<%BANNER%>

Evaluating and Improving Patient-Specific QA for IMRT Delivery

Permanent Link: http://ufdc.ufl.edu/UFE0024310/00001

Material Information

Title: Evaluating and Improving Patient-Specific QA for IMRT Delivery
Physical Description: 1 online resource (155 p.)
Language: english
Creator: Yan, Guanghua
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2009

Subjects

Subjects / Keywords: arccheck, imrt, qa, vmat
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: Modern radiation therapy techniques such as intensity-modulated radiation therapy (IMRT) and newly-emerging volumetric modulated arc therapy (VMAT) aim to deliver highly conformal radiation dose to the target volume while sparing nearby critical organs as much as possible with the complex motion of multi-leaf collimator (MLC) leaves. Pre-treatment patient specific quality assurance (QA) has become an essential part of IMRT in making sure the delivered dose distributions agree with the planned ones. This dissertation evaluates the performance of current patient-specific QA process and proposes solutions to improve its sensitivity, accuracy and efficiency. In step and shoot IMRT, the study on the sensitivity of patient-specific QA to minor MLC errors reveals tighter criterion such as 2%/2mm must be employed to detect systematic MLC positioning errors of 2 mm. However, such criterion results in low average passing rate which leads to excessive false alarms, mainly due to inadequate treatment planning system (TPS) beam modeling on beam penumbra. An analytical deconvolution approach is proposed to recover true photon beam profiles to obtain a true beam model which significantly improves agreement between calculated and measured dose distributions. Thus a tighter criterion could be employed to enhance the sensitivity of patient-specific QA to minor errors in the delivery system. Measurement based patient-specific IMRT QA is a time-consuming process. A fast and accurate independent planar dose calculation algorithm is proposed to replace measurement based QA. The algorithm analytically models photons coming out from the accelerator and computes dose distribution from first principles. Accuracy of the algorithm is validated against 2D diode array measurements. The algorithm is found to be fast and accurate enough to replace time consuming measurement based QA. Patient-specific QA for VMAT differs significantly from step and shoot IMRT due to the increased use of dynamic components (dynamic gantry and collimator, dynamic MLC and variable dose rate) in VMAT. A novel four dimensional (4D) diode array is developed by Sun Nuclear Corp for patient-specific VMAT QA. This work develops effective calibration procedures for this novel device by accounting for diode sensitivity and angular response dependence. A real time algorithm to derive gantry angle is developed to interpolate corresponding angular correction factors. Clinical applications of the diode array are demonstrated with IMRT as well as VMAT plans. Excellent agreement ( > 95% passing rate with 1%/2mm criterion) is achieved between diode array measurement and TPS calculation. The 4D diode array is proved to be a valuable tool for both IMRT and VMAT patient specific QA.
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 Guanghua Yan.
Thesis: Thesis (Ph.D.)--University of Florida, 2009.
Local: Adviser: Liu, Chihray.
Local: Co-adviser: Li, Jonathan.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2010-05-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2009
System ID: UFE0024310:00001

Permanent Link: http://ufdc.ufl.edu/UFE0024310/00001

Material Information

Title: Evaluating and Improving Patient-Specific QA for IMRT Delivery
Physical Description: 1 online resource (155 p.)
Language: english
Creator: Yan, Guanghua
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2009

Subjects

Subjects / Keywords: arccheck, imrt, qa, vmat
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: Modern radiation therapy techniques such as intensity-modulated radiation therapy (IMRT) and newly-emerging volumetric modulated arc therapy (VMAT) aim to deliver highly conformal radiation dose to the target volume while sparing nearby critical organs as much as possible with the complex motion of multi-leaf collimator (MLC) leaves. Pre-treatment patient specific quality assurance (QA) has become an essential part of IMRT in making sure the delivered dose distributions agree with the planned ones. This dissertation evaluates the performance of current patient-specific QA process and proposes solutions to improve its sensitivity, accuracy and efficiency. In step and shoot IMRT, the study on the sensitivity of patient-specific QA to minor MLC errors reveals tighter criterion such as 2%/2mm must be employed to detect systematic MLC positioning errors of 2 mm. However, such criterion results in low average passing rate which leads to excessive false alarms, mainly due to inadequate treatment planning system (TPS) beam modeling on beam penumbra. An analytical deconvolution approach is proposed to recover true photon beam profiles to obtain a true beam model which significantly improves agreement between calculated and measured dose distributions. Thus a tighter criterion could be employed to enhance the sensitivity of patient-specific QA to minor errors in the delivery system. Measurement based patient-specific IMRT QA is a time-consuming process. A fast and accurate independent planar dose calculation algorithm is proposed to replace measurement based QA. The algorithm analytically models photons coming out from the accelerator and computes dose distribution from first principles. Accuracy of the algorithm is validated against 2D diode array measurements. The algorithm is found to be fast and accurate enough to replace time consuming measurement based QA. Patient-specific QA for VMAT differs significantly from step and shoot IMRT due to the increased use of dynamic components (dynamic gantry and collimator, dynamic MLC and variable dose rate) in VMAT. A novel four dimensional (4D) diode array is developed by Sun Nuclear Corp for patient-specific VMAT QA. This work develops effective calibration procedures for this novel device by accounting for diode sensitivity and angular response dependence. A real time algorithm to derive gantry angle is developed to interpolate corresponding angular correction factors. Clinical applications of the diode array are demonstrated with IMRT as well as VMAT plans. Excellent agreement ( > 95% passing rate with 1%/2mm criterion) is achieved between diode array measurement and TPS calculation. The 4D diode array is proved to be a valuable tool for both IMRT and VMAT patient specific QA.
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 Guanghua Yan.
Thesis: Thesis (Ph.D.)--University of Florida, 2009.
Local: Adviser: Liu, Chihray.
Local: Co-adviser: Li, Jonathan.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2010-05-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2009
System ID: UFE0024310:00001


This item has the following downloads:


Full Text

PAGE 1

EVALUATING AND IMPROVING PATIENTSPECIFIC QA FOR IMRT DELIVERY By GUANGHUA YAN A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2009 1

PAGE 2

2009 Guanghua Yan 2

PAGE 3

To my beloved wife, Xiuyao and daughters, Maggie and Sarah 3

PAGE 4

ACKNOWLEDGMENTS I would like to gratefu lly and sincerely thank my major advisor, Dr. Jonathan Li, for his guidance. His insights and suggestions are always inspiring. I really apprec iate his willingness to offer prompt help at any time. His diligence, pe rseverance and serious at titude on both research and clinic set a good example for me. The person who always advises me, supports me and helps me is my other advisor, Dr. Chihray Liu. Dr. Liu introduced me into the exciting medical physics field and generously provided financial support throughout my entire gra duate study in medical physics. His unmatched enthusiasm inspired me to devote my career to medical physics research and practice. I really thank him for his continuous encourag ement and support. My thanks extend to my committee members, Dr. Jatinder R. Palta and Dr. William M. Mendenhall for their input and many valuable suggestions. I thank Dr. Darren L. Kahler and Dr. Sanjiv S. Samant for their comments, suggestions and help. I need to express my special gratitude to Mr. Bo Lu. He initiated the planar dose calculation algorithm based on which we receive d funding from Sun Nuclear Corporation which made my graduate study financia lly possible. I enjoyed numerous in spiring discussions with him. I wish to thank all post-doc and fellow gradua te students working in the Radiation Oncology student lab, Christopher Fox, Thomas A. Simon, Lee-Cheng Peng and Bart Lynch for friendship, collaboration and numerous suggestions. I would like to sincerely thank Mr. William Si mon from Sun Nuclear Corporation for his generous support and continuous encouragement. Mr. Jakub Kozelka and Dr. Jie Shi from Sun Nuclear Corporation provide me a lot of techni que help which I am really indebted to. Most of all, I thank my wife and my daughters, my parents and my sisters for their great unconditional support, love and encouragement. 4

PAGE 5

TABLE OF CONTENTS page ACKNOWLEDGMENTS ............................................................................................................... 4 LIST OF TABLES ...........................................................................................................................7 LIST OF FIGURES .........................................................................................................................8 ABSTRACT ...................................................................................................................... .............10 CHAPTER 1 INTRODUCTION ................................................................................................................ ..12 General Introduction .......................................................................................................... .....12 Patient-specific IMRT QA ......................................................................................................15 Study Aims .............................................................................................................................17 2 ON THE SENSITIVITY OF PATIENT-SPECIFIC IMRT QA TO MLC POSITIONING ERRORS ......................................................................................................23 Introduction .................................................................................................................. ...........23 Materials and Methods ...........................................................................................................25 Patient Plan Selection ......................................................................................................25 Leaf Positioning Errors Simulation .................................................................................26 Dose Distribution Measurement ......................................................................................26 Planar Dose Comparison .................................................................................................28 Criteria for Identifying Errors .........................................................................................28 Results .....................................................................................................................................29 Discussion .................................................................................................................... ...........30 Conclusions .............................................................................................................................33 3 THE EXTRACTION OF TRUE PH OTON BEAM PROFILE FOR TPS COMMISSIONING ................................................................................................................3 9 Introduction .................................................................................................................. ...........39 Methods and Materials ...........................................................................................................42 Beam Data Measurement ................................................................................................42 Beam Profile Deconvolution ...........................................................................................42 Detector response function .......................................................................................43 Profile deconvolution ...............................................................................................44 TPS Commissioning ........................................................................................................46 Patient Specific IMRT QA Comparison ..........................................................................47 Results .....................................................................................................................................48 Discussion .................................................................................................................... ...........51 Conclusions .............................................................................................................................54 5

PAGE 6

4 INDEPENDENT PLANAR DOSE CALCULATION FOR IMRT QA ................................63 Introduction .................................................................................................................. ...........63 Methods and Materials ...........................................................................................................64 Head Scatter Source Models ...........................................................................................64 Planar Dose Calculation Algorithm .................................................................................67 Comparison for Head Scatter Calculation .......................................................................72 Comparison for IMRT Dose Calculation ........................................................................73 Results .....................................................................................................................................74 Comparison of Head Scatter Factors ...............................................................................74 Comparison of Dose Calculation .....................................................................................75 Computation Time ...........................................................................................................76 Discussion .................................................................................................................... ...........76 Conclusions .............................................................................................................................80 5 THE CALIBRATION OF A NOVEL 4D DIODE ARRAY ..................................................89 Introduction .................................................................................................................. ...........89 Methods and Materials ...........................................................................................................91 Description of the 4-D Diode Array ................................................................................91 Diode Array Calibration ..................................................................................................92 Diode sensitivity correction factors ..........................................................................93 Diode directional respon se correction factors ..........................................................94 Independence of Calibration Factors on Field Sizes .......................................................97 Gantry Angle Derivation Algorithm ...............................................................................97 Diode Array Calibration Validation ................................................................................99 Results ...................................................................................................................................100 Discussion .................................................................................................................... .........102 Conclusions ...........................................................................................................................105 6 CLINICAL APPLICATION OF A NOVEL 4D DIODE ARRAY FOR IMRT AND VMAT QA ............................................................................................................................117 Introduction .................................................................................................................. .........117 Materials and Methods .........................................................................................................118 Detector characterization ...............................................................................................118 Clinical Application .......................................................................................................120 Sensitivity to errors ........................................................................................................1 22 Results ...................................................................................................................................123 Discussion .................................................................................................................... .........126 Conclusions ...........................................................................................................................129 7 CONCLUSIONS ................................................................................................................. .141 LIST OF REFERENCES .............................................................................................................146 BIOGRAPHICAL SKETCH .......................................................................................................155 6

