the mirror calibration fact ors can be calculated with 1 EnEn n nncfmt t++=1. (4-4) 82

PAGE 83

where n = [(E+1)/2]+1 th rough E, when E is odd. Two solutions now exist for each mirror calibration factor which allows us to solve for DS in Eq. 4-3 The mean DS value for all mirror pairs is used to reduce any error in this term. The fi nal expression for calibration factors is expressed as (1 1 1 1 = + =n n i i i nSD cf ). (4-5) For convenience, cf is re-normalized to the center detector, i.e. cfcenter = 1. The calibration algorithm was recreat ed in the MATLAB (The MathWorks, Natick MA, USA) software environment, whic h allowed complete control over the calibration procedure. Calibration factors created in MATLAB match those created in the panels software. Limiting calibration error Since the WF calibration algorithm can produce reproducible results [see Fig. 41 B], the primary source of calibration e rror is related to the delivered beam and violations of the first postulate. Limiting t hese postulate violations should reduce WF calibration errors. For comprehensiveness, the other two pos tulates are also examined. Effects of postulate failure The effect of small postulate violations was quantified by applying a sine shaped perturbation () p erty to hypothetical calibration fields effectively simulating a field perturbation due to a postulate violati on. The perturbati on was created with ()0.001sin 1 32cmy pertyn= +, (4-6) 83

PAGE 84

where y { } :16.5,16,15.5,...,16yb is the in-line posit ion (cm) on the LINAC coordinate system. The pertur bation reaches a maximum of () 0.1% at the array edges. A single perturbed field was paired with unperturbed fields in the calibration algorithm to produce cfX, where X represents the perturbed field ( or ). For example, applying pert(y) to but not to and results in cf. Using all unperturbed fields creates a baseline calibration factor (cf ). The effect of perturbati on was quantified by determining the percent error (X_,n) p errorcfcf between cfX and cf for each detector n. This is expressed as () X X_, 1100n ncf perrorcfcf cf = (4-7) It will be shown in the results section that the WF calibration algorithm is sensitive to postulate violations. Specifically with the magnitude defined in Eq. 4-6 perturbations to or result in calibration e rrors of up to 2%, while is nearly immune. Limiting violations of the first postulate The first postulate assumes that the dose distribution does not change between measurements. The validity of this assumpti on was evaluated by determining the in-line beam reproducibility of the two LINACs (from Fig. 4-1 ) over 10 consecutive measurements. Reproducibil ity was quantified by determining the percent error [(_n) p errornmeas] between each detectors measurements to their mean value. This is expressed as () 1100n nnmeas perrornmeas nmeas = (4-8) 84

PAGE 85

where nmeas is the measurement no rmalized to the center detector and nmeasis the mean nmeas value for each detector n. It will be shown in the results section that the dose distribution reproducibility is improved while the beam is continuously ru nning, instead of being cycled on and off. Because of this and the sensitivity of and to postulate violations, the array was calibrated on the Elekta wit h a continuous beam during and Movement was provided by a linear stage motor (model NL S4-2.5-16; Newmark Systems, Mission Viego, CA USA) that was capable of sub-mm movement. Measurement was a standard cycling of the beam on and off. This process was repeated four times for a total of five calibrations. Limiting violations of the second postulate The second postulate assumes that if t here is a change in detector sensitivity during the calibration, then all of the relative detector sensitivities change by the same amount. Failures of this postulate are difficult to quantify since the detector sensitivities are the values being pursued in the WF calibra tion. Actions can be taken to reduce the likelihood of a postulate failure. These includ e storing the panel in the environment that it will be measuring and also maintaining a constant power supply to the panel. If this is impossible, power should be applied fo r several minutes prior to making measurements.53 For example, if a panel was stored in an area whose temperature is significantly different than the temperature of the treatment vault, then as thermal equilibrium takes place, the temperature gradients across t he array may differentially influence the relative sensitivities from one region to another. Another and more di fficult example is 85

PAGE 86

the beams off-axis energy distribution and it s effect, if any, on the energy response of the detectors. With the rotational detector substitution ( to ), there will be no effect if the energy distribution has radial symme try. But with the linear array shift ( to ), all the detectors move into a new energy distribution, albeit ever so slightly different. Neither of these effects is likely to be large; however we do see significant calibration errors induced with very small beam symmetry changes. Limiting violations of the third postulate The third postulate assumes that t he detector arrays movement between measurements does not change the scatter c onditions encountered by the array. While the array is radially symmetr ic, that symmetry begins to br eak down near the distal ends of the panel. This is well illustrated by t he panels y-axis. The positive y-axis has a reduced amount of side-scatter relative to the negative y-axis, which is adjacent to the panels electronics (see Fig. 4-2 ). Furthermore, the recommended field size for calibrating the panel is 35 x 35 cm2 at a source to panel surface distance of 100 cm; the panel size is 35.3 x 35.3 cm2. The proximity of the fiel d to the panels edge creates a dynamic scatter condition as the panel is moved through the steps of calibration ( and ). The effect that additional side-scatter has on the measured dose distribution was quantified by taking the ratio between the mean of five normalized measurements with (ss) and without additional side-scatter. This is expressed as () ,ss ss n nnmeas responsenmeasnmeas nmeas = (4-9) 86

PAGE 87

where (,ss nresponsenmeasnmeas) is the effect that additi onal side-scatter has on the measured beam shape. The side-scatter was cr eated with a 4 x 4.2 cm (width x height) acrylic border that completely surr ounded the three open sides of the panel, and therefore the array. The width of the addi tional side-scatter was chosen to mimic the scatter conditions of the negative y-axis, while the height matched the thickness of the panel and the additional buildup (1 cm) that was used during calibrations. It will be shown that additional side-scatter at the positive y-axis perturbs the measured dose distribution. The effect this has on the WF calibration was determined by calibrating the array five times with and without the additional side-scatter. The Varian LINAC was used for this portion in an effort to reduce the influence of symmetry variations from the Elekta during beam on/off cycles. The agreement [ (,n ssagreementcfcf)] between the two sets of calibra tion factors was evaluated using () ,ss n ss ncf agreementcfcf cf = (4-10) where s scf and cf are the mean calibration fa ctors determined with and without additional side-scatter, respectively. Evaluating calibration factors The effect of modifications to the WF calibration methodology was evaluated in two areas: reproducibility and accuracy. Repr oducibility was quantified by determining a) the coefficient of variation (CV, relati ve standard deviation) and b) the percent error [p_errorn(cf )] of each detectors calibration factors (cf ) relative to their mean value ( cf). This is expressed as 87

PAGE 88

() 1100n ncf perrorcf cf = (4-11) Long term reproducibility was evaluated in t he previous work on the IC PROFILER using the standard WF calibration pr otocol and the Varian LINAC.53 Since the calibration algorithm was unchanged and the results were obtained with the Varian LINAC, which provided stable array calibra tions, the long term reproducibility was not re-evaluated in this study. The WF calibration inter-LINAC reproducibility was evaluated with the Varian and Elekta LINACs; five calibrations were done on each unit. The calibrations on the Elekta used a continuous beam (during and ) and additional side-scatter. Those done on the Varian used additional side-scatter but not a continuous beam; thereby testing the effects of using a continuous beam as well. Their agreement (),n EVagreementcfcf was quantified by determining the ratio between the mean of the two sets of calibrations. This is expressed as () ,E n EV V ncf agreementcfcf cf = (4-12) where Ecf and Vcf are the mean calibration factor s obtained with the Elekta and Varian, respectively. The accuracy of the array calibration was evaluated by taking two full field measurements with a 180 panel rotati on between measurem ents. The 180 measurement had its data rota ted about the center detecto r to maintain the LINAC coordinate system of the 0 meas urement. The calibration accuracy 88

PAGE 89

(180 0,posaccuracynmeasnmeas ) was then quantified by taking t he ratio of the 180 to the 0 measurement. This is expressed as () 180 180 0 0,pos posnmeas accuracynmeasnmeas nmeas = (4-13) where nmeas0 and nmeas180 are the normalized measur ements for the 0 and 180 panel orientations, respectively, and pos is the spatial position on the LINAC axis. An accurate calibration would result in mini mal differences between the two measurements since measurements at a point in space should be detector invariant if the detectors response is known by calibration. However, there may be a subtle effect across the panel from the off axis energy distribution. It would probabl y not be detected in a mirror symmetry test using a 180 rotation measur ement comparison. To address this, we compared to a scan in water. The calibration accuracy was also evaluated by comparing panel and scanning water tank [Blue Phantom and CC13 (IBA, Schwarzenbruck, DE)] profile measurements of a 6 MV X-ray beam (Elekt a). Four square field sizes were measured (with sides of 5, 10, 20, and 30 c m) at two depths (1.5 and 10 g/cm2) and a source to surface distance (SSD) of 90 cm Two array calibrations were evaluated that used the continuous beam (Elekta LINAC) in either the presence or absence of additional sidescatter. The profile agreement was quantified by taking the ratio of panel to water tank measurements. This is expressed as (,pos panel wateraccuracynmeasnmeas )() ,panel pos panel water water posnmeas accuracynmeasnmeas nmeas = (4-14) 89

PAGE 90

where nmeaswater and nmeaspanel are the normalized measurem ents for the water tank and panel, respectively, and pos is the spatial position on the LINAC axis. Results and Discussion Effects of Postulate Failure The perturbation that was applie d to the hypothetical calibration fields is seen in Fig. 4-3 A; the inset offers an expanded view. This shape was chosen because it closely resembles the actual symmetry variati ons in beam on and off cycling that were observed with the Elekta LINAC (these result s are presented in the next sub-section). Within real measurement and data precision limits, it can be di fficult to track the effects of a field perturbation on the calibration. The val ue in this hypothetical analysis lies in its precise ability to quantify the effects that small perturbations have on the calibration algorithm. The calibration error due to simulat ed perturbations is seen in Fig. 4-3 B. When applied to the perturbation resulted in minimal c hanges [() 0.07%] relative to the baseline calibration. The relative immunity of this measurement to postulate violations is due to its role in determining the DS correction (see Eq. 4-4 ), which has limited potential for error propagati on. Perturbation applied to or resulted in calibration errors of up to () 2%. This strong respons e highlights the potent ial for error propagation due to the product sequence term in the WF calibration algorithm (see Eq. 4-5 ). While not covered in this work, the effect of perturbations to the orthogonal (x-axis) calibration factors is minimal. Their method of calculation lacks a linear shift and hence the same potential for error propagation.43 The x-axis cfs do, however, use the y-axis cfs in their calculation and are hence suscept ible to the errors associated with them. 90

PAGE 91

Limiting Violations of the First Calibration Postulate The potential of a 2% calibration error due to a 0.1% dose distribution error seems to preclude the WF calibration algori thm from generating consistent results. Measurement errors that exceed this are unavoidable. However, if we recall the results from Fig. 4-1 B, we know that the WF calibration can produce reproduc ible calibration factors. Therefore, the variations in calibration reproducibil ity that were observed in Fig. 4-1 A must be related to the LINAC and the first calibration postulate. This conclusion is supported by compari ng the reproducibility of the Elekta and Varian dose distributions. In Fig. 4-4 A, we see the in-line dose distribution variations for the Elekta LINAC. The overa ll variation was small, () 0. 15%. However, some of the variations closely resembled the perturbati on from the previous subsection, making them the likely source of the WF calibration errors. Similarly, we would expect the Varian LINA C to have a tighter error spread for its beam. In Fig. 4-4 B, we see this to be the case. T he first two measurements (1 and 2) have a higher noise component compared to the subsequent measurements. The decrease in noise from the first to the second measurement and the relative absence thereafter indicates that this phenomenon is due to the pre-irradiation effect.54 While the first two measurements have increased noise, t he shape of their dose distributions was consistent with the subsequent measurements. This indicates a remarkably small amount of beam variation. It must be emphasized that these two mach ines represent sample sizes of one. Each is in clinical use and the observed va riation of 0.15% is insignificant to the intended use of clinical treatment. 91

PAGE 92

The increased error observed with the center detector [see Fig. 4-4 B] has a limited effect on WF calibration reproducibility. A noise event at a single detector will only contribute to error propagation once; all su bsequent detectors will be off by that amount. However, that level of error (0.15%) is minimal. The increased error in the other detectors is biased both positively and negatively. These cancel each other and therefore do not contribute significantly to calibration errors. While the pre-irradiation effect is small, it should be considered when performing WF calibrations. The likely cause of the beam variations in Fig. 4-4 A is small fluctuations in the LINACs electron spot. As the beam is cycl ed on and off, the spot has micro variations in its location and shape. This causes variations in the impingement of X-rays on the flattening filter, which consequently alters the delivered dose distribution. Turning the beam on and leaving it on produces less variation in the electron spot and ultimately a more stable dose distribution. Evidence of this is seen in Fig. 4-4 C. The cross-line symmetry is also likely to be more stable t han the in-line, making it a better candidate for the WF calibration. However, this did not always hold true for other machines that were tested. The p_error of calibrations done with a continuous beam (during and ) on the Elekta are shown in Fig. 4-5 The maximum error decreased fr om () 1.6% to () 0.68%, Table 4-1 This method did not entirely eliminate the error trends that were observed in Fig 4-1 A, which indicates a reduced but continued variation in the dose distribution between and Limiting Violations of the Third Calibration Postulate The effect that additional side-scatter has on the measured calibration beam is shown in Fig. 4-6 A. An increase in signal begins at approximately (+) 10 cm on the y92

PAGE 93

axis and increases up to 0.35% at the array s end. Additional side-scatter had a minimal effect on the negative y-axis, which is expec ted since no additional side-scatter was added here due to the presence of t he panels electronics (see Fig. 4-2 ). The effect that additional side-scatter had on calibration factors is shown in Fig. 46 B. An asymmetric response occurred in which the negative y-ax is cfs under responded by up to (-) 0.35% and the positive y-axis cfs under responded by a maximum of (-) 2.21%. These results indicate that the sidescatter conditions are not adhering to the third postulate. Evaluating Calibration Factors The inter-LINAC agreement of calibrations is shown in Fig. 4-7 The two calibrations agree within () 0.4% of each other. Since both calibrations used sidescatter, this tested the viability of using a continuous beam (Elekta) to match the calibration results obtained with a standar d on and off beam (Varian). The general agreement between the two sources shows that the algorithm produces consistent results for different LI NACs when the continuous beam approach is used. The accuracy of the continuous calibration factors was determined with four measurement (meas) and calibr ation (cal) combinations. These were created with (w/ss) and without (n/ss) additi onal side-scatter. The calibration accuracy is shown in Fig. 4-8 ; the symmetric error indicates calibrati on errors. The smallest deviations from unity [() 0.4%] occurred when addition al side-scatter was used during both the calibration and the measurement. The wo rst combination [() 2%] involved measurements with additional side-scatter paired with a calibration that did not use additional side-scatter. These results indicate that additional side-scatter should be used during calibration and for best result s, during measurements. However, the 93

PAGE 94

penalty for not using side-scatter during meas urements is approximat ely 0.5% at the extreme array ends. In Fig. 4-9 we see the profile agreement between the panel and water tank measurements. Two panel calibrations were used that were acqui red with a continuous beam (Elekta LINAC) in the presence and ab sence of additional side-scatter, Figs. 4-9 A and B, respectively. The additional side-sca tter calibration improved the agreement between the two detector systems for t he shallow measur ement (30 x 30 cm2), reducing the maximum error from to 1.3% to 1.0%. However, for the 10 g/cm2 measurement (30 x 30 cm2), the side-scatter calibration reduced t he agreement from a maximum error of 0.87% to 1.0%. For smaller fields (< 30 x 30 cm2) the panel measurements were typically within () 0.5% of the water tank. Judging by the absolute change in erro r, using additional side-scatter during calibration improved the overall agreement. However, the relatively steep increase in error for the 10 g/cm2 measurement beginni ng at approximately (+) 8 cm is suspect. That position coincides with the beginning of the WF calibrations response to sidescatter [see Fig. 4-6 B]. The benefit of using additional side-scatter during calibration is not entirely clear. It provided the most reproducible calibrations, the best accuracypos(nmeas180,nmeas0) values, and improvement in the shallow pr ofile agreement. It did, however, worsen the agreement for the 10 g/cm2 measurement. The use of addi tional side-scatter during WF calibration is under continued investigation. Other Factors Affecting the WF Calibration The response of the arrays detectors to external stimuli (e.g. beam energy and additional buildup) may produce WF calibra tion dependence to t hat stimulus. In 94

PAGE 95

previous work to characterize the IC PROFILER, the dependence of the WF calibration to these two variables was quantified.53 The Varian LINAC was used with the standard calibration procedure since it produc ed stable calibrations. The methodology of these experiments was covered in the text of that manuscript. There was minimal calibration response to addi tional buildup (< 0. 8% for 0.9 to 4.9 g/cm2) and a slight response to energy (< 1.5% for 18 to 6 MV X-rays).53 The recommendations of that work were to calib rate the array with buildup that met or exceeded the energy specific d-max and to use an energy specific calibration. Other Arrays and IMRT The majority of arrays using the WF ca libration algorithm have a reduced number of detectors in their translational arrays relative to the panel. For example, the MapCHECK 2 (SNC) has 33 detectors in it s translational array, versus 65 for the panel. This lower number reduces the amount of error propagation that can occur in Eq. 4-5 As a result, the MapCHECK 2 calibration error would likely be half of the panels on the same LINAC. An important question to ask is: how does this affect patient specific QA for IMRT? For the case of a MapCHECK 2 on a LINA C with symmetry variations of () 0.15%, the error would likely be half of what was observed with the panel, making it roughly () 0.8%. Using passing criteria of 3%/3mm, a calib ration error of this size is unlikely to artificially pass an unsuitable plan. Howeve r, certain conditions could lead to biased clinical decisions if errors stack up in the same direction. Conclusion The aim of this work was to reduce the effect that minor measurement perturbations have on the calibration reproducibi lity. These occur via violations of the 95