PAGE 7

LIST OF TABLES Table page 2-1 P-values from two-tail ed Wilcoxon rank-sum test. ...........................................................35 3-1 Best-fit shape parameter k of the detector re sponse function. .........................................55 3-2 Average passing rate and standard deviation of the IMRT QA results .............................55 4-1 The comparison of average passing rates. .........................................................................82 5-1 Pseudo algorithm to derive instant beam angle. ..............................................................107 5-2 Beam incident angle derivation for four Step and shoot IMRT beams. ..........................108 6-1 Effect of using averaged di rectional correction factors. ..................................................131 7

PAGE 8

LIST OF FIGURES Figure page 1-1 Multi-leaf Collimator and a patient undergoing IMRT. ....................................................21 1-2 Flow chart of the dissertation. ............................................................................................22 2-1 Average passing rates of Gafchromic EBT film measurement with DTA. .......................35 2-2 Average passing rates of Gafchromic EBT film measurement with Gamma. ...................36 2-3 Average passing rates of MapC HECK measurement with DTA. ......................................36 2-4 Average passing rates of MapC HECK measurement with Gamma. .................................37 2-5 Drop of average passing rates with film measurement. .....................................................37 2-6 Drop of average passing rates with MapCHECK measurement. .......................................38 2-7 Average passing rates when comparing error free plans to plans with errors. ..................38 3-1 Comparison of cross beam profiles m easured with different dosimeters. .........................56 3-2 Diode measured beam profiles and the co rresponding TPS calculate d beam profiles. .....57 3-3 Convolution to obtain kernel parameters. ..........................................................................58 3-4 Deconvolution results for sm all and medium field sizes. ..................................................59 3-5 Deconvolution results for large field size. .........................................................................60 3-6 Deconvolution results for asymmetric MLC field. ............................................................60 3-7 Comparison of the passing rates. .......................................................................................61 3-8 Example of calculated pl anar dose distributions. ..............................................................62 4-1 Flowchart of the proposed IMRT pl anar dose calculation algorithm. ...............................82 4-2 Comparison of the calculated fluence. ...............................................................................83 4-3 Comparison between the measured a nd calculated head scatter factors. ...........................84 4-4 Comparison of head scatter factors with modified three source model. ............................85 4-5 Comparison of head scatter factors with modified single source model. ..........................86 4-6 Comparison between the measured a nd calculated cross beam profiles. ..........................87 8

PAGE 9

4-7 Comparison of MapCHECK-measured and calculated planar dose distributions. ............87 4-8 Example of a calculated planar dose distribution. .............................................................88 4-9 Source intensity distribution. .............................................................................................88 5-1 Prototype of a 4D diode array. .........................................................................................109 5-2 Diagram of a p-n junction diode. .....................................................................................109 5-3 Definition of relative gantry angle and normalized measured beam profile. ..................110 5-4 Field size dependence of diode response. ........................................................................111 5-5 Algorithm for deriving instant beam incident direction. .................................................112 5-6 Convert dose measured in water to dose in acrylic. .........................................................112 5-7 Relative diode response as a function of field size with normal incidence. ....................113 5-8 Relative diode response as a function of field size with oblique incidence. ....................113 5-9 Relative diode sensitivity of all diodes on the diode array. .............................................114 5-10 Directional response as a func tion of beam incident angle. .............................................115 5-11 Comparison between Pinnacle calculat ion and diode array measurement. .....................115 5-12 Detector response isotropicity. .........................................................................................116 6-1 Electron pulses and setup to change dose ra te by varying source to detector distance. ..132 6-2 Linear detector response as a f unction of delivered monitor units. .................................133 6-3 Pulse rate and dose rate dependence of diode response. ..................................................134 6-4 Dose profile calculated by Pinnacle and measured by diode rings. .................................136 6-5 Dose profile for conformal arc plan. ................................................................................138 6-6 Average passing rate change of three IMRT beams with gantry rotation errors. ............139 6-7 Average passing rate change of three IMRT beams with MLC errors. ...........................140 9

PAGE 10

Abstract of Dissertation Pres ented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy EVALUATING AND IMPROVING PATIENTSPECIFIC QA FOR IMRT DELIVERY By Guanghua Yan May 2009 Chair: Chihray Liu Cochair: Jonathan G. Li Major: Nuclear Engineering Sciences Modern radiation therapy techniques such as intensity-modulated ra diation therapy (IMRT) and newly-emerging volumetric modul ated arc therapy (VMAT) aim to deliver highly conformal radiation dose to the target volum e while sparing nearby critical organs as much as possible with the complex motion of multi-leaf collimator (MLC) leaves. Pre-treatment patient specific quality assurance (QA) has become an essential part of IMRT in making su re the delivered dose distributions agree with the planned ones. This dissertation evaluates the performance of current patient-specific QA process and proposes solutio ns to improve its sensitivity, accuracy and efficiency. In step and shoot IMRT, the study on the sens itivity of patient-specific QA to minor MLC errors reveals tighter criterion such as 2%/2mm must be employed to detect systematic MLC positioning errors of 2 mm. However, such crite rion results in low average passing rate which leads to excessive false alarms, mainly due to inadequate treatment planning system (TPS) beam modeling on beam penumbra. An analytical deco nvolution approach is proposed to recover true photon beam profiles to obtain a true beam model which significantly improves agreement 10

PAGE 11

11 between calculated and measured dose distributions. Thus a tighter criterion could be employed to enhance the sensitivity of patient-specific QA to minor errors in the delivery system. Measurement based patient-specific IMRT QA is a time-consuming process. A fast and accurate independent planar dose calculation algorithm is proposed to replace measurement based QA. The algorithm analytic ally models photons coming out from the accelerator and computes dose distribution from first principles. Accuracy of the algorithm is validated against 2D diode array measurements. The algorithm is found to be fast and accurate enough to replace time consuming measurement based QA. Patient-specific QA for VMAT differs significantly from step and shoot IMRT due to the increased use of dynamic components (dynami c gantry and collimator, dynamic MLC and variable dose rate) in VMAT. A novel four dime nsional (4D) diode array is developed by Sun Nuclear Corp for patient-specific VMAT QA. This work develops effective calibration procedures for this novel device by accounting for diode sensitivity and angular response dependence. A real time algorithm to derive gantry angle is deve loped to interpolate corresponding angular correction factors. Clin ical applications of the diode array are demonstrated with IMRT as well as VMAT plans. Excellent agreement (>95% passing rate with 1%/2mm criterion) is achieved between diode array measurement and TPS calculation. The 4D diode array is proved to be a valuable tool for both IMRT and VMAT patient specific QA.

PAGE 12

CHAPTER 1 INTRODUCTION General Introduction Radiation therapy uses ionizing radiation as part of cancer tr eatment to control malignant cells. It destroys or kills cancer cells in the targeted area ( tumor) by damaging their genetic materials and preventing them from growing and dividing. Al though radiation can destroy both cancer cells and normal cells, most normal cells will recover from radiation damage due to regrowth process. It is crucial to deliver adequate dose to the tumor ce lls to achieve cancer killing while avoiding surrounding healthy tissue to minimize side effects. Historically, maximum dose prescribed to th e tumor is limited by the sensitivity and limits of nearby organ at risk (OAR) and thus the e ffectiveness of radiation therapy is reduced. A number of recent refinements and technologies have dramatically increased the effectiveness of external radiation therapy by sh aping radiation to the target Three-dimensional conformal radiation therapy (3D-CRT) and intensity modul ated radiation therapy (IMRT) are among those new technologies. Three dimensional CRT utilizes modern computed tomography (CT) techniques to acquire a 3D image of the patient which allows the physician to delineate the tumor and critical organs precisely. Ra diation beams are optimized to conform to the targets with jaws and multi-leaf collimators (MLC). Sufficiently high radiation dose could be delivered to the tumor while surrounding healthy tis sues are largely sp ared. The effectiveness of 3D-CRT is significantly improved compared to traditional two dimensional (2D) treatments where only one or a few axial slices of the patient anatomy were acquired for planning. Three dimensional CRT is a dramatic change from traditional practice in that it utilizes tumor and critical structures delineated on multiple transverse CT images, radi ation fields designed in beams eye view and volumetric dose calculation and evaluation tools.1 12

PAGE 13

IMRT is a further extension of 3D-CRT in that non-uniform fluences generated by the superposition of multiple MLC segments or dyna mic MLC movements are used to improve dose conformity and OAR sparing. The modulated flue nces are obtained from inverse planning and computer optimization with the guidance of clinic al goals and objectives. It inherits all the new tools and methodologies from 3D-CRT. Previous radiation therapy tec hniques including 3DCRT do not provide a method for sparing critical struct ures that either push into the target or are tightly surrounded by targets.2 IMRT does allow for highly c onformal dose distribution with sharp dose gradient for complex target volumes with concave surfaces. This is accomplished by the complex motion of MLC leaves (Figure 1-1 a) equipped on a medical linear accelerator (LINAC).3 This advantage has been utilized to escal ate dose to the tumor. As a consequence, each IMRT field has many small, irregular and as ymmetric subfields defined by MLC leaves and are characterized by high dose gradient. In IMRT delivery, the mechanical component s that can be varied are gantry positions, collimator positions, MLC configurations, couch positions and dose rates. IMRT delivery techniques based on MLC can be typically classified into two categories: fixed gantry IMRT and dynamic gantry IMRT.2 If the gantry remains stationary wh en radiation beam is on, the delivery is referred to as fixed gantry IMRT. Otherwise, it is referred to as dyna mic gantry IMRT. Fixed gantry IMRT includes sliding window techniques and segmental IMRT. Sliding window techniques allow MLC motion when beam is tu rned on while segmental IMRT turns off the beam when MLC is in motion (so called step and shoot delivery). In both sliding window techniques and segmental IMRT, gantry angles are usually selected manually with the guidance of beams-eye-view for optimal target cove rage and OAR sparing (Figure 1-1 b). Both techniques in the fixed gantry IMRT category have gained wide application in clinical practice 13

PAGE 14

for more than a decade. In the category of dynamic gantry IMRT, a special type called volumetric-modulated arc therapy (VMAT) is rapidl y proliferating recently in both academic and community practice due to its shorter treatment time and less monitor units than fixed gantry IMRT.4 It utilizes all the dynamic components (dynamic gantry and collimator, dynamic MLC and variable dose rate) to achieve these advantages. IMRT represents one of the most exciting t echnology advancements in radiation therapy since the introduction of CT imagi ng in radiation treatment planning.5 It is a new treatment paradigm to deliver high doses of radiation to tumor while providing maximum sparing for nearby OAR. The obvious dosimetric advantage and increased reimbursement have fueled the speed of clinical dissemination of IMRT.6 A recent survey shows IMRT treatment was used by 73% of the responding radiation oncologi sts while the number was 32% in 2002.7 It has been estimated that 30%~60% of the cancer patients in the United States are currently being treated with IMRT.8 Improved local control and reduced complications as compared with 3D-CRT has been observed for all common anatomical sites with the largest number of applications for prostate and head and neck treatment.7 However, the advancement in IMRT delivery doesnt come without a risk. The clinical efficacy of IMRT relies on the ability of the planning system and the delivery system to accurately deliver planned dose to the target. The proximity of critical organ to tumor leads to high dose gradient which puts st ringent requirements on modeli ng radiation beam penumbra as well as radiation dose outside beam opening. The complicated motion of MLC leaves to modulate beam segments makes leaf positioni ng accuracy more crit ical than 3D-CRT.3 Patient delivery quality assurance (QA) has become an integral part of IMRT treatment process. Increased effort has to be made to understa nd IMRT planning and de livery process and its 14