PAGE 96

three WF calibration postulates, which st ate that a) the beam shape does not change between measurements; b) the relative sensit ivities of the detectors do not change; c) the scatter does not change as the array is moved between measur ements. During WF calibration, perturbations to the beam shape of 0.1% can lead to calibration errors of () 2%. Postulate violations were limited by us ing a continuously on beam during portions of the calibration procedure. This increased the beams stability and therefore limited the error propagation that occurs in the algorit hm. Additional side-scatter was also added to the IC PPROFILER to increase the scatter uniformity. The overall effect was to reduce the calib ration error from appr oximately () 1.6% to () 0.48%. The agreement between m easurements done with the panel and a scanning water tank were () 0.9%. The ca libration error for arrays commonly used in patient specific QA (IMRT) would likely be half of what was observed with the IC PROFILER, which would be unlikely to affect a clinical decision using a distance to agreement of 3%/3mm. This work increased the reproducibility of the WF calibration procedure for LINACs that hav e minor symmetry variations. 96

PAGE 97

Table 4-1. The short term reproducibility of WF calibrations performed on the Elekta using either the standard protocol or the continuous beam [with and without additional side-scatter ( ss)]. Standard Continuous without ss Continuous with ss Mean CV (%) 0.56 0.25 0.15 Maximum CV (%) 1.4 0.57 0.33 Maximum p_error (%) 1.6 0.68 0.48 97

PAGE 98

A B Figure 4-1. Wide field (WF) calibration reproducibility on LINACs with beam symmetry variations of (A) () 0.15% (Elekta LI NAC) and (B) () 0.05% (Varian LINAC). Figure 4-2. Oblique view of the panels arrays and electronics. The panels y-axis is parallel to the long ax is of the device. 98

PAGE 99

A B Figure 4-3. (A) The perturbation that wa s applied to the hypot hetical calibration measurements. (B) The effe cts of perturbations applie d, separately, to each calibration measurement. A B C Figure 4-4. The percentage error between t en consecutive measurements and their means for (A) a LINAC that produces ( ) 1.6% calibration reproducibility, (B) a LINAC that produces () 0.5% calibration reproducib ility, and (C) the same machine as used in (A) but with the beam continuously on. The spikes in the center detector of (B) were due to the pre-irradiation effect of ionization chambers and do not contribute significantly to calibration errors. 99

PAGE 100

Figure 4-5. Calibration reproducibility usin g a continuous beam during measurements and on an Elekta LINAC; no addi tional side-scatter was used. A B Figure 4-6. The effect of additional si de-scatter on (A) beam measurements and (B) array calibrations. The measurements and calibrations were performed on a Varian LINAC. Figure 4-7. The agreement between calibration factors obtained with side-scatter and either a continuous beam (Elekta) or a standard on/off delivery (Varian). 100

PAGE 101

101 Figure 4-8. The calibration accuracy was evaluated for four measurement combinations. These included the pres ence (w/ss) or absence (n/ss) of additional side-scatter in the measurem ent (meas) and/or calibration (cal). A B Figure 4-9. The calibration accuracy expressed as the ratio between water tank measurements and panel measurements. The panel calibrations were performed using a continuous radiation s ource in the (A) presence of and (B) absence of additional side-scatter.

PAGE 102

CHAPTER 5 A QUALITY ASSURANCE PROGRAM FOR A DETECTOR ARRAY Introduction A vital, but oft overlooked, component of an effective quality assurance (QA) program is evaluating and ma intaining the measurement equipment. The importance of this is immediately apparent for certain systems (a local standar d ionization chamber) and in general was addressed by the American Association of Physicists in Medicine (AAPM) in a report by Task Group (TG) 40 of the Radiation Therapy Committee.5 However, since TG-40 was published in 1994 measurement equipment has changed dramatically, tending toward more electronic and complex systems. Detector arrays are a prime example of th is trend. An argument can be made that their role in the clinic is every bit as impor tant as that of more traditional systems. Their use as a benchmark dosimetry system for in tensity modulated radiation therapy (IMRT) plans helped to usher in that modalities widespread use.22, 45, 55, 56 Detector arrays play various other important roles including the measurement of linear accelerator (LINAC) symmetry and flatness levels and ca librating mechanical components.33, 44 Since TG-40 was published, there have been few new QA guidelines suggested for measurement systems. In contrast, there have been numerous characterization papers that have accompanied new products.44, 51-53, 57, 58 While these are important pieces of work that establish the behavior of the system in its intended environment, they do not provide practical guidelines that a physicist c an use to evaluate the devices continued operation. Recommendat ions are also typically lacking in the device manuals. The purpose of this work was to creat e a QA program for detector arrays. 102

PAGE 103

First, a question needs to be asked: why is there a lack of guidance in this area? The primary reason is that it is difficult to establish recommendations that are based on hard science. Instead, most QA recommendat ions, including those for LINACs, are based on experience and/or conv ention. This work is not, and cannot be, completely different from this behavior. But it will, wher e practical, use sound scientific reasoning to establish tests and tolerances. Materials and Methods Materials This work is intended to be general and app licable to most detector arrays. A specific array was used to illustrate certain portions of the QA program. That array is the IC PROFILER [Sun Nuclear Corporati on (SNC), Melbourne, FL USA] and will henceforth be referred to as the panel.53 The panel contains 251 semi-cylindrical ionization chambers and has an active area of 32 x 32 cm2. A detector array is located on each coordinate axis (x and y) and on t he negative and positive diagonal axes (nd and pd, respectively). The panel performs measuremen t updates at near integer multiples of a preset collection interval. This effectively allows the panel to measure an indefinite amount of radiation by not saturating the capacitors. The measurement file can either preserve these updates (in the form of a running cumulati ve signal) or the final cumulative update is kept. A file that preserves each update is referred to as a multi-frame measurement. The option between multiple and single frames is available due to file size. For example, a 30 s multi-frame measurement with 125 ms collection interval will produce a file that is approximately 450 kb, while a single fram e measurement is approximately 10 kb. 103

PAGE 104

Panel measurements can also be operated in either pul sed or continuous mode. Pulsed mode measurements ar e synced with radiation pul ses. Continuous mode measurements are clock synced. Both have intended uses and associated qualities.53 The 6 MV X-ray beams used in this study were produced by an Elekta Synergy (Elekta Limited, Crawley, SXW UK). Methods The QA program is broken into four ar eas: physical, personal computer software and device firmware, electronics, and array calib ration. When a test is described, it will generally refer to an array; portions of t he program will then be applied to the panel in the results and discussion. The tests, their frequencies, and acti on levels specific for the IC PROFILER, at our institut ion, are consolidated in Table 5-1 Physical Detector arrays are relatively robust compared to typical dosimetry systems (thimble ionization chambers, ra dio sensitive film, scanning wa ter tanks, etc.). This is primarily due to their packaging and an inherent lack of moving parts and water. Despite this, they are still complex systems and need to be treated accordingly. Before each use of the array, it should be examined for any physical damage. Relatively little can be done on site if severe physical damage occu rs. Two items that can be continually checked and taken care of are the arrays buildup and power/data cables. Buildup: Extended use will accumulate grime on an arrays top surface (including additional buildup). While this poses little to no risk to the actual device, it does increase the friction between the array and added buil dup. Consequently, there is a higher likelihood of the array being pushed out of position wh ile adding or sliding buildup across the top surface, thereby skewing the measurement geometry. 104

PAGE 105

As accumulation occurs, the top surface (and additional buildup) is cleaned with a micro-fiber cloth moistened with a small amount of water. A ll power to the array should be disconnected during cleaning and extreme care should be taken so that water is not allowed to flow or collect on any surface. Cables: Power and data cables are a common failure point for devices and occur almost entirely due to impr oper handling. Cables should always be disconnected by grasping and pulling the connector rather than the cable. Pulling the cable will stress the connection between the connector and the co rd, which can expose conductors. Failures also occur due to repetitive circular movement. This is caused by taking an array (attached to a cable) out of a cabi net and turning clockwise to put it on the treatment couch; after the measurement session, the array is then put back into the cabinet using a second clockwise movement. The result is a cable that is continually twisted and strained, and can quickly wear out through frequent repetitions. Cables that are separating from the connec tor or that are severely twisted should be replaced immediately to prevent personal injury, physical damage to the array, and loss of measurement capability. Firmware and software It is important to stay current with softw are for reasons of defect correction and improved performance. It is al so important to report defec ts and desired improvements that contribute to confidence in measurem ents. Firmware and software updates occur at different frequencies for different manufactu rers. Ideally, an arrays software would automatically inform a user when an update is available. A solution that SNC is incorporating is predefined bi-annual release dates. Knowing these dates, the user can 105

PAGE 106

efficiently schedule an update after the pl anned release date has passed. Otherwise, updates are checked for based on the manufacturers release patterns. Measurement consistency: Before applying an update, a benchmark setup measurement (meas1) is made twice and check ed for consistency. In this situation a benchmark setup measurement is defined as a set of invariant beam and measurement parameters [e.g. for the panel the beam par ameters could be a 6 MV X-ray beam, 33 x 33 cm2 field size, no added buildup, and a 100 cm source to surface distance (SSD)]. Without disturbing the array, the update is app lied and a second benchmark measurement (meas2) is made twice; thes e are also checked for consistency. An inconsistency between the two measurements could be the result of a number of different problems: instable beam delivery, a temperature gr adient in the array, or the pre-irradiation effect.54, 59 If an inconsistency is identified and corrective actions are taken, two measurements should be made again and checked for consistency. The goal is to be certain that the array measurements are not influenced by source outside of the software update, which could invalidate this test. Using the first measurement from each set, the percent e rror (p_error) is determined for detector n with () 1 1,2 1100 2n nmeas perrormeasmeas meas = (5-1) Since the geometry was unchanged between measurements the difference in data should be due to LINAC and measur ement noise. Large differences are the result of the software or the firmware update. Over a short time frame, the panels ma ximum percent error versus a mean was reported as () 0.15%;53 similarly the LINACs maximu m error was also reported as 106

PAGE 107

approximately () 0.15%.59 Accounting for other possible sources of error,54, 59 we can estimate a maximum amount of measured erro r at approximately () 0.5%, this value can be looked on as a tolerance level. Similar analyses can be used to set tolerance levels for other arrays. This test is only va lid for detectors within the radiation field. Detectors outside of the direct beam have a lower signal to noi se ratio, which can result in false positives. A software update that results in a legiti mate difference should be accompanied by an explanation and also should be investi gated for its impact on past and future measurement applications. For example, a software bug that improperly calculates symmetry may have produced a symmetry value t hat is used to justify IMRT patient plan QA agreement. Data consistency: Following an update, t he consistency of previously measured data is evaluated. A baseline measurement that was taken upon first receiving the array is opened in the arrays software and the data is copied (or exported) to a spreadsheet program (Excel; Microsoft Corporation, R edmond, WA USA) with previous exports of the same measurement; each export should be identical. Any difference is a deviation and should be accompanied by an explanation from the vender. A software update that results in a difference should be investigat ed for how it impacts past and future measurements. Software features: The proper function of so ftwares analysis features (flatness, symmetry, index, etc.) is an important aspect of a QA program. Their QA is beyond the scope of this work given their variety amongst all manufacturers and devices. 107

PAGE 108

However, fundamental and essential QA c an be accomplished by trending their output of a benchmark meas urement setup. Electronics The operation of the arrays electroni cs and detectors is evaluated background and beam measurements, both of which contain unique and separable information about the system performance. Visual examination of background: At the beginning of a measurement session, a background measurement is performed and visually examined. The appearance and amplitude of the signal should be consistent with previous background measurements. Frequent use of the detector array provides the user a visual memory of prior background appearance; however, if the ar ray is used infrequently, a previous background measurement can be opened for compar ison. Despite its simplicity, this is a powerful test since a systematic influence wi ll result in a deviation from the arrays normal behavior. This test is done bef ore each measurement session. Abnormal behavior should be investigated before the array is used. Test of background normality: A more rigorous test of the measured background is to examine its probability density function (PDF) for normality. A convenient method of doing this is to make a background measurem ent with the array oper ating in continuous multi-frame mode. The raw counts that occur during eac h update (accumulated charge measurement) are then calculated by taki ng the difference between the successive measurement updates. This is expressed as () () () () ()11 a n d 1,nn nnnUCRawCounts UCiiRawCountsiiRawCountsii = = (5-2) 108

PAGE 109

where UC is the raw counts per update ii { } :2,3,4,..., ii Lb, L is the last measurement update, and n is the detector. Since UC is a measure of background, its PDF should be normal. The normality of UC was evaluated using two goodness of fit (GOF) tests: Chisquared and Anderson-Darling.60, 61 Chi-squared was chosen because it is a common GOF test and serves as a benchmark; Anderson-Darling is better suited at assessing the normality of an observed PDF.62 The null hypothesis (H0) was that UCs PDF was normal for each detector; t he alternative hypothesis (Ha) was that each detectors PDF was non-normal. By convention, a significance level ( ) of 0.05 was used; this level can easily be increased, but with the risk of increasi ng Type I errors (i.e. rejecting H0 when it is true). Both GOF tests are easily implemented in a spread sheet program such as Excel. A rejection of H0 at the selected significance level is evaluated for the magnitude of rejection. Rejected values that are close to should be repeated. A second rejection likely indicates a biased system. Ideally, this test would be performed befor e each measurement session. However, if a highly integrated version of the test is not available, then it can be used after an anomaly is encountered during the visual inspection of the background measurement. The test should also be per formed at least annually. Dose linearity: The dose linearity of the ar ray is evaluated with LINAC deliveries of 100 and 200 monitor units (MU). Each is de livered fives times; these measurements will also be used to evaluate the arrays reproducibility. The measur ements are performed at an SSD of 100 cm to the arrays surface a nd with a buildup that is at least as much 109

PAGE 110

as the energy specific d-max. The deliver ies are simultaneously measured with a local standard instrument (LSI ) in a 4 x 30 x 30 cm3 slab of water equivalent material that is located below the array and align ed to the LINACs cross hairs. The protocol of TG-51 is used to conv ert the charge measured by the LSI to dose.42 While this setup violates the protocol of TG-51, the measur ement is proportional dose and can be used as a normalization method to compensate any changes or deviations in the delivery system and the air density. The measured signal from the panels center chamber and the independent system are normalized to the 200 MU delivery. The arrays response is then normalized to the LSIs response using, () () () () () 100 200 100 200panel LSIMEASMU MEASMU responseMU MEASMU MEASMU = (5-3) where response(MU) is the arrays dose lineari ty relative to a Farmer-type chamber and MEAS is the measured signal for each detector system (CC or cGy for the array or LSI, respectively). This test is done annually. The toleranc e level is based on the users accepted level of error. For example, TG-142 recommends a dose linearity tolerance level of () 2% for exposures greater than or equal to 5 MU.17 Therefore the array should have a dose linearity error less than 2%. For example, if the arrays dose linearity is 0.5%, then the balance of the error budget for setup and LI NAC is 1.5% which effectively places a tighter limit on the reco mmendations of TG-142. 110

PAGE 111

Measurement reproducibility: The measurements from the dose linearity section are used to evaluate the arrays short and l ong term reproducibility. The protocol from TG-51 is once again used to convert the charge measured by the LSI to dose.42 The short term reproducibility is eval uated using five 200 MU dose linearity measurements. Each panel measurement is normalized to the LSI measurement, resulting in NORM. This is done to correct fo r any variations in the LINACs output and will be of value for the long term reproducib ility. The percent error [p_error(NORMii)] between each normalized measurement ii and the mean of all the normalized measurements is determined using () 200_ 1100ii ii MUNORM perrorNORM NORM = (5-4) where NORMii is the normalized measurement and NORM is the mean of all normalized measurements. Long term reproducibility is evaluated for the mean of a meas urement set kk versus the mean of the baseline measur ement set. This is expressed as () 200_ 1100kk kk baseline MUNORM perrorNORM NORM = (5-5) where NORM is the mean of measurement set kk or the baseline set. These tests are done annually. The tolerance level is based on the users accepted level of error. Array calibration An important property of detector arrays is a unifo rm exhibition of detector sensitivity to radiation. Otherwise, array measurements will be prop ortionally biased by each detectors sensitivity. Many met hods have been developed to normalize these 111