PAGE 15

associated QA procedures compared with 3D-CRT.3 The AAPM guidance document on IMRT points out that IMRT QA consists of three task s: commissioning and testing of the treatment planning and delivery systems, routine QA of the delivery system, and patie nt-specific validation of treatment plans1. The first task concerns with the inte grity of the planning system in modeling the delivery system. The second task deals with the mechanical and dosimetric accuracy of the delivery system (gantry and collimator rotati on, MLC positioning, etc). The third task is to ensure safe and accurate treatment of a patien t. In general, IMRT is a complex technique including patient simulation, treatment planni ng, leaf sequencing, plan transfer, patient positioning, online verification and treatment deli very. Patient specific IMRT QA, as a safe guard and total system check, play s an essential role in making su re IMRT delivery is carried out as prescribed and planned. Patient-specific IMRT QA The clinical efficacy of IMRT treatment depends on the ability of the delivery system to faithfully deliver the planned dose distributi on to the desired location. IMRT process as described above has multiple steps with potential errors arising from each step. Patient-specific IMRT QA, as a total system check, provides a unique opportunity to identify these potential sources of errors and plays an essential role in ensuring the safe and accurate delivery of IMRT. Typical patient-specific IMRT QA involves th e measurement of a point dose or 2D dose distributions in a homogeneous phantom and compar ed with the treatment planning system (TPS) calculation. The whole delivery sequ ence of an IMRT treatment is transferred onto the CT image of the phantom and calculated with TPS. At th e same time, the entire delivery sequence is delivered on the phantom and a planar dose at a cer tain depth is measured with some dosimeters. The measured and calculated planar dose distri butions are compared and analyzed with dose difference, distance to agreement9 or gamma index10, etc. The percentage of points, area or 15

PAGE 16

volume passing a pre-selected criterion is used to indicate the quality of the whole planning and delivery procedure. IMRT planning and delivery systems are cha nging rapidly and there have been no well established standards on measuring devices, criter ia or acceptance levels for patient-specific IMRT QA. Different dosimeters have been empl oyed or developed for IMRT QA measurement. Ion chamber combined with film is the early popular choice. An ion chamber could be placed in a high dose and low dose gradient region for abso lute point dose measurement. Film can be irradiated to measure a relative dose distribution. An absolute planar dose map can be obtained by combining ion chamber and film measurement. Accuracy of the measurement depends on the selection of the measuring point. This is a time consuming process. Film is gradually replaced by online 2D detectors such as diode arrays and ion chamber arrays. Absolute planar dose distribution could be obtained dur ing a single delivery which make s measurement more accurate and efficient. The downside of these online 2D detectors is their limited spatial resolution compared with film. Portal dosimetry based on electronic portal imagi ng device (EPID) is another alternative for IMRT QA measurement. It has excellent resolution with immediate readout. However, designed as an imaging device EPID requires a lot of correction factors to achieve accurate dosimetry which ma y limit its wide application in IMRT QA. Just like there are a lot of different choices of dosimeters for QA measurement, there is no consensus on what criteria one should use in evaluating the agr eement between measured and calculated dose distribution. A recent survey shows the majority responding clinical institutions use 3%/3mm criterion in their practice.11 At the same time, stricter criterion such as 2%/2mm and looser criterion such as 5%/5mm are both used at di fferent institutions. With the same comparison criterion (for example, 3%/3mm), there have be en very different achievable agreements (or 16

PAGE 17

action levels) reported in literature. The survey conducted by Nelms et al shows majority of the community employing 3%/3mm criterion for IMRT QA analysis can achieve 90%-95% passing rates between measurem ent and calculation.11 Basran et al reported that greater than 95% agreement could be achieved for non-head and neck IMRT with same criterion, while the number for head and neck was only 88%.12 The AAPM summer school proceedings13 provide general guidance on IMRT QA, tolerance limits a nd action levels. However, it doesnt seem practical to build standards across all institutions with all these di versities in the IMRT practice. It is recommended that each facility offeri ng IMRT should develop its own guidelines and criteria for the acceptance and QA of IMRT planning and delivery.3 This will inevitably raise the question of how one should evalua te the performance of their own IMRT QA process in making sure the delivered dose distribution agrees with the prescribed and planned dose distribution. In 2008, Radiological Physics Center (RPC) re ported that 30% of 250 head and neck IMRT irradiations of an anthropomorphic phant om performed by academic institutions applying for credentialing failed to agree with their own treatment plan to within 7% difference and 4 mm DTA accuracy criteria14. In other words, the irradiation of an anthropomorphic phantom demonstrated failings in their own QA proce dures usually performed well on homogeneous virtual water phantom with a tighter criterion (such as 3%/3mm). These institutions were those who had confidence in their dosimetry accuracy and expected to pass the credentialing tests. The RPC credentialing test results clearly show the current IMRT QA process in these institutions fail to achieve their goal, i.e., identif ying sources of errors in the system. Study Aims This dissertation consists of two parts. The first part eval uates the ability of current IMRT QA practice for fixed gantry IMRT in catching minor errors in the delivery system and proposes solutions to improve its sensitivity to errors and its efficiency. The second part focuses on 17

PAGE 18

calibrating a novel four-dimensional dosimeter for dynamic gantry IMRT (VMAT) patient specific QA. Specifically, the first part of this dissertation focuses on these aspects of patient specific QA for fixed gantry IMRT: (1) Sensitivity to errors in delivery system. Patient specific IMRT QA is expected to catch both gross errors and minor errors in the planning and delivery system. Minor MLC positioning errors are of particular interest since MLC positi oning reproducibility and accuracy are critical to the success delivery of IMRT. It is not uncommon to observe these errors15 and they typically lead to large dosimetric impact16. Can the current IMRT QA process effectively catch these errors when they are present in the system? Our study shows that with either film or online 2D diode arrays the QA process fails to identify systematical MLC errors of 2 mm using popular 3%/3mm criterion.17 (2) Accuracy. Our results show that patient-s pecific IMRT QA with a stricter criterion (such as 2%/2mm) exhibits stronger sensitiv ity to minor errors in delivery system.17 However, not all institutions can achieve acceptable passing rate with such strict criterion which leads to large number of false alarms. One of the main reasons causing large discrepancy between calculation and measurement is the inadequate beam modeling by using beam profiles measured with finite-sized ion chambers whic h suffer from volume averaging effect.18 In this dissertation, we develop algorithms to extract true photon beam profiles by removing volume averaging effect to improve treatment planning system (TPS) beam modeling. The improved TPS beam model significantly improves dose calculation accuracy. The improved agreement between calculated and measured dose distributions en ables the use of stricter criteri on in patient specific QA which enhances its sensitivity to minor errors in the delivery system. 18

PAGE 19

(3) Efficiency. IMRT QA with offline detect ors (film, gel dosimeter) is extremely timeconsuming. With fast online electronic dosimeters (diode or IC arrays), it still takes 30 min on average for each patient QA. It has been pointed out that patient-specifi c verification could be reduced once confidence in dosimetric accuracy has been accumulated19. In this dissertation, we develop a fast and accurate independent planar dose calculation algorithm to replace measurement based patient specific IMRT QA.20 The independent planar dose calculation is validated against diode array measurement. Wh en performing patient-specific QA, TPS dose calculation is compared with the fast independe nt dose calculation instead of time-consuming delivery and measurement after clinic hours. The second part of this dissertation deals wi th the calibration and testing of a novel device for patient-specific QA for the newly emerging VMAT. Patient-specific QA for VMAT differs significantly from IMRT due to its increased use of dynamic components (dynamic gantry and collimator, dynamic MLC and variable dose rate).21, 22 Ideally, patient-specific QA for VMAT should be performed in arc mode with an isotr opic 3D dosimeter to check dosimetric impact of the interplay of these dynamic components. At pr esent, there has been no such isotropic 3D dosimeter available in the market. Sun Nuclear Co rp developed the first isotropic 3D dosimeter (ArcCHECKTM) for patient-specific QA for VMAT. Howe ver, effective calibration procedure, dosimetric characterization and clinical application of this device are yet to be established. In this dissertation, we develop calibration procedures for this novel devi ce and demonstrate its clinical application for both IMRT and VMAT patient-specific QA. In summary, this dissertation is organized with five specific aims (Figure 1-2): Specific aim 1 : Investigate the sensitivity of current patient-specific IMRT QA to MLC leaf positioning errors (Chapter 2). 19

PAGE 20

Specific aim 2 : Develop deconvolution algorithms to recover true beam profiles to improve Pinnacle beam modeling and dose calculati on accuracy. Tighter criteria could thus be employed in IMRT QA for enhan ced sensitivity (Chapter 3). Specific aim 3 : Develop independent planar dose calculation algorithms to replace measurement based IMRT QA to impr ove its efficiency (Chapter 4). Specific aim 4 : Develop effective calibration proce dure for ArcCHECK for its use in VMAT QA (Chapter 5). Specific aim 5 : Study ArcCHECK dosimetric characteri stics and validate its clinical application for both IMRT and VMAT QA (Chapter 6). 20

PAGE 21

Figure 1-1. (a) Multi-leaf Collimator designed by Varian Medical Systems. Typical MLC leaf has a width of 1 cm and each leaf can be moved independently to modulate the beam. (b) A patient undergoing IMRT. The LINAC can rotate around the patient to shoot the beam from the best angles. 21

PAGE 22

22 Figure 1-2. Flow chart of the dissertation. The pu rpose of this work is to propose solutions for existing issues with patient specific IMRT QA. Major work towards improving QA for fixed gantry IMRT includes the evaluati on of its sensitivity (SA 1), development of deconvolution algorithm to extract true photon beam profiles for TPS commissioning (SA 2) and an independent planar dose calculation (SA 3) aiming to replace measurement based IMRT QA. For VMAT, a novel 4D diode array is characterized and calibrated (SA 4). Algorith ms for its clinical application in both IMRT and VMAT QA are developed (SA 5).