PAGE 112

sensitivity variations;43, 49, 50 among these, the wide field (WF) calibration is commonly used due to the ease and efficiency with which it can be performed on site.45, 52, 59 Since many arrays use WF calibration, this work will focus on it, but in general the QA program should still be applicable to other calibration techniques. Reproducibility: Since many arrays are user calibrated, it is the users responsibility for verifying the arrays calibration accuracy. An initial step is determining the calibration reproducibility; this can be quickly evaluated for two successive WF calibrations using () 2 21 1_, 1100n ncf perrorcfcf cf = (5-6) where p_errorn(cf2,cf1) is the percent error between the two successive WF calibrations (cf1 and cf2, respectively) for detector n. The user needs to determine the level of calibration error acceptable for the application, as derived from requirements in task group reports, original equipment maker (OEM) standards, acceptance testing, etc. For example, IMRT QA that uses tight passing criteria will have less room to acco mmodate calibration error. A 2% calibration error coupled with IMRT passing criteria of 3%/3mm will leave a window of 1% for all other errors (TPS algorithm, measurement, set up, etc.). The array ca libration error must at least be less than the passing criteria or the tolerance of the item that is being measured. In contrast, an array that is solely used for tra cking profile deviations from a baseline measurement has no real calibration reproducibility constraints. The frequency for evaluating the calibration reproducibility is during an array calibration, which should occur annually. During the annual calibration, the short and 112

PAGE 113

long term reproducibilitys are determined. For long term stability, Jursinic and Nelms estimated that annual calibrations would be sufficient to mainta in an accurate calibration.57 If there is reason to suspect that the array calibration has drifted and needs to be repeated, its accuracy can be checked with the following test. Calibration accuracy rotational substitu tion: The accuracy of the calibration factors can be tested by performing two measurem ents with setups that differ by a 180 array rotation. This locates eac h detector in the same field location as its mirror detector about the arrays center. Since the array ca libration is intended to provide measurement invariance with respect to detector, there should be minimal difference between the two measurements. This is assessed by taki ng the ratio between the two measurements using () 180 180 0 0_,pos posnmeas cfaccuracynmeasnmeas nmeas = (5-7) where cf_accuracy is the accuracy of the ca libration factor for axis position pos and nmeas is the normalized measurement for array orientation 0 or 180. The tolerance level for this test of the calibration accuracy is dependent on the tolerance level established for the calibra tion reproducibility assuming there are no systematic influences that resu lt from an array rotation. To determine the accepted level of error, the tolerance from the reproducibility test should be doubled since the error is a result of two detectors. This is not stochasti c error, it is the ma ximum expected error, and therefore should not be added via quadrature. This test is conducted during an array calib ration and if there is a specific reason to believe the calibration is invalid (an increase in the failure rate of patient specific QA or a deviation from a baseline measurement ). Due to the natur e of the calibration 113

PAGE 114

procedure,59 small shifts in the off-axis X-ray spectrum may introduce a symmetric error to the calibration factors. This rotational test cannot detect this or other types of symmetric calibration errors; a direct comparison to a water scan is needed. Calibration accuracy water scan: The accu racy of the calibration factors can be checked against a water scan using () ,_,narray na r r a yw a t e r poswaternmeas cfaccuracynmeasnmeas nmeas =, (5-8) where cf_accuracy is the calibration factor accuracy for detector n, axis position pos; nmeaswater and nmeasarray are the normalized measurem ents acquired with the water tank and array, respectively. The array measurements can be checked against annual QA measurements using a sim ilar geometry as the water t ank. However, if the LINAC beam varied significantly for any reason (a major component failu re) since the water scans, those water scans should not be used in comparison to the array measurements. Once again the tolerance is dependent on the type of test that is being conducted by the array and is user defined. This test is only recommended during the initial use (or acceptance) of the array. Afterward, the reproducibility and the rotational substitution tests will be sufficient to main tain calibration accuracy. Results and Discussion The QA program described in Materials and Methods was applie d to the panel. The results and experiences are described below. Physical During the course of one year, the panel s top surface was cleaned twice due to buildup of grime. Cable damage due to improper handling is seen in Fig. 5-1 The cable 114

PAGE 115

handling recommendations (discussed in the pr evious section) should be followed to avoid these types of damage. Firmware and Software Measurement consistency: An example of the panels measurement reproducibility on the Elekta Synergy is shown in Fig. 5-2 The methodology of Eq. 5-1 was carried out for five measurement sets. Ther e were no software updates during these measurements; these results purely represent the magnitude of measurement variation that occurs with the panel on the Elekta Synergy over a short time period. Two interesting effects are present in se t 1. The first is a sine shaped symmetry variation that is likely the result of micro-variations in the electron beams spot size and location. The second is the pre-irradiation ef fect on the panels ioni zation chambers; this results in increased noise throughout the me asurement, specifical ly for the center ionization chamber for this particular device. Electronics Visual examination of background: In Fig. 5-3 A we seen a screen shot from the panels software (pd-axis) of a background measurement with a spike in one of the detectors responses. A contaminant had ent ered the chamber and was bridging the gap between the anode and the cathode, causi ng a high leakage rate. The panel was serviced by the manufacturer, resolv ing the singular detector response. In Fig. 5-3 B we see a comparison of backgrounds that were taken seven months apart on a panel that was operating correctly, the black line represents the older measurement and the red li ne represents the newer m easurement. While their magnitudes are different, the over all shape is similar, which is a good indication that the panel is working correctly. 115

PAGE 116

Over time, the panels sensitivity to background gradually changed with accumulated dose, Fig. 5-3 C. These values represent the mean bias value (leakage rate) from the background measurement that is taken when the panels software is turned on at the beginning of a measurement se ssion. This analysis was not conducted as a formal study, which accounts for the use of days instead of accumulated dose and the somewhat inconsistent conditions of measurement. The inconsistent measurement conditions specifically refers to the time when power was applied to the panel before the detectors background bias was m easured, this can affect the bias amplitude. While it is not a formal study, it does give some indi cation that the background sensitivity changes with accumulated dose. Whether this occurs for all arrays is unknown; but it should be considered when one visually compares background measurements to previous versions. Test of background normality: Upon initially receiving the panel, five background measurements were made, each approximat ely 23 s in duration and containing 180 updates (125 ms collection interval). For each measurement, the UC values (see Eq. 52 ) were determined for each of the 251 indivi dual ionization chambers. On these, the GOF tests were used to evaluate the norma lity of UCs PDF. The percentage of detectors that rejected H0 are shown in Table 5-2 under the row heading of Leaky. The normality of these UCs is poor, as can be seen in the percentage of detectors that rejected H0; all were in the upper 90s. The panel was discovered to have a contaminant (the same as discussed in visual examinat ion of the background) on several of the electrodes that were bridging the gap between the anode and t he cathode, thus introducing a bias into the measurements. 116

PAGE 117

The panel was returned to manufacture r for service. After which, the measurements and GOF tests were repeated. The PDF of UC fo r the (-) 12.5 cm x-axis detector (X8) is shown in Fig. 5-4 and appears normal. The percentage of detectors that rejected H0 are shown in Table 5-2 under the row heading of Post. The percentage of detectors rejecting H0 dropped to 23 and 15% for the Chisquared test with significance values of 0.1 and 0.05, respectively. Using the Anderson-Darling test, no detectors rejected H0. The Anderson-Darling GOF test is bette r suited to assess the normality of a PDF than the Chi-squared test due to a higher power score.62 Based on these results, the panels behavior was deemed acceptable for use. After a year of heavy use, the measurem ents and GOF tests were repeated, Table 5-2 under the row heading of Post 1 year. T he results show that despite a change in the global bias value (see Fig. 5-3 ), the behavior of the background measurements remained normal. The two GOF tests were tested against 251 randomly generated normal distributions (generated in MATLAB), si mulating background measurements for each of the panels 251 ionization chambers. The percentage of simulated detectors that rejected H0 are shown in Table 5-2 under the row heading of RND. These values illustrate the ability of each test to a ssess the normality of a PDF. Once again the Anderson-Darling test proved to be a better test for assessing normality. Dose linearity: The dose linearity of the panel was determined relative to the dose linearity of a LSI, in this case a Farmertype chamber and electrom eter (model FC65-G, IBA, Schwarzenbruck, DE; model K602, CNMC Company, Nashville, TN USA). In previous work to characterize the IC PROFILER,53 a dose non-linearity was identified 117

PAGE 118

and corrected with new firmware. Using this unc orrected version of the panel, the dose linearity test identified a maximum panel devi ation from the LSI of 1.1%. While this value is less than the monitor chamber linea rity tolerance [() 2%], and therefore could theoretically pass, there was a cause for concern and it was inve stigated further. Using the corrected firmware, the maximum deviation of the panels dose linearity relative to a LSI decreased from 1.1% to 0. 48%. Once again using the tolerance set in TG-142 as a guideline, this value left 1.52% error available to the machine QA test before the tolerance was crossed or a machine with a dose-nonlinearity was not identified. Measurement reproducibility: T he short term reproducibility of the panel relative to the LSI resulted in a maximum percent er ror of 0.08% and a mean percent error of 0.04%. These values are consistent with previously determined values and small enough to be a non-factor.53 Setting a tolerance value is dependent on the quantity being measured. Measuring output with the recommended toler ances of TG-142 [3%, 2%, and () 1% (absolute), for daily, monthly, and annual QA, respectively] w ould require a panel reproducibility that is less than those tolerances. The short term p anel reproducibility of 0.08% is well below these values. The long term panel reproducibilit y established by Simon et al. was 0.84% using a cobalt source that was not normalized to a LSI. Instead a se t amount of time to integrate the signal was used as the standard; the timer standard deviation was insignificant. Using 0.84% gives an error play of 2.2% and 1.2% fo r measuring daily and monthly output before the output is falsely failed or a failure is not identified. 118