PAGE 23

CHAPTER 2 ON THE SENSITIVITY OF PATIENT-SPECIFIC IMRT QA TO MLC POSITIONING ERRORS Introduction Intensity-modulated radiati on therapy (IMRT) has become the treatment technique of choice for many types of cancers receiving radiat ion therapy. The clinical efficacy of IMRT relies on dose escalation to the tumor while avoiding toxicity to the surrounding critical structures. Accurate multi-leaf collimator (MLC ) leaf positioning plays a crucial role in the effective implementation of MLC-based IMRT13. Tolerance limits for leaf position accuracy and reproducibility have been suggested for IMRT wh ich are more stringent than for conventional radiation therapy23. Several authors have studied the dosimet ric effect of leaf positioning errors 16, 24-27. Luo et al studied the correlation between leaf position errors and dosim etric impact in prostate cancer treatment24. They found a linear correlati on between the target dose error and the average MLC position error, with 1% target dose change aris ing from 0.2 mm systematic leaf position errors. LoSasso et al also reported that a 0.2 mm gap variation leads to 1% dose variation with an average gap width of 2 cm with dynamic beam delivery28. Mu et al studied the dosimetric effect of leaf position errors on head and neck patients by deliberatel y introducing random (uniformly sampled from 0 mm, mm and mm) and systematic (.5 mm or mm) leaf positioning errors into the plan26. They found no significant dosimetric effect (<2% dose change to both target and critical organs) introduced by ra ndom leaf position errors up to 2 mm, while significant effects (8% change in D95% and ~12% in D0.1cc to critical organs ) were observed by 1 mm systematic leaf position errors in complex IMRT plans. Zygmanski et al studied the dosimetric effect of truncated Gaussian (with 0. 1 cm standard deviation) shaped random leaf position errors27. They found that although the average com posite dose to the ta rget of a nine 23

PAGE 24

field IMRT plan was changed only by 3%, fluen ce change resulted from each single field was commonly > 10%. Woo et al found that leaf position uncertainty could lead to dose variations of up to 13% when positioning the ion chamber on the field edge16. All these studies emphasized the importance of the MLC positioni ng accuracy and reproducibility. Several authors have reported excellent accuracy of MLC leaf position by analyzing MLC log files for both dynamic MLC and static MLC24, 27, 28. For dynamic MLC, Zygmanski et al reported <0.05 cm leaf position error27, while LoSasso et al found that the average leaf gap error was much smaller than 0.02 cm28. For static MLC, Luo et al reported average leaf position errors of ~0.05 cm based on the analysis of MLC log files24. On the other hand, by using a fast video-based electronic portal imaging device, Zeidan et al observed a maximum unplanned leaf movement of 3 mm during static MLC delivery15. The same groups study, based on MLC log file analysis, reported that in approximately 80% of the total segment deliveries, at least one collimator leaf had unplanned movement of at least 1 mm (projected at isocenter) during segment delivery29. Significant dosimetric impact could arise from these MLC position errors which suggests that periodic MLC quality assura nce (QA) should be performed to ensure the accuracy and reproducibility of MLC leaf positions. It is both challenging and time-consuming to check the position accuracy of every single MLC leaf pair at all possible off axis positions30-33. In practice, dedicated MLC QA is conducted bi-weekly, or even less frequently, whereas pati ent-specific IMRT QA is usually done for every new patient before the start of IMRT treatment. Th e aim of this work is to assess the sensitivity of patient-specific IMRT QA to leaf position er rors. A common method of patient-specific QA is to re-calculate the treatment plan in a QA phantom with all beams at 0o gantry angle (IEC convention) and normal to the phantom surface. Th e planar dose distribution is measured under 24

PAGE 25

the same geometry using film or two-dimensiona l (2D) detector arrays The agreement between the calculated and measured planar dose distribu tions is then quantified using parameters like the Gamma index10, percent dose difference (%Diff) or dist ance-to-agreement (DTA), etc. Patientspecific IMRT QA is required for every new IM RT patient to ensure the accuracy of the treatment plan. Childress et al have investigated the feasibilit y of using a 2D dosimetric system to detect gross errors such as beam ener gy change, wrong patients beam data, one beam collimator angle change of 90o, gantry angle change of 10o and omitting the delivery of one beam34. However, the sensitivity of commonly used patient-specific IMRT QA systems to subtle MLC position errors has not been fully examined. Sastre-Padro et al studied the consequences of subtle leaf positioning errors on IMRT delivery but whether the errors could be detected by patient-specific IMRT QA systems was not answered35. In this work, random MLC position errors up to 2 mm as well as systematic errors ( mm, mm) were introduced into treatment plans for 8 H&N IMRT patients (tot aling 53 fields). Planar dose distributions calculated before and after introducing errors were compared to dose distributions measured with both film and 2D diode arrays using gamma index as well as DTA cr iterion on a field-by-fie ld basis. The change in the passing rate distribution was used as an indicator to study the sensitivity of the IMRT QA procedure to MLC position errors. Materials and Methods Patient Plan Selection Eight head and neck IMRT plans (total of 53 fields) previously used for patient treatment with 6 MV photon beams of a linear accelerator (Trilogy, Varian Medical Systems, Palo Alto, CA) with 120-leaf Millennium MLC were randomly selected for this study. The step-and-shoot treatment plans were generated with a commer cial treatment planning system (TPS, Pinnacle3, version 8.0d, Philips Medical Systems, Madison, WI) with direct machine parameter 25

PAGE 26

optimization (DMPO) option which directly op timizes the shape and weight of each MLC segment. Minimum segment area and minimum segment MU were set to 4 cm2 and 3 MU, respectively. The adaptive convol ution dose calculation algorithm with inhomogeneity correction was used for all the plans. Leaf Positioning Errors Simulation Random and systematic leaf positioning errors were simulated in the same manner as in Mu et al.26 The original treatment plans were ex ported from Pinnacle using the Pinnacle scripting language. Leaf position errors were intro duced into the plans by directly modifying the leaf positions in the exported treatment plans. Random errors were uniformly sampled from [0 mm, mm, mm] for each leaf from both leaf banks. Negative and positive errors make the leaf end-to-end distance smaller a nd larger, respectively. When leaf collisions happen, the leaves were assume closed and they were assigned the same position. When introducing systematic errors, the position of each open leaf was cha nged by the same amount ( mm and 2 mm). A total of 6 plans were generated for each patient: the original plan without leaf positioning error, the plan with random errors and four plans wi th different amount of systematic errors. The modified treatment plans were imported back in to Pinnacle for planar dose calculation. Planar dose distribution was calculated for each field at 10 cm depth and 90 cm source-to-surface distance (SSD) in a rectangular water phantom with 0o gantry angle (IEC convention) and normal to the phantom surface. All the calculation was done using a 1.0 x 1.0 mm2 dose grid. Dose Distribution Measurement In standard practice for patient-specific IMRT QA, both film and 2D detector arrays are widely used. In this work, both radiochromic films and a 2D diode array were employed to measure the planar dose distribut ion under the same geometry as in calculation. Radiochromic films (Gafchromic EBT film, ISP Corp., Wayne NJ) were sandwiched between solid water 26

PAGE 27

pieces at 10 cm depth with a source-to-film di stance of 100 cm. Small marks were placed on the edges of the films for geometri c registration. After ir radiation, the films were left for 24 hours before analysis as recommended by Niroomand-Rad et al36 to minimize post irradiation coloration effects. A commercial flatbed scanner (EPSON 1680) was used to digitize the film at a resolution of 100 m/pixel and was sampled at a spatial resolution of 1 mm at both directions using commercial data analysis software (Matla b v7.0, Mathworks, Inc.). An adaptive 2D wiener filter of 4x 4 pixel region was used to reduce the inherent noise. Film images were acquired from the red CCD channel only. Sensitometric curves were obtained by deliver ing varying amounts of radiation which covered th e range of the doses of interest. When irradi ating the radiochromic films, all doses were rescaled such that the maximum dose was around 100 cGy to avoid the uncertainty at low doses37. Convenient 2D detector arrays are replacing film for patient-specific IMRT QA. A 2D diode matrix (MapCHECK 1175, Sun Nuclear Corp., Melbourne, FL) was also used to measure the planar dose distribution unde r the same geometry as the film measurement. The MapCHECK device has been demonstrated to be an excellent tool for the verification of IMRT fields with little energy or dose rate dependence38, 39. An intercomparison between film and MapCHECK showed that MapCHECK can effectively replace film dosimetry in routine IMRT QA, even though it has a limited spatial resolution40. The MapCHECK device wa s calibrated for absolute dose measurement. The original clinical treatm ent plans without MLC positioning errors were delivered for measurement. Each of the 53 IMRT fields was meas ured with the diode array and 10 of the fields were measured with the radiochromic films. 27

PAGE 28

Planar Dose Comparison Planar dose distributions without and with the leaf position errors were calculated for each of the 53 fields. The calculated planar dose di stributions were imported into the MapCHECK analysis software38 and compared with either MapCHECK measurement or film measurement. When calculation was compared with film meas urement, film measurement was loaded into MapCHECK software and used as reference. Ea ch of the MapCHECK (o r film) measured points above 10% of the maximum dose level was compared with calculation usi ng absolute distanceto-agreement (DTA) comparison as well as index in the absolute dose comparison mode9, 10. Two sets of criteria, 2%/2mm and 3%/3mm, were employed, and the percentage of the points passing the acceptance criteria was evaluated. Criteria for Identifying Errors The decision regarding whether the procedur e could identify the MLC errors was made based on the comparison between passing rate dist ributions before and after the errors were introduced. In this work, the change in average passing rates and a significance test were used to make the decision. A sudden drop of the average pa ssing rate from the baseline value serves as a first indication of errors in the system. In th is work, the threshold of average passing rate decrease was arbitrarily set at 5%. A non-parametric significance test, Wilcoxon rank-sum test41, was employed to compare the distribution of th e passing rates before and after the introduction of the MLC position errors. The Wilcoxon rank-sum test was chosen over the popular student ttest was because the passing rates didnt fo llow the normal distribution and the standard deviation varied significantly. The p-value of th e test gives the probabi lity the two compared variables are from the same distribution with eq ual median. The significance level of the test was set to 0.01. In other words, the difference was sign ificant only when the pvalue of the test was less than 0.01. In summary, the in troduced MLC position errors woul d be considered identified 28

PAGE 29

when (1) the drop in average passing rates was la rger than 5% and (2) the p-value from the Wilcoxon rank-sum test was smaller than 0.01. Results The average passing rates and the standard devi ations using both films and the diode array are displayed in Figure 2-1 to 2-2 and Figur e 2-3 to 2-4, respectively, for different MLC positioning errors, and the drop of average pa ssing rate caused by MLC positioning errors is shown in Figure 2-5 to 2-6. The p-value from the Wilcoxon rank-sum test is summarized in Table 2-1. For film measurements, the average passing rate with DTA, 3%/3mm criterion was 93%% (1 standard deviation) without MLC errors. With random leaf position errors of up to 2 mm and systematic errors on the order of 1 mm the average passing rate dropped by less than 5% and was still around 90%. For 2 mm systematic errors, an asymmetric behavior was observed with 5% drop in average passing rate for positive errors but ~15% drop for negative errors. The p-values from the rank-sum test indicated significant difference only for negative 2 mm systematic error, with a p-valu e of 1.10E-03. In this case, we conclude that only the -2 mm systematic MLC positioning error could be identified by patient-s pecific IMRT QA procedure when using EBT film and DTA, 3%/3mm. When using DTA, 2%/2mm criterion in film result analysis, both the -1 mm and mm systematic errors would cause the average passing rates to drop by more than 5%. But rank-sum test showed that only -2 mm erro r caused significant difference in passing rate distribution with a p-value of 6.22E-04. For the other cases, there is significant over lap between the distributions. Based on our criteria for identifying errors, only the -2 mm errors could be identified. The same can be concluded when employing test for film result analysis. 29