PAGE 119

Array Calibration Reproducibility: The primary factor affect ing WF calibration reproducibility is variation in the delivered beam symmetry.59 A beam that has a higher degree of variation may only be capable of producing a rray calibrations with error approaching 1 to 2%.45, 59 Simon et al. described st eps to reduce this effect ; on a machine with minor symmetry variations [() 0.15%] the panels ca libration error was reduced from () 1.6% to () 0.48% (short term reproducibility). Simon et al. established the WF calibrati on reproducibility (for the panel) as () 0.5%.53, 59 If the panel is used as an absolute comparison to water scans, then this reproducibility would allow a combined error of 0.5% due to other sources before the tolerance level of 1% (from a baseline value) is crossed.17 This level of calibration error seems acceptable not only for profile m easurements, but also for other common detector array uses (IMRT QA). Calibration accuracy rotati onal substitution: Using the calibration reproducibility of () 0.5% established above, the panels tolerance for calibration factor accuracy (determined by rotational subs titution) was set at () 1%. The cf_accuracy results for the panel are seen in Fig. 5-5 The symmetric error that occu rs on the yand pd-axes is a calibration error less than the tolerance; the maximum deviation from unity was 0.87%. Since each position is a combination of the mi rror detectors, the calibration error is half of the cf_accuracy value or 0.44%. Using the recommended tolerance of 1% fr om TG-142, this leaves a combined error of 0.56% before that tolerance level is reached. Whether this is an acceptable level of error depends on the institution. 119

PAGE 120

Calibration accuracy water scan: Profile measurements (10 g/cm2) with the panel and a scanning water tank [Blue P hantom and CC13 ionization chamber (IBA, Schwarzenbruck, DE)] are seen in Fig. 5-6 A. The calibration factor accuracy values are shown in Fig. 5-6 B. The maximum disagreement between the panel and the CC13 for the 1.5 and 10 g/cm2 measurements was (+) 1.3% and (-) 0.71%, respectively. The agreement for the 10 g/cm2 measurements is relatively uniform, with some noise involved; there is a symmetric concave error associated with the 1.5 g/cm2 measurement. This is the type of error that the rotational substitution test cannot detect. Contrary to the 1.5 g/cm2 measurement, the 10 g/cm2 results are relatively uniform and possibly even slightly convex in shape. This indicates that the calibration factors or the actual detectors have a slight off-axis dependence relative to the CC13 response at these depths. A likely cause of this are t he spectral changes that occur in an X-ray beam as the off-axis distance is increased. This is highlighted by a slight energy dependence of the panels detectors to low energy scatter and/or contaminant electrons.53 Using the tolerance level of 1%, from TG-142, the agr eement values for the 10 g/cm2 comply while the 1.5 g/cm2 do not. This is taking the stance of one to one comparison with a water tank. However, if we take the ro le of a baseline comparison, then the primary purpose of this test was to explore the po ssibility of a large symmetric calibration error, which is not pres ent for the reasons already stated. Conclusion We have developed a QA program that is general to detector arrays. It focuses on four test areas: physical, software and firmw are, electronics, and array calibration. 120

PAGE 121

Specific tests were described with guidelines for determining the tolerance values and frequencies were also recommended. Firm tolerance values were not estab lished; instead guideli nes and discussion were given for physicists to set their own values. While specific tolerances would be beneficial, the complication associated with t heir determination would be extraordinary and not universally applicable. Logically, the acceptable error must be less than the test parameter being measured; ultimately, the physicist is responsible for making an informed and responsible decision in setting the tolerance levels. The QA program was applied to the IC PROFILER. The test of background normality effectively identified a backgr ound bias in the electronics that the manufacturer repaired. This QA program is basic and can be expanded on by the community. Ideally, this work will increase interest in this topic to the point of official recommendations from the medical physics community and the vendors. 121

PAGE 122

Table 5-1. Quality assurance program for the IC PROFILER; E, before every use of the detector array; I, upon initial use of the array or following service; B, biannual; A, annual; AN, as needed Test Area Test Frequency Action Physical Panels top surface and buildup E Clean Cables E Replace Firmware / software Check for updates B Before and after measurements AN 0.5% File integrity AN 0% Electronics Background Visual E Visual difference in shape and/or magnitude Goodness of fit A Failure of H0 Beam measurements Dose response A 1% Reproducibility A 1% Array calibration Reproducibility A () 0.5% Rotational substitution A () 1% Water I () 1% Table 5-2. Percent of detectors rejecting H0 (= the distribution of UC is normal) using the chi-squared and AndersonDarling goodness of fit tests. The hypothesis was tested for two significance levels (0.1 and 0.05). Four data sets were evaluated: measurements of the panel with leaky c hannels, post repair by one day, post repair by one year; a random normal distribution (RND). Chi-square test Anderson-Darling test Measurements = 0.1 = 0.05 = 0.1 = 0.05 Leaky (%) 98 1.3 97 2.0 97 1.6 96 2.8 Post (%) 23 3.2 15 2.4 0 0 0 0 Post 1 year (%) 14 2.7 8.1 2.0 0 0 0 0 RND Test (%) 5.0 0.60 2.2 0.73 0 0 0 0 122

PAGE 123

A B Figure 5-1. Cable damage as a result of (A) disconnecting the cable by pulling the cable instead of the connector and (B ) repetitive circular movement. Figure 5-2. The measurement reproducibility for the panel on the Elekta Synergy. The methodology of Eq. 5-1 was used for five meas urement sets; two unique measurements were in each set. 123

PAGE 124

A B C Figure 5-3. (A) An example of two backgr ound measurements that were separated by seven months and have a similar shape and amplitude. (B) An example of an anomaly in the measurement background. (C) The quantitative change in bias (leakage rate) over one year. Figure 5-4. The PDF of UC for the (-) 12.5 cm x-axis detector (X8). 124

PAGE 125

125 Figure 5-5. The accuracy of the panels ca libration factors were determined using rotational substitution. A B Figure 5-6. (A) Profile m easurements with the panel and a scanning water tank. (B) Calibration factor accuracy results, obt ained by taking the ratio of the panel to the CC13 profile measurements.

PAGE 126

CHAPTER 6 SUMMARY AND FUTURE WORK Recently, the quality of radiation therapy (s pecifically instances of overdoses) have been brought to the attention of the American public.6 Several instances of human error that led to a lethal dose bei ng delivered to patients were illustrated in the New York Times. At the heart of these incidents is an inherent lack of time in this profession to complete all aspects of QA and quality control. The purpose of this work was to illustrate the increase in efficiency of LINAC QA and radiation data collection that can be accomplished with the use of detector arrays. Two time consuming measurement processes were used as a showcase: MLC calibration and water tank scans. Specific Aim 1: Multi-leaf collimator calibration is traditionally done using time consuming equipment and/or techniques. An MLC calibration method (RDRL) was created that was capable of quantitative and efficient measurement of the MLC. The method is based on measuring each leaf posit ion with a detector in a detector array (PROFILER 2) that was prev iously aligned to the LI NACs radiation coordinate system. The PROFILER 2 was chosen due to the high spatial resolution of its diode detectors. Multi-leaf collimator calibrati on using the RDRL method takes on average 30 minutes to complete, which is a marked impr ovement over traditional methods that can take several hours and are not as quantitative. Specific Aim 2: Water scans, by definit ion, are acquired using a scanning water tank. This detector system, while time test ed, is accompanied by inefficiencies and unfamiliarity amongst clinical medical physicist s. Detector arrays in contrast, are much more familiar to the clinical physicist due to regular use and are inherently more efficient. Accordingly, an ionization cham ber detector array was characterized in a 126