PAGE 30

Figure 2-3 to 2-4 show the results when us ing MapCHECK as the detector. The drop of average passing rates is displayed in Figure 6 and the p-values from Wilcoxon rank-sum test are summarized in Table 2-1. The p-values indicated that both mm errors and mm errors caused significant difference in the passing rate distributions for both 2%/2mm and 3%/3mm criterion, using either DTA comparison or analysis. However, Figure 2-6 shows that larger than 5% drop in average passing rate was observed only for mm error. Thus we conclude that with MapCHECK, the IMRT QA procedure could identif y systematic errors on the order of 2 mm. Discussion Due to the complexity of planning and delivery of IMRT, patient-specific QA is recommended for every new patient before the st art of IMRT delivery. This could potentially detect any gross errors such as the wrong beam energy, wrong patient plan, or any data transfer errors from the TPS to the delivery system. Ho wever, the ability of patient-specific QA in detecting subtle errors, such as TPS calculation error or MLC positioning error, is more difficult to quantify. In practice, after commissioning of the IMRT pl anning and delivery system, baseline values for patient-specific QA acceptance criteria, which might be site-dependent, could be established over a period of time. A sudden drop in passing rates would signify deviations in one or more of the IMRT planning and delivery com ponents and would warrant further investigation. Therefore, whether the commonly-used QA method of field-by-field planar dose comparison is sensitive to MLC positioning errors on the order that would signifi cantly affect the treatment is of practical interest. Both radiochromic film and MapCHECK de vice were used in this study. When no MLC positioning errors were introduced, the passing ra te distribution from MapCHECK device had a less variation than radiochromic film: 1% vs. 5% with the 3%/3mm criterion and 3% vs. 9% with 30

PAGE 31

the 2%/2mm criterion. With larger MLC positioni ng errors, the variation associated with MapCHECK passing rate distribution gradually increased to around 6% with 3%/3mm and 9% with 2%/2mm, while it stayed constant for radioc hromic film. If the average passing rates didnt decrease significantly when the MLC errors were introduced, the passing rate distribution would overlap the one without errors. Fi gure 2-1 to 2-4 showed that the overlapping was larger for film than for MapCHECK device due to the larger variation associated with film. This fact should explain the larger p-value from rank-sum test with radiochromic film than with MapCHECK. From this perspective, MapCHECK device showed larger sensitivity than radiochromic film to MLC positioning errors. The difference in variation associated with passing rate distribution was mainly due to different detector response. The variation of film sensitivity is on the order of 1.5%~4% depending on dose level37, while MapCHECK has variation of only 0.15%38, 39. Larger variation in the passing rate dist ribution was found when using 2%/2mm than 3%/3mm. At the same time, a larger drop in th e average passing rate was noticed with 2%/2mm than with 3%/3mm as shown in Figure 2-5 to 2-6. Table 2-1 showed sm aller p-values with 2%/2mm than with 3%/3mm in general, which meant a higher probability of distinguishing the passing rate distributions with 2%/2mm. Therefore, a tighter criterion shows more sensitivity to MLC position errors. With 2%/2mm, the average passing rate was ~93% and ~80% with MapCHECK and radiochromic film, respectively. It was relatively difficult to achieve acceptable passing rates using film with tighter criterion37. Both the DTA analysis and index were studied in this work. DTA analysis is more stringent than index and on average gives 2~3% lower passing rates. However, no significant difference was found in the decrease of average pa ssing rate and the p-values from rank-sum test 31

PAGE 32

were in general on the same order, which suggested that the choice of using DTA analysis or index didnt affect the sensitivity of the IMRT QA to MLC position errors. Asymmetric response to +2 mm and -2 mm systematic MLC positioning errors was noticed with both MapCHECK and film. The decr ease in average passing rate was ~3% larger with -2 mm systematic errors than with +2 mm errors when using MapCHECK as the detector. When using film, a 10% larger decrease in aver age passing rate was noticed with -2 mm errors than with +2 mm errors (25% vs. 10% when using 2%/2mm crit erion for both DTA analysis and index). To investigate the asymmetric behavior with respect to systematic MLC positioning errors, the following simulation was conducted to reveal the cause of such behavior. The planar dose distribution calculated from plan without MLC errors was re-sampl ed and only those points corresponding to a detector position on the MapC HECK device were kept. The re-sampled dose distribution was compared to other four dose di stributions from plans with systematic MLC errors (mm and mm) using 2%/2mm DTA criterion. The aver age passing rates of 15 IMRT plans were shown in Figure 2-7. An obvious asymme tric behavior was noticed. Compared to an error free plan, plans with -1 mm symmetric ML C errors had ~3% lower average passing rates than plans with +1 mm symmetric MLC errors (93% vs. 96%). While for 2 mm systematic MLC errors, 17% lower average passing ra tes were observed for plans w ith negative errors than plans with positive errors (66% vs. 83%). A similar trend was observe d when comparing error-free plans to plans with errors in full resolution (i .e., no re-sampling). This simulation exercise showed that the asymmetric behavior is not caused by MLC calibration error or TPS modeling errors, but rather indicated that the dose distribution is more sensitive to negative systematic MLC errors (which makes the MLC segment opening smaller) than positive errors. The 32

PAGE 33

magnitudes of the passing rate differences betw een the negative and positive MLC errors are very similar to those observed in Figure 2-1 to 2-4. Childress et al investigated the feasibility of auto matically detecting both gross delivery errors and small delivery errors (5 mm lateral shift of isocenter, etc) with patient-specific IMRT QA procedures34. They found that most scalar do se comparison metrics, including index, failed in identifying isocenter shift error of 5 mm. Our results indicated that the studied IMRT QA procedures could detect systematic MLC positioni ng errors of 2 mm. A major difference in their work was that they compared composite dose di stribution while our analysis was done on a fieldby-field basis. One reason for poor detection ab ility when using composite dose distribution in comparison could be attributed to the fact that when composite dose distribution was analyzed, errors from individual fields c ould compensate each other. As TPS beam modeling, dose calculation and dose measurement become more accurate, tighter criterion could be employed in the an alysis and still achieve good agreement between calculated and measured dose distribution18, which will lead to better and earlier IMRT delivery error detection, especially with MapCHECK whic h provides enhanced sensitivity compared with EBT film. However, considering the dosimetric imp act of such errors, patient-specific IMRT QA should be combined with a peri odic MLC positioning check program31-33 in order to guarantee the accuracy of IMRT delivery. Conclusions In this work, we investigated the sensitivity of patient-specific IMRT QA procedure to MLC positioning errors. Random errors of up to 2 mm and systematic errors of mm and mm were simulated in the treatment plans. Plan ar dose distributions ca lculated with plans containing errors were compared to dose measurement with both Gafchromic EBT films and 33

PAGE 34

MapCHECK device, acquired while irradiating th e original error-free plan. DTA analysis and index with 2%/2mm or 3%/3mm criteria were employed to evaluate the agreement between calculation and measurement. The change of passi ng rates distribution was used as an indicator of the sensitivity to simulated MLC positioning errors. It was found that the studied patientspecific IMRT QA procedure with both radiochrom ic film and MapCHECK device was just able to detect systematic errors on the order of 2 mm. Significantly larger sensitivity to MLC position errors was found with the 2%/2mm criterion than with the 3%/3mm criterion. Since clinical acceptable agreement between calculation and meas urement using film dosimetry with 2%/2mm is relatively difficult to achieve, MapCHECK is more appropriate to be employed with a tight criterion. As the accuracy of dose calculation and measurement is further improved, the sensitivity of IMRT QA procedure to MLC positioning errors will be greatly enhanced. However, considering the dosimetric impact of such errors, patient-specific IMRT QA should be combined with a periodic MLC QA program in order to guarantee the accuracy of IMRT delivery. 34

PAGE 35

Table 2-1. P-values from two-tailed Wilcoxon ranksum test. The test co mpared the distribution of passing rates before and after th e introduction of MLC position errors. p-Value Random 1 mm -1 mm 2 mm -2 mm EBT Film DTA 2%/2mm 3.28E-01 9.59E-01 1.05E-01 2.81E-02 6.22E-04 3%/3mm 5.56E-01 7.98E-01 1.61E-01 7.74E-02 1.10E-03 Gamma 2%/2mm 5.05E-01 9.81E-01 1.30E-01 2.81E-02 6.22E-04 3%/3mm 6.29E-01 1.00E+00 3.14E-01 9.95E-02 1.90E-03 MapCHECK DTA 2%/2mm 1.10E-02 1.29E-05 2.60E-03 2.35E-11 6.86E-10 3%/3mm 3.20E-02 1.90E-05 3.60E-03 2.84E-11 1.62E-08 Gamma 2%/2mm 1.76E-02 1.08E-05 3.45E-04 3.97E-11 3.28E-10 3%/3mm 7.59E-05 3.85E-06 3.57E-04 4.33E-12 9.20E-10 0 R +1 -1 +2 -2 50 60 70 80 90 100 MLC positioning error (mm)average passing rate (%) DTA 2%/2mm DTA 3%/3mm Figure 2-1. Average passing rates when compar ing Pinnacle calculated planar dose distribution and Gafchromic EBT film measurement us ing 2%/2mm and 3%/ 3mm criterion with DTA analysis for a total of 53 IMRT fields. Th e x-axis, from left to right, stands for no MLC positioning error, random errors and sy stematic errors of mm and mm, respectively. The error bar indi cates one standard deviation. 35

PAGE 36

0 R +1 -1 +2 -2 50 60 70 80 90 100 MLC positioning error (mm)average passing rate (%) Gamma 2%/2mm Gamma 3%/3mm Figure 2-2. Average passing rates when compar ing Pinnacle calculated planar dose distribution and Gafchromic EBT film measurement us ing 2%/2mm and 3%/ 3mm criterion with Gamma index for a total of 53 IMRT fields. Th e x-axis, from left to right, stands for no MLC positioning error, random errors and sy stematic errors of mm and mm, respectively. The error bar indi cates one standard deviation. 0 R +1 -1 +2 -2 50 60 70 80 90 100 MLC positioning error (mm)average passing rate (%) DTA 2%/2mm DTA 3%/3mm Figure 2-3. Average passing rates when compar ing Pinnacle calculated planar dose distribution and MapCHECK measurement using 2%/2mm and 3%/ 3mm criterion with DTA analysis for a total of 53 IMRT fields. The x-axis, from left to right, stands for no MLC positioning error, random errors and syst ematic errors of mm and mm, respectively. The error bar indi cates one standard deviation. 36

PAGE 37

0 R +1 -1 +2 -2 50 60 70 80 90 100 MLC positioning error (mm)average passing rate (%) Gamma 2%/2mm Gamma 3%/3mm Figure 2-4. Average passing rates when compar ing Pinnacle calculated pl anar dose distribution and MapCHECK measurement using 2%/2mm and 3%/ 3mm criterion with Gamma index analysis for a total of 53 IMRT fields. The x-axis, from left to right, stands for no MLC positioning error, random errors and sy stematic errors of mm and mm, respectively. The error bar indi cates one standard deviation. 0 R +1 -1 +2 -2 0 5 10 15 20 25 30 MLC positioning error (mm)drop of average passing rate (%) DTA 3%/3mm DTA 2%/2mm Gamma 3%/3mm Gamma 2%/2mm Figure 2-5. Drop of average passing rates with re spect to the clinical plan (no MLC positioning errors) with the introduction of random and systematic errors. The comparison was between Pinnacle calculated planar dose distribution and film measurement using DTA 2%/2mm and 3%/ 3mm as well as 2%/2mm and 3%/3mm criterion. The xaxis, from left to right, stands for no MLC positioning error, random errors and systematic errors of mm and mm, respectively. 37

PAGE 38

0 R +1 -1 +2 -2 0 5 10 15 20 25 30 MLC positioning error (mm)drop of average passing rate (%) DTA 3%/3mm DTA 2%/2mm Gamma 3%/3mm Gamma 2%/2mm Figure 2-6. Drop of average passing rates with re spect to the clinical plan (no MLC positioning errors) with the introduction of random and systematic errors. The comparison was between Pinnacle calculated planar dose distribution and MapCHECK measurement using DTA 2%/2mm and 3%/ 3mm as well as Gamma 2%/2mm and 3%/3mm criterion. The x-axis, from left to right, stands for no MLC positioning error, random errors and systematic errors of mm and mm, respectively. 0 +1 -1 +2 -2 60 65 70 75 80 85 90 95 100 105 MLC positioning error (mm)average passing rate (%) DTA 2%/2mm Figure 2-7. Average passing rates when comparing error free plans (re-sampled according to MapCHECK detector positions) to plans w ith +/1 mm and +/2 mm systematic MLC errors using DTA 2%/2mm criterion. Th e error bar stands for one standard deviation. 38

PAGE 39

CHAPTER 3 THE EXTRACTION OF TRUE PHOTON BE AM PROFILE FOR TPS COMMISSIONING Introduction The aim of intensity modulated radiation th erapy (IMRT) is to de liver highly conformal radiation dose to the planning target volume while avoiding criti cal structures as much as possible. Complicated IMRT fields are charac terized by many small segments and steep dose gradients which require treatment verification. The main purpose of patient-specific quality assurance (QA) is to detect potential errors which could arise in the chain of IMRT process: treatment planning system (TPS), treatment planning, dose calculation, plan parameter transmission and treatment delivery.2 A common method of patient specific QA is to re-calculate the treat ment plan in a QA phantom with all beams at 0o gantry angry and normal to the phantom surface. The planar dose distribution is measured under same geometry us ing film or two-dimensional (2D) detector arrays. The agreement between th e calculated and measured planar dose distribution is then quantified using parameters like the Gamma index,10, 11, 42, 43 percent dose difference (%Diff), or distance-to-agreement (DTA), etc. The existence of various sources of errors in the chain of the IMRT process makes it difficult to analyze unsatisfactory IMRT QA results. Large clinical effort has been put in identifying the sources of discrepancy. If necessary, the plan is changed to achieve acceptable IMRT QA analysis statistics11. Studies44, 45 have shown that the main source of discrepancy is detector size effect in T PS commissioning data, which makes it even more difficult to detect errors due to mechanical inaccuracy or TPS dose calculation error. An essential issue in the beam modeling of T PS is to ensure the calculated dose matches the dose delivered to the patient. A key com ponent in successful TPS commissioning is the accurate measurement of beam profiles1, especially the penumbra44. Measurement error in the 39

PAGE 40

beam profile will propagate to TPS dose calculation directly. A well-known source of error during cross beam profile measurement is the volu me averaging effect caused by the finite size of the measuring detectors, which causes the ar tificial broadening of the profile penumbra. Depending on the specific detector and radiatio n field, the commonly used standard ionization chambers with volumes on the order of 0.1~0.2 cm3 can cause penumbra increase by 0.2 ~ 0.3 cm,44, 46, 47 which is obviously large enough to influe nce the IMRT dose calc ulation accuracy. In the effort to rectify the measurement, two appro aches have been attempted to recover the true profiles in the literatures46: (1) perform the measurement with variable size detectors and extrapolate to zero size detector46-50 (2) deconvolve the detector response kernel from the measured profiles.46, 51-56 The first approach utilizes detector s of variable sizes to measure the same beam profiles. The relationship betwee n penumbra width and detector sizes can be established and the true penumbra can be r ecovered by extrapolating down to zero size detector. The disadvantage of this method include s: (1) several different sized detectors are required; (2) it is time-consuming. The second a pproach either calculat es or measures the detector response kernel and its effect on penumbra broadeni ng can then be eliminated by numerical or analytical dec onvolution. Numerical deconvolution suffers from measurement noise and becomes very unstable.51, 52, 54, 55, 57 Analytical deconvolution pr oduces stable outcome but it depends on the choice of analytic al functions which would fit bot h the profiles and the kernel and the deconvolution can be solved analytically.51, 57 Although the detector size effect on the prof ile measurement has been well understood, its clinical impact on IMRT QA has not been quantified. Laub and Wong pointed out the importance of the detector size effect in small field dosimetry in IMRT but the effect on patient specific IMRT QA was not studied.48 Arnfield et al used film dosimetry to improve beam profile 40

PAGE 41

penumbra measurement and observed improved agreement between calculation and measurement.44 Esch et al studied patient specific IMRT QA results among five European departments and concluded that the detector size effect was one of the main sources for discrepancy. But the degree of influence was not determined.45 A recent survey11 showed that the majority of the community which uses planar dose verification method em ploys 3%/3mm criteria for IMRT QA analysis with the number of points passing the criteria (passing rate s) of 90%-95% as acceptable. Yang et al58 proposed an acceptable criterion of 85% passing rates with %3 and DTA, which are considered achievable even with the most complicated cases. Given that the leaf positioning accuracy for mo st MLC systems can achieve better than 1 mm accuracy,15, 29 the use of as distance to agreement does not seem justified. Is detector size effect the limiting factor? Could more stringent cr iteria be applied if detector size effect is removed? The impact of the detector size effect on the results of IMRT pl anar dose verification is the focus of this work. mm 3 mm 3The aim of this work is two fold: (1) to de velop a method of extrac ting the true cross beam profiles, free of volume averaging effect, using analytic fitting/deconvolution; and (2) to investigate the clinical impact of detector size effect on patient specific IMRT QA. Photon beam profiles were obtained with a 6 mm and a 4 mm di ameter ion chambers, respectively. An analytic function was used to fit only the high dose gradie nt regions of the profiles (penumbra regions) and the true profile set was r ecovered via an analytic deconvol ution approach. Each of these beam profile sets was used to commission a photon beam model. Planar do se distributions were calculated for 53 IMRT fields using each of th e three beam models. The agreement between calculated and measured planar dose distribution was evaluated us ing %Diff and DTA. 41

PAGE 42

Significantly better agreement between calculated and measured planar dose distributions was achieved with the beam model commi ssioned using the deconvolved profiles. Methods and Materials Beam Data Measurement A commercial treatment pla nning system (TPS, Pinnacle3, version 8.0d, Philips Medical Systems, Madison, WI) which employs a convolution-superposition based dose calculation algorithm was commissioned for the 6 MV photon beam of a linear accelerat or (Trilogy, Varian Medical Systems, Palo Alto, CA) with 120-leaf Millennium MLC. Data requirement for the TPS commissioning includes depth dose cu rves at different field sizes, cross beam dose profiles for collimatorand MLC-defined fields at different field sizes and depths calibration and relative output factors, block and tray tr ansmission factors. Depth dose and cross beam profiles were acquired with two scanning ionization chambe rs: one with 6 mm diameter, 5.8 mm length and 0.13 cm3 volume (CC13, Scanditronix Wellhofer, Ba rtlett, TN), and the other with 4 mm diameter, 3.6 mm length and 0.04 cm3 volume (CC04, Scanditron ix Wellhofer) in a 48x48x48 cm3 scanning water tank (Wellhofer Dosimetrie Shwarzenbruck, Germany). All beam profiles were collected at a source-tosurface distance (SSD) of 90 cm. Beam Profile Deconvolution The effect of the finite size of the detector on beam profile measurement is described by a convolution of the dete ctor response function K(x) to the true profile Pt(x)51, 52, 55 duuxKuPxPt m)()()( (3-1) where Pm(x) is the measured profile. The true profile can be derived by using the convolution theorem, )]([)]([)]([ xKFxPFxPFt m (3-2) 42

PAGE 43

and inverse Fourier transform, )]([ )]([ )(1xKF xPF FxPm t (3-3) Where and denote Fourier transform and inverse Fourier transform, respectively. So there are two steps to recover the true profile. F1 F 1). Determine the detector response function K(x) 2). Perform de-convolution on measured profile. As mentioned above, discrete deconvolu tion suffers from measurement noise and numerical errors. In practice, it is very complex and the results depend on the noise level, the measurement uncertainty and the cut-off frequency chosen to filter the inherent noise.51, 57 This approach becomes impractical in TPS beam mo deling where a large num ber of profiles are needed.59-61 Garcia-Vicente et al51 provided an exact analytic solution to the deconvolution problem. The solution was limited to small field size profiles. In this work, we extended and applied their method to profile of arbitrary fi eld size and depth for the purpose of TPS beam modeling. Detector response function The detector response function for arbitrary de tector could be deri ved from Equation 3-1 and Equation 3-2 if the true profile is known. A common approach to deri ve the true profile is to measure the same profile using variable size detectors an d extrapolate down to zero size detector.46-50 The downside of this approach is that measurements with a large number of variable size detectors are requi red in order to get good confiden ce in the extrapolation. Pappas et al 57 studied the relationship between detector size and penumbra width using gel dosimeter and concluded that detector sizes of 0.5 mm to 0.7 mm were adequate for relative dose 43

PAGE 44

measurements in high gradient regions of a 5 mm diameter photon beam. In this study, we used a small diode detector (Edge Detector, S un Nuclear Corp., Melbourne, FL) with 0.8 x 0.8 mm2 effective detection area to co llect the true profile, Pt( x ), for small field sizes. Comparison between the profiles measured using the diode and radiochromic films showed excellent agreement for small field sizes62. The same profiles measured with CC13 were denoted as Pm( x ). The detector response function is as sumed to have a Gaussian shape,51 22/)(kxexK (3-4) where k is the shape parameter. k is related to the sigma of a standard Gaussian function form s by multiplying k by a factor of 0.4. The optimal k was obtained by minimizing the difference between Pm( x ) and the convolution of Pt( x ) and K ( x ) in a least square sense. The modified Levenberg-Marquardt algorithm wa s used to solve this nonlinear problem.63 Cross beam profiles of collimator defined field sizes of 4 x 4 cm2 and 10 x 10 cm2 were measured with the Edge Detector at SSD of 90 cm and at depths of 1.5 cm, 10 cm, and 25 cm to derive the detector response function parameters. This procedure was restricted to small (10x 10 cm2) field sizes since diode measured prof iles are known to deviate from chamber measured profiles at large field size due to the energy dependence of its response.48, 64-66 The dependence of the detector response function on field size and de pth was also investigated. Profile deconvolution The essential part of analytical deconvolution is to find an an alytic function that can fit the measured beam profiles well and Equation 3-3 can be solved in a closed form. Garcia-Vicente et al51 used the difference of error f unctions, originally proposed by Cho et al ,67 to fit small photon fields, c xLerf xLerfxFm m m )]}2/()2()[()]2/()2()[({2/1)(2/1 2/1 (3-5) 44

PAGE 45

where L is the field width, m the shape parameter and erf (x) the error function, dte xerfx t0 2/122 )( (3-6) They worked out an analytic solution to E quation 3-3 which has the same functional shape as Equation 3-5 but with a new shape parameter t, 222 kmt (3-7) Thus, if the measured beam profile could be fit reasonably well using Equation 3-5, the true profile could be recove red once the shape parameter k for the detector response function was determined and deconvolution becomes a simple replacement of the shape parameter with the new one. It was demonstrated that good fits could be achieved for small photon beams using Equation 3-5.51, 67 However, large field photon beams are characterized by the well known horn effect in the high dose region due to the fl attening filter. The difference of error functions will fail in reproducing this featur e which is present in all larg e field photon beams at various depths. Computer simulation68 showed that the effect of volume averaging on low dose gradient regions (including the tail regi on and horn-shaped high dose re gion) was negligible; only the high dose gradient region was a ffected significantly. This motivat ed us to exclude the low dose gradient region and only keep the high dose grad ient region in consideration when rectifying measured profiles. Meanwhile, the function used to fit the measured profile could be extended as following, N n n mxFxF1)( )( (3-8) where each Fn(x) is the difference of error functions as expressed in Equation 3-5. We found that the use of gave the best fit for all the prof iles with no improvement when more 5 N 45