PAGE 127

radiation environment to assess its abilities as a system for acquiring beam profiles that are water tank equivalent. The detector ar ray chosen was the IC PROFILER, which uses ionization chambers for its detector s. The IC PROFILER was found to be a suitable system for acquiring beam profiles with an average error relative to a water tank acquired profile of () 0.8%; this error was positively offset for certain measurement conditions. The benefit of using the IC PROFILER over a scanning water tank to acquire beam profiles is the time savings. The required time to acquire a series of profile measurements was 180 mi nutes for a scanning water tank versus 30 minutes for the IC PROFIELR. The time savings increa ses with the number of measured profiles. The next step in this research is to begin evaluating the ability of the IC PROFIELR to perform current tasks that are assigned to scanning water tanks. Specifically, the abilities of the IC PROFILER to perform monthly and annual QA are going to be investigated. The ability of t he IC PROFILER as a commissioning device will also be evaluated. While the data density is too low for direct measurements to be used for TPS commissioning, various compensa tory methods will be investigated. One promising method involves fitting previous ly measured golden beam data to a small subset of measurements that were made with the IC PROFIL ER. This is possible due to the dosimetric similarity of modern LINACs, amongst a particular make and model.63, 64 Specific Aim 3: During the course of characterizing the IC PROFILER, it was noticed that the WF array ca libration produced unstable calibration factors when the calibration was acquired on a beam with minor symmetry instabilities. A machine with clinically insignificant symmetry variations [() 0.1%] can readily cause calibration errors 127

PAGE 128

of () 2%. In contrast, a machine with symme try variations on the order of () 0.05% produced WF calibrations that varied by less than () 0.5%. This occurs because the WF calibration procedure requires an invariant beam during each calibration measurement. The stability of WF calibration s was increased by combining the use of a continuously on LINAC beam during portions of the calibrati on procedure with the addition of side-scatter to the IC PROFIELR. A cont inually on LINAC beam has a reduced amount of symmetry variation when compared to a beam that is cycled on and off. The reproducibility of the WF calibration increased from () 1.6% to () 0.5% on a LINAC with symmetry variations of () 0.15%. Further work needs to be done on the WF calibration to fully automate the calibration procedure, increasing its reproduc ibility. Additionally, small changes that occur in the spectral energy of the beam as the off-axis distance is increased should be investigated for their impact on calibration accuracy, particu larly with different amounts of buildup. Specific Aim 4: With the increased use of detector arrays comes an increase in their importance in the clinic and a need to QA their performance. A QA program was developed that was general to detector arrays and included suggestions for establishing tolerances and frequencies for testing. Four QA regions were designated: physical, software and firmware, electronic, and array ca libration. The QA program was applied to the IC PROFILER, identifying a leaky chamber that occurred due to a contaminant in the ionization chamber. Due to the acknowledged lack of time to perform all of the recommended QA tests for LINACs (and other modalities),15 the entirety of this wo rk needs to be continued. 128

PAGE 129

129 Ideally, more efficient measurement systems and methods should be developed to perform every test that is recommended in TG-40 and TG-142. De tector arrays have shown an adaptability and proficiency for st reamlining the QA and radiation data collection processes. It is the aim of th is candidate to continue research in this endeavor.

PAGE 130

LIST OF REFERENCES 1 S. Henderson and P. Biggs, "The rate of ev olution of radiation therapy planning and delivery systems is exceeding the evolution rate of quality assurance processes," Med Phys 26, 1439-1441 (1999). 2 T. R. Mackie, T. Holmes, S. Swerdloff, P. Reckwerdt, J. O. Deasy, J. Yang, B. Paliwal and T. Kinsella, "Tomotherapy: a new concep t for the delivery of dynamic conformal radiotherapy," Med Phys 20, 1709-1719 (1993). 3 J. F. Dempsey, B. Dionne, J. Fitzsimmons, A. Haghigat, J. G. Li, D. Low, S. Mutic, J. R. Palta, H. E. Romeijn and G. Sjoden, "A Real-Time MRI Guided External Beam Radiotherapy Delivery System," Med Phys 33, 2254 (2006). 4 P. L. Petti, C. F. Chuang, V. Smith and D. A. Larson, "Peripheral doses in CyberKnife radiosurgery," Med Phys 33, 1770-1779 (2006). 5 G. J. Kutcher, L. Coia, M. Gillin, W. F. Hans on, S. Leibel, R. J. Mort on, J. R. Palta, J. A. Purdy, L. E. Reinstein, G. K. Svensson and et al., "Comprehens ive QA for radiation oncology: report of AAPM Radiation T herapy Committee Task Group 40," Med Phys 21, 581-618 (1994). 6 W. Bogdanich, "Radiation Offers New Cures, and Ways to Do Harm," in The New York Times, (New York, 2010), pp. A1. 7 C. Burman, C. S. Chui, G. Kutcher, S. Lei bel, M. Zelefsky, T. LoSasso, S. Spirou, Q. Wu, J. Yang, J. Stein, R. M ohan, Z. Fuks and C. C. Ling, "P lanning, delivery, and quality assurance of intensity-modul ated radiotherapy using dyna mic multileaf collimator: a strategy for large-scale implementation for t he treatment of carcinoma of the prostate," Int J Radiat Oncol Biol Phys 39, 863-873 (1997). 8 X. Wang, S. Spirou, T. LoS asso, J. Stein, C. S. C hui and B. Mohan, "Dosimetric verification of intensity-modulated fields," Med Phys 23, 317-327 (1996). 9 V. M. Systems, "http://www.varian.com/us/oncology/treat ments/treatment_techniques/rapidarc/ ," (2008). 10 S. N. Corporation, http://www.sunnuclear.com/documents/ProC atB06252010.pdf ," (2008). 11 A. Van Esch, C. Clermont, M. Devillers, M. Iori and D. P. Huyskens, "On-line quality assurance of rotational radiotherapy treatment delivery by means of a 2D ion chamber array and the Octavius phantom," Med Phys 34, 3825-3837 (2007). 12 I. B. A., http://www.iba-dosimetry .com/MULTICube.1362.0.html ," (2008). 130

PAGE 131

13 S. N. Corporation, http://www.sunnuclear.com /documents/arccheck.pdf ," (2009). 14 ScandiDos, "http://www.scandidos.com/Documents/Brochures.aspx ," Delta 4 (2009). 15 J. R. Palta, presented at the QA & Do simetry Symposium, Lake Buena Vista, FL, 2009 (unpublished). 16 S. Huq, "The Philosophy of TG100: What It Is and What It Is Not," Med Phys 35, 2932 (2008). 17 E. E. Klein, J. Hanley, J. Bayouth, F. F. Yin, W. Simon, S. Dresser, C. Serago, F. Aguirre, L. Ma, B. Arjomandy, C. Liu, C. Sandin and T. Ho lmes, "Task Group 142 report: quality assurance of medica l accelerators," Med Phys 36, 4197-4212 (2009). 18 B. D. Lynch, J. Kozelka, M. K. Ranade, J. G. Li, W. E. Simon and J. F. Dempsey, "Important considerations for radiochromic film dos imetry with flatbed CCD scanners and EBT GAFCHROMIC film," Med Phys 33 4551-4556 (2006). 19 O. H. Suleiman, P. C. Bunch, P. F. Butler, J. E. Gray, A. G. Haus, M. H. P. McKetty, E. L. Nickoloff, M. V. Pina, R. J. Pizzutiello, S. J. Shepar, D. C. Spelic, K. J. Strauss, S. R. Szeglin and R. A. Uzenoff, Report No. 94, 2006. 20 A. Niroomand-Rad, C. R. Blackw ell, B. M. Coursey, K. P. Gall, J. M. Galvin, W. L. McLaughlin, A. S. Meigooni, R. Nath, J. E. Rodgers and C. G. Soares, "Radiochromic film dosimetry: recommendations of AAPM Radiation Therapy Committee Task Group 55. American Association of Ph ysicists in Medicine," Med Phys 25 2093-2115 (1998). 21 S. Pai, I. J. Das, J. F. De mpsey, K. L. Lam, T. J. Losasso, A. J. Olch, J. R. Palta, L. E. Reinstein, D. Ritt and E. E. Wilcox, "TG-69: radiographic film for megavoltage beam dosimetry," Med Phys 34, 2228-2258 (2007). 22 B. E. Nelms and J. A. Sim on, "A survey on planar IMRT QA analysis," J Appl Clin Med Phys 8 2448 (2007). 23 D. D. Leavitt and L. Larsson, "Evaluation of a diode detector array for measurement of dynamic wedge dose distributions," Med Phys 20, 381-382 (1993). 24 H. H. Liu, E. P. Lief and E. C. Mc Cullough, "Measuring dose distributions for enhanced dynamic wedges using a mult ichamber detector array," Med Phys 24, 15151519 (1997). 25 T. C. Zhu, L. Ding, C. R. Liu, J. R. Palta, W. E. Simon and J. Shi, "Performance evaluation of a diode array for enhanced dynamic wedge dosimetry," Med Phys 24, 1173-1180 (1997). 131

PAGE 132

26 S. N. Rustgi and D. M. Frye, "Dosimetric characterization of r adiosurgical beams with a diamond detector," Med Phys 22, 2117-2121 (1995). 27 G. Yan, C. Fox, C. Liu and J. G. Li, "The extraction of true profiles for TPS commissioning and its impact on IMRT patient-specific QA," Med Phys 35, 3661-3670 (2008). 28 I. J. Das, C. W. Cheng, R. J. Watts, A. Ahnesjo, J. Gibbons X. A. Li, J. Lowenstein, R. K. Mitra, W. E. Simon and T. C. Zhu, "Accelerator beam data commissioning equipment and procedures: repor t of the TG-106 of the Therapy Physics Committee of the AAPM," Med Phys 35, 4186-4215 (2008). 29 G. Mu, E. Ludlum and P. Xia, "Impact of MLC leaf position errors on simple and complex IMRT plans for head and neck cancer," Phys Med Biol 53 77-88 (2008). 30 ICRU, Report No. 24, 1976. 31 A. Boyer, P. Biggs, J. M. Galvin, E. Kl ein, T. LoSasso, D. Low, K. Mah and C. Yu, Report No. 72, 2001. 32 M. Pasquino, V. C. Borca, P. Catuzzo, F. Ozzello and S. To fani, "Transmission, penumbra and leaf positional accuracy in commissioning and quality assurance program of a multileaf collimator for step-and-shoot IMRT treatments," Tumori 92, 511516 (2006). 33 L. Parent, J. Seco, P. M. Evans, D. R. Dance and A. Fielding, "Evaluation of two methods of predicting MLC leaf positi ons using EPID measur ements," Med Phys 33, 3174-3182 (2006). 34 M. Mamalui-Hunter, H. Li and D. A. Low, "MLC qualit y assurance using EPID: a fitting technique with subpixel precision," Med Phys 35, 2347-2355 (2008). 35 M. Mohammadi and E. Bezak, "Evaluation of MLC leaf positioning using a scanning liquid ionization chamber EPID," Phys Med Biol 52, N21-33 (2007). 36 M. C. Lopes, A. Chaves and M. Capela, "A dosimetric calibration method for a double-focused multileaf collimator," Med Phys 34, 3473-3474 (2007). 37 T. J. Jordan and P. C. Williams, "The des ign and performance characteristics of a multileaf collimator," Phys Med Biol 39, 231-251 (1994). 38 J. R. Palta, D. K. Yeung and V. Frouhar, "D osimetric consideratio ns for a multileaf collimator system," Med Phys 23, 1219-1224 (1996). 39 I. E. Commission, Report No. IEC 1217, 1996. 132

PAGE 133

40 J. F. Dempsey, H. E. Romeij n, J. G. Li, D. A. Low and J. R. Palta, "A fourier analysis of the dose grid resolution required for accurate IMRT fluence map optimization," Med Phys 32, 380-388 (2005). 41 H. E. Johns and J. R. Cunningham, The physics of radiology 4th ed. (Charles C. Thomas, Springfield, Ill., U.S.A., 1983). 42 P. R. Almond, P. J. Biggs, B. M. Coursey, W. F. Hanson, M. S. Huq, R. Nath and D. W. Rogers, "AAPM's TG-51 protocol for clin ical reference dosimetry of high-energy photon and electron beams," Med Phys 26, 1847-1870 (1999). 43 W. E. Simon, J. Shi and C. A. I annello, U.S.A. Patent No. 6125335 (2000). 44 R. J. Watts, "Evaluation of a diode detector array for use as a linear accelerator QC device," Med Phys 25 247-250 (1998). 45 D. Letourneau, M. Gulam, D. Yan, M. Oldham and J. W. Wong, "Evaluation of a 2D diode array for IMRT quality assurance," Radiother Oncol 70, 199-206 (2004). 46 E. Yorke and AAPM Radiation Ther apy Committee. Task Group No. 62., Diode in vivo dosimetry for patients receiving exte rnal beam radiation t herapy: report of Task Group No. 62, Radiation Therapy Committee, AAPM. (Published by Medical Physics Publishing for the American Association of Physicists in Medicine, College Park, MD, 2005). 47 G. F. Knoll, Radiation detection and measurement 3rd ed. (Wiley, New York, 2000). 48 K. R. Kase, B. E. Bjrngard and F. H. Attix, The Dosimetry of ionizing radiation (Academic Press, Orlando, 1985). 49 L. Parent, A. L. Fielding, D. R. Dance, J. Seco and P. M. Evans, "Amorphous silicon EPID calibration for dosimetric applications: comparison of a method based on Monte Carlo prediction of response with ex isting techniques," Phys Med Biol 52, 3351-3368 (2007). 50 G. Yan, B. Lu, J. Kozelka, C. Liu and J. G. Li, "Calibration of a novel fourdimensional diode array," Med Phys 37, 108-115 (2010). 51 S. Amerio, A. Boriano, F. Bourhaleb, R. Cir io, M. Donetti, A. Fidanzio, E. Garelli, S. Giordanengo, E. Madon, F. Marche tto, U. Nastasi, C. Peroni, A. Piermattei, C. J. Sanz Freire, A. Sardo and E. Trevisiol, "Dosimet ric characterization of a large area pixelsegmented ionization chamber," Med Phys 31, 414-420 (2004). 52 R. Sadagopan, J. A. Bencomo, R. L. Martin G. Nilsson, T. Matz en and P. A. Balter, "Characterization and clinical evaluation of a novel IMRT quality assurance system," J Appl Clin Med Phys 10 104-119 (2009). 133

PAGE 134

53 T. A. Simon, W. E. Simon, D. Kahler, J. G. Li and C. Liu, "Characterization of a multiaxis ion chamber array," Med Phys Under Review (2010). 54 J. P. McCaffrey, B. Downton, H. Shen, D. Niven and M. McEwen, "Pre-irradiation effects on ionization chambers used in radiation therapy," Phys Med Biol 50 N121-133 (2005). 55 B. Poppe, A. Blechschmidt, A. Djouguela, R. Kollhoff, A. Rubach, K. C. Willborn and D. Harder, "Two-dimensional ionization chamber arrays for IMRT pl an verification," Med Phys 33, 1005-1015 (2006). 56 M. Stasi, S. Giordanengo, R. Cirio, A. Boriano, F. Bourhaleb, I. Cornelius, M. Donetti, E. Garelli, I. Gomola, F. Marchetto, M. Po rzio, C. J. Sanz Freire, A. Sardo and C. Peroni, "D-IMRT verification with a 2D pixel ionization chamber: dosimetric and clinical results in head and neck cancer," Phys Med Biol 50, 4681-4694 (2005). 57 P. A. Jursinic and B. E. Nelms, "A 2-D diode array and analysis software for verification of intensity modulated r adiation therapy de livery," Med Phys 30, 870-879 (2003). 58 E. Spezi, A. L. Angelini, F. Romani and A. Ferri, "Characterization of a 2D ion chamber array for the verification of radiotherapy treatments," Phys Med Biol 50, 33613373 (2005). 59 T. A. Simon, W. E. Simon, D. Kahler, J. G. Li and C. Liu, "Wide field array calibration dependence on the stability of measur ed dose distributions," Med Phys Under Review (2010). 60 J. L. Devore, Probability and statistics for engineering and the sciences 7th ed. (Thomson/Brooks/Cole, Belmont, CA, 2008). 61 T. W. Anderson and D. A. Darling, "A Test of Goodness of Fit," Journal of the American Statistical Association 49, 765-769 (1954). 62 M. A. Stephens, "EDF Statistics for G oodness of Fit and Some Comparisons," Journal of the American Statistical Association 69, 730-737 (1974). 63 D. A. Jaffray, J. J. Battist a, A. Fenster and P. Munro, "X -ray sources of medical linear accelerators: focal and extra-focal radiation," Med Phys 20, 1417-1427 (1993). 64 R. J. Watts, "Comparative m easurements on a series of accelerators by the same vendor," Med Phys 26 2581-2585 (1999). 134

PAGE 135

135 BIOGRAPHICAL SKETCH Thomas Allan Simon was born in Euclid Ohio. In 1985, his family moved to Satellite Beach, Florida. He graduated from Satellite Hi gh School in June of 1997 and then enrolled at the Florida State University. After one y ear he transferred to the University of Florida to pursue a degree in microbiology, graduating with a B.S. in 2002. In 2002 he enrolled in the medical physics program of the Depar tment of Nuclear and Radiological Engineering at t he University of Florida. His initial research involved anthropomorphic phantom constr uction with Drs. David Hintenlang and Wesley Bolch. In 2005 he passed the qualifying exam and began working for his doctoral advisor Dr. Chihray Liu. In 2006, he became involved in a small busin ess innovative research grant that was tasked with collecting golden beam data sets (also serving as benchmark data) for each of the three primary linear accelerato r (LINAC) manufacturers (Elekta, Siemens, and Varian). During this time he was educated in and first began to appreciate metrology. Upon graduating, he plans on moving to Me lbourne, FL to work at Sun Nuclear Corporation on a calibration l aboratory. His current research interests lie in array calibration methodology and investigating t he boundaries of replacing water tanks with detector arrays.