PAGE 46

terms were used. The selection of high dose gradie nt region could be eith er gradient based or distance based. In gradient based selection, only the region where lo cal gradient is above 10% of the gradient at the inflection point (the point where the derivative of the gradient changes signs) is included. In distance based approach, the re gion falling in between 80% and 120% of field width is included. Both methods were implemented and results showed little difference. In this study, the gradient based selection was used. Once the measured profile is fit using Equation 3-8, the deconvolution is simply to adjust the shape parameter of each Fn(x) according to Equation 37. In-plane and cross-plane beam profiles for symm etric fields with fiel d sizes ranging from 4 x 4 cm2 to 40 x 40 cm2 at depths of 1.5 cm, 10 cm, and 25 cm were deconvolved using this approach. Asymmetric profiles were divided into two half profiles at the field center and each half was deconvolved separately and then combined t ogether. The deconvolved profiles for the two smaller field sizes of 4 x 4 cm2 and 10 x 10 cm2 were validated using the Edge Detector measured profiles. TPS Commissioning External photon beam commissioning for Pinnacle3 has been well documented in the literature.59-61 The convolution-superposition algorithm employed in the TPS computes dose distributions from first principl es. Photon fluence exiting the head of the linac was characterized by a set of photon beam model parameters. Differ ent regions of the depth dose curves and dose profiles were used to adjust the pa rameters which characterize the beam.59 The rounded leaf end effect of the MLC was characterized by a set of parameters which modeled the penumbral difference between the jaw-define d and MLC-defined profiles. Thes e parameters were iteratively adjusted during the modeling process such that the calculated profiles match the measured ones as much as possible. 46

PAGE 47

Three different beam data sets (CC13-m easured, CC04-measured and deconvolved beam data) were used to commission Pinnacle3 for the 6 MV photon beam of the Trilogy and three photon beam models (BM6, BM4 and BM08, respectively) were created. A 2 x 2 x 2 mm3 voxel size was used during the commissioning. Task Group report 53 (TG-53) of The Radiation Therapy Committee of the American Asso ciation of Physicists in Medicine69 was followed during the commissioning process. Patient Specific IMRT QA Comparison Five head & neck (H&N) IMRT plans and thre e prostate IMRT plans (for a total of 53 fields) previously used for patient treatment on the Trilogy were randomly selected for this study. The plans were generated using the DMPO (Direct Machine Parameter Optimization) option in the TPS, which optimized the shape and weight of each MLC segment directly. The three beam models BM6, BM4 and BM08 were used to calculate planar dose distribution for each field at 10 cm depth and 90 cm SSD in a rectangular water phantom with 0o gantry angle (IEC convention). All the calcula tion was done using a 1.0 x 1.0 mm2 dose grid. The measurement was performed in a solid water ph antom using a 2-dimensional (2D) diode array (MapCHECK 1175, Sun Nuclear Corp., Melbourne, FL) under the same geometry. The MapCHECK device has been demonstrated to be an excellent tool for th e verification of IMRT fields with little energy or dose rate dependence.38, 39 The small size of its detectors (0.8x0.8 mm2) ensures minimal volume averaging effect An intercomparison between film and MapCHECK showed that MapCHECK can re place film dosimetry in routine IMRT QA.40 The MapCHECK device was calibra ted for dose distribution measurement based on the manufacturer-recommended procedures. The calcul ated planar dose distributions were imported into the MapCHECK analysis software. Each of the MapCHECK measured points above 10% of the maximum dose level was compared with calculation using %Diff and DTA acceptance 47

PAGE 48

criteria in the absolute dose comparison mode.9, 10 Two sets of criteria, 2%/2mm and 3%/3mm, were employed, and the percentage of the point s passing the acceptance criteria was evaluated. Results Figure 3-1 shows the comparison of the cros s beam profiles measured with the Edge Detector, CC04, and CC13 for a 10 x 10 cm2 field at 10 cm depth in water phantom with SSD=90 cm. The 80%-20% penumbra widths are 0.40 cm 0.61 cm and 0.72 cm for the diode, CC04 and CC13, respectively. The penumbra was broadened by 0.2 ~ 0.3 cm using the commonly used ion chambers. The difference is cons istent with published results.44, 46, 47 This necessitates the need to deconvolve the profiles measured with finite size ion chambers. The beam modeling process is as important as the beam data accuracy, especially for MLC-delimited small field sizes. Figure 3-2 (a) and (b) show diode measured beam profiles and TPS calculated beam profiles with three different beam models for MLC-delimited 2x2 cm2 and 4x4 cm2, respectively. The agreement in penumbra (80% to 20%) between diode measured data and beam model BM08 calculated data was with in 1 mm, which met the criteria recommended by TG-5369. Similar agreement was achieved between ion chamber measured data and TPS calculation with the co rresponding beam models. The derived parameters for the detector response function for CC13 are summarized in Table 3-1 and demonstrated in Figure 3-3 for the two field sizes and at th ree depths. These were determined using Equation 3-1 in a least squa res fit and assuming that the diode measured profiles are the true cross beam profiles. Figure 3-3(a) shows the results for 4 x 4 cm2 field size. The convolution between the diode measured data and the best fit Gau ssian function at each depth agrees well with the CC 13 measured profile. The Gaussian function parameters k obtained at depths of 1.5 cm, 10 cm and 25 cm were 0.68 cm, 0.68 cm a nd 0.70 cm, respectively. 48

PAGE 49

When converted to the sigma of a standard Gaussian function ( s) by multiplying k by a factor of 0.4, they correspond to 0.27 cm, 0.27 cm and 0.28 cm, respectively. Figure 3-3(b) shows the results for the 10 x 10 cm2 field size. The optimal parameters k obtained were 0.68 cm, 0.72 cm and 0.69 cm at depths of 1.5 cm, 10 cm and 25 cm, respectively, which correspond to 0.27 cm, 0.28 cm and 0.28 cm when converted to s. These are very close to those obtained with the 4 x 4 cm2 field size. The results indica te that the detector response function is independent of both field size and depth in th e ranges studied. The derived s is also very close to the radius (0.3 cm) of the CC13 chamber, which agrees with the findings of other works.46, 54 The analytic fit using Equati on 3-8 to the CC13 measured cr oss beam profiles for field sizes 4x 4 cm2 and 10 x 10 cm2 at the three different depths are shown in Figure 3-4. By restricting the analytic fit to high gradient regions of the profiles and exte nding the analytic function from Equation 3-5 to Equation 3-8, exce llent fit can be achieved at all depths (1.5 cm, 10 cm and 25 cm). The deconvolved profiles, obtained by deconvolving the analytically fitted profiles with the detector response function obtaine d above, agree well with the diode measured profiles at each depth. The advantage of using the analytic functional form of E quation 3-8 to fit the measured profiles is that deconvolution can be solved analytically and is simply a replacement of the shape parameter m with the new one based on Equation 3-7. The results validated the procedures of extracting the true profiles with the analytic fit/deconvolution method. The results of applying the above method to extr act the true profiles for larger field sizes and assuming that the detector response functio n is independent of field size and depth are illustrated in Figure 3-5, where results for field size 30 x 30 cm2 at three depths are shown. Excellent analytic fit was also obtained for field sizes 20 x 20 cm2 and 40 x 40 cm2 at all depths (not shown). Figure 3-6 shows the results of extracting the true pr ofiles for MLC-defined 49

PAGE 50

asymmetric fields. Asymmetric profiles were divided into two halves along the center of the field and were treated separately. The deconvolved prof iles were then merged together to form the complete profiles. Figure 3-6 shows that the same functional form that fit well the collimator defined fields was equally effective in fitting the MLC defined fields. By restricting the analytic fit to high dose gradient regions of the profiles, we were able to apply the analytic fitting method to larger field sizes which would otherwise fail in trying to reproduce the horn effect. The passing rates between the measured and cal culated planar dose distributions for the 8 clinical cases (53 treatment fi elds) using both the 2%/2mm and 3%/3mm criteria are summarized in Figure 3-7, where the passing rates were binned into 5% steps for the three different beam models. Table 3-2 lists the average passing rates for all the cases along wi th their associated standard deviations. The data demonstrates a clear trend where beam models commissioned using profiles obtained with smaller detectors have higher passing rate s than those commissioned using profiles obtained with larger detectors. The deconvolved beam model BM08 gave the best passing rates among the three beam models w ith an average passing rate of 96.8% when evaluated with the 2%/2mm criterion, compared with average passing rates of 92.9% and 81.6% for beam models BM4 and BM 6, respectively. For beam mode l BM08, all the cases achieved passing rates of > 90% and ~76% of which were above 95%. For beam model BM4, 81% of the cases had passing rates greater than 90% and 31% of which were above 95%; while for beam model BM6, only 10% of which were above 90% and none of which was above 95%. When the criterion of 3%/3mm was employed, both BM4 and BM08 achieved excellent results with passing rates for all the cases greater than 95%. The average passing rates for both beam models were above 98%. However, for beam model BM 6, although above 86% of all the cases had 50

PAGE 51

passing rates greater than 90%, only 37% of which were above 95%. The average passing rate for BM6 was only 93.8% with th e criterion of 3%/3mm. Figure 3-8 shows an example of calculated plan ar dose distributions using three different beam models and their comparison to MapCHE CK measurement using 2%/2mm criterion. In this case, the passing rates were 85%, 96% and 98% for BM6, BM4 and BM08, respectively. The failing points were indicated on each calcul ated dose distribution with stars and squares representing hot spots and cold spots, respectively. It was clear ly shown on Figure 3-8(a) that most failing points occurred around the high gradient region. Also pl otted in Figure 3-8(d) was a comparison among calculated in-plane dose profiles using three different beam models and the MapCHECK measurement. The calculated profil es were taken from the same position of each calculated planar dose distributi on and the position was indicated by the arrows. The calculated beam profile with BM6 was characterized with more rounded peaks and valleys and slower gradient compared with calculated profiles with th e other two beam models due to detector size effect. The detector size effect was less pr onounced with BM4. Similar dose values were achieved with all three beam models in low dose and high dose but low gradient regions. Discussion There has been a great interest in extracting the true profiles from profiles measured with finite sized detectors. In this study, an analytic deconvolution procedure was developed based on the work of Garcia-Vicente.51 The analytic function was extende d and a gradient based selecting approach was introduced to achieve g ood fit to the measured profiles. The true profiles for small field sizes (4 x 4 cm2 and 10x 10 cm2) were collected using a small diode detector. In literatu re, it was usually obtained by pr ofile measurement using various size detectors and extrapolati on down to zero size detector.46-50 This process requires many detectors and is very time consuming. In addition, error arises around ze ro size region where the 51

PAGE 52

linear relationship assumed duri ng the extrapolation may not hold.48, 57, 68, 70 Other options are film and diode detector. Compared to ion chambe r, film has excellent sp atial resolution and has been used in the determination of the detector response function.44, 46 However, accurate film dosimetry is time consuming and it has the problem of over-response to low energy photons for radiographic films. Diode detectors have over-response for broad beams (over 10 x 10 cm2) due to energy and directional dependence. But the influen ce is minimal when the diode detector is used to measure the penumbra of small field sizes ( 10x 10 cm2). The active measuring area of the diode detector is small enough to re solve the volume averaging effects.57 It has been shown that the penumbra widths measured by silicon diodes are very close to those measured by film and diamond detectors.44, 48, 71 When the true profile is obtai ned, numerical procedures such as deconvolution or inverse Fourier transform can be used to derive the parameter of the detector response function.52, 55 As mentioned above, numerical procedures suffer from the measurement noise or uncertainty and the results could be disconcerting. In this work, a curve fitting procedure was used to derive the Gaussian parameter for the detector response func tion. The true profiles we re compared with the CC13 measured profiles after c onvolution with the Gaussian func tion in a least squares sense. The influence of the measurement noise was mini mized. Very reliable an d consistent Gaussian parameters were achieved throughout di fferent geometries (field sizes of 4 x 4 cm2 and 10 x 10 cm2 and depths of 1.5 cm, 10 cm, and 25 cm, see Table 3-1). The sigma of the corresponding standard Gaussian function falls in 0.27~0.29 cm, wh ich is very close to the radius of CC13 ion chamber (0.3 cm), similar to the findings of previous works.46, 54 Bloch and Wallace72 found that the broadening of the penumbra was independent of the depths of scan in phantom. The results of this work are in agreement with this finding. In addition, we found no ch ange in the detector 52

PAGE 53

response function from field sizes 4 x 4 cm2 to 10 x10 cm2. We therefore assumed that the detector response function derived from small field si zes can be applied to large field sizes. Gaussian function and the difference of error functions have been used to fit photon beam profiles51, 57. However, they are limited to small field sizes and cant reproduce asymmetric or horn shaped profiles. In this work, the extende d difference of error functions was used to fit only the high dose gradient regions of an arbitrary profile. The gr adient based selection approach was motivated from the observation that only high dose gradient regions of the profiles suffer from volume averaging. The high dose, low gradient region, characterized by unflatness or horns, is excluded in the fitting process. Asymmetric profiles were divided into two half profiles along the field center and treated sepa rately. In this way, the analytic fitting can be performed for profiles of arbitr ary field sizes. The purpose of IMRT patient specific QA is to detect any errors in the chain of IMRT planning and delivery process. When the errors introduced by the volume averaging effect of the beam commissioning data are significant, the QA results will be less se nsitive to any other dosimetric errors. This would de feat the purpose of patient sp ecific QA. When 3%/3mm criterion was used, the averaging passing ra tes for 57 IMRT fields were in creased by 5% from ~94% to ~99% when the detector size was decreased from 0.6 cm to less than 0.4 cm. When using 2%/2mm criterion, 10% increase in averaging passing rates wa s achieved with detector size decrease from 0.6 cm to 0.4 cm. A further increase of 4% was achieved from the CC04measured beam model to the deconvolved beam model. These results confirmed the conclusion of Esch et al that the main source of discrepancy between calculated and measured dose distribution was volume averaging effect caused by the finite si ze of the detectors used for collecting the data for TPS beam modeling.45 Unsatisfactory IMRT QA results caused mainly by 53

PAGE 54

detector size effect could l ead to wrong clinical decisions48 and probably a waste of clinical resources (e.g., in trying to identify the sources of error for each questionable IMRT field, redo the IMRT plan11, etc). On the other hand, when the detect or size effect is removed, a stringent acceptance criteria (such as 2%/2mm) in IMRT pa tient specific QA could be employed. Thus the QA procedure would be more sensitive to systema tic uncertainty in the system. For example, a recent study showed that a 1 mm systematic error in MLC positioning could lead to clinically significant dosimetric errors.26 The deconvolved beam mode co mbined with more stringent acceptance criteria such as 2%/2mm would be mo re likely to detect the MLC positioning error. This is the subject of our future study. Conclusions A simple analytic approach has been deve loped to deconvolve ion chamber measured profiles and extract the true profiles at arbitrary field size and depth. A series of difference of error functions and a gradient base d fitting approach has been used to fit the measured profiles. The deconvolution procedure has been demonstrat ed to be effective under different geometries. CC04 and CC13 measured beam profiles as well as the deconvolved profiles were used to commission a commercial TPS. Agreement between calculated and diode array measured planar dose distribution were evaluate d using percent dose differen ce and DTA criteria. The passing rates were increased significantly with the decr ease in the detector size used to obtain the commissioning beam profiles, with an average passing rate of ~96% using the 2%/2mm criterion for the beam model using the d econvolved beam profiles. These results confirm the hypothesis that detector size effect is the main source of discrepancy in IMRT patient specific QA. When detector size effect was rem oved from the measured beam profiles for commissioning, more stringent criteria can be employed clinically wh ich will likely result in higher chances of detecting dosimetric errors caused by TPS or MLC positioning or other mechanical errors. 54

PAGE 55

Table 3-1. Best-fit shape parameter k of the detector response f unction for CC13 at depths of 1.5 cm, 10.0 cm and 25.0 cm for field sizes of 4x4 cm2 and 10x10 cm2. The sigma of the corresponding standard Gaussian function ( s) is obtained by multiplying k by a factor of 0.4. Depth 1.5 cm 10 cm 25 cm k (cm) s(cm) k (cm) s (cm) k (cm) s (cm) 4x4 cm2 0.68 0.27 0.68 0.27 0.70 0.28 10x10 cm2 0.68 0.27 0.72 0.29 0.69 0.28 Table 3-2. Average passing rate and standard deviation of the IMRT QA results for 53 fields for the three photon beam m odels. The measurement was performed with the MapCHECK device. IMRT plan was generated using the BM4 beam model for each patient. The same beam parameters (MUs, se gment weights, etc) were then employed to calculate planar dose dist ributions with each of the three beam models. Percent dose difference (%Diff) and distance-to-ag reement (DTA) accepta nce criteria were used to evaluate the agreement between calculated and measured dose distribution. CC13 model (BM6) CC04 model (B M4) Deconvolved model (BM08) 2%/2mm 81.6.8 92.6.3 96.8.3 3%/3mm 93.8.1 98.9.0 99.4.6 55

PAGE 56

3.5 4 4.5 5 5.5 6 6.5 0 10 20 30 40 50 60 70 80 90 100 off axis distance (cm)relative beam intensity 10x10 cm2 CC04 CC13 diode Figure 3-1. Comparison of cro ss beam profiles measured with the Edge Detector, CC04, and CC13 for a 10 x 10 cm2 field of a 6 MV photon beam from Varian Trilogy at 10 cm depth and 90 cm SSD. The 80%~20% penum bra distance is 4.0 mm, 6.1 mm and 7.2 mm, respectively. 56

PAGE 57

-4 -3 -2 -1 0 1 2 3 4 0 20 40 60 80 100 120 off axis distance (cm)relative beam profile (%) (a) MLC 2x2 cm2 Diode CC13 BM CC04 BM True BM -4 -3 -2 -1 0 1 2 3 4 0 20 40 60 80 100 120 off axis distance (cm)relative beam profile (%) (b) MLC 4x4 cm2 Diode CC13 BM CC04 BM True BM Figure 3-2. Diode measured beam profiles and the corresponding TPS calculated beam profiles using three different beam models ( BM6, BM4 and BM08) for (a) MLCdelimited 2x2 cm2 and (b) MLC-delimited 4x4 cm2. 57

PAGE 58

1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3 0 10 20 30 40 50 60 70 80 90 100 110 off axis distance (cm)relative beam intensity (%) (a) 4x4 cm2 diode CC13 convolved 1.5 cm 10 cm 25 cm 3.5 4 4.5 5 5.5 6 6.5 0 10 20 30 40 50 60 70 80 90 100 110 off axis distance (cm)relative beam intensity (%) (b) 10x10 cm2 diode CC13 convolved 1.5 cm 10 cm 25 cm Figure 3-3. Comparison of the diode measured (diode) and CC13 m easured (CC13) cross beam profiles at three depths (1.5 cm, 10.0 cm, and 25.0 cm) for field sizes of 4 x 4 cm2 (a) and 10x 10 cm2 (b) at 90 cm SSD. The convolved profiles (convolved), obtained by convolving the diode measured profiles with the best-fit detector response function, are also shown. The optim al Gaussian Sigmas obtained at three different depths for the two field sizes are listed in Table 4-1. 58

PAGE 59

1 1.5 2 2.5 3 3.5 4 0 10 20 30 40 50 60 70 80 90 100 110 off axis distance (cm)relative beam intensity (%) (a) 4x4 cm2 diode CC13 analytic fit deconvolved 1.5 cm 10 cm 25 cm 3 3.5 4 4.5 5 5.5 6 6.5 7 7.5 8 0 10 20 30 40 50 60 70 80 90 100 110 off axis distance (cm)relative beam intensity (%) (b) 10x10 cm2 diode CC13 analytic fit deconvolved 1.5 cm 10 cm 25 cm Figure 3-4. Analytic fit using Equation 3-8 to the CC13 measured cross beam profiles (CC13) for field sizes of 4 x 4 cm2 (a) and 10x 10 cm2 (b) at depths 1.5 cm, 10.0 cm, and 25.0 cm. Also plotted are the deconvolved prof iles from the analytic fit and the diode measured profiles (diode). 59

PAGE 60

11 12 13 14 15 16 17 18 19 0 10 20 30 40 50 60 70 80 90 100 110 off axis distance (cm)relative beam intensity (%) 30x30 cm2 CC13 analytic fit deconvolved 1.5 cm 10 cm 25 cm Figure 3-5. Analytic fit using Equation 3-8 to the CC13 measured cross beam profiles (CC13) for field size 30 x 30 cm2 at depths 1.5 cm, 10.0 cm, a nd 25.0 cm. Also plotted are the deconvolved profiles from the analytic fit using the detector response function obtained from the smaller field sizes of 4 x 4 cm2 and 10x 10 cm2. -5 0 5 10 15 0 10 20 30 40 50 60 70 80 90 100 110 off axis distance (cm)relative beam intensity (%) CC13 analytic fit deconvolved Figure 3-6. Analytic fit using Equation 3-8 to the CC13 measured cross beam profiles (CC13) at 10 cm depth for MLC-defined asymmetric fields. One MLC bank was fixed at 15.0 cm and the other one was placed at 5.0 cm, 0 cm, and -5.0 cm in the cross-plane direction. The profiles were measured in the cross-plane direction, which is parallel to the direction of the MLC leaf travel. Also plotted are the deconvolved profiles from the analytic fit using the detector response function obt ained from the collimator defined field sizes of 4 x 4 cm2 and 10 x 10 cm2. 60

PAGE 61

75 80 85 90 95 100 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 passing rate (%) with 2%/2mm criterionfraction (a) BM6 BM4 BM08 75 80 85 90 95 100 0 0.2 0.4 0.6 0.8 1.0 passing rate (%) with 3%/3mm criterionfraction (b) BM6 BM4 BM08 Figure 3-7. Comparison of the passing rates be tween the MapCHECK measured and calculated planar dose distributions for 8 clinical cases (53 treatment fields) using three different beam models (BM6, BM4 and BM08). Comparison was done using the MapCHECK analysis software (version 2.01) in the AD mode with both (a) 2%/2mm and (b) 3%/3mm DTA criteria. The pa ssing rates were binned into 5% steps. 61

PAGE 62

(a) CC13 Beam Model, 85% (b) CC04 Beam Model, 96% (c) True Beam Model, 98% 0 50 100 0 5 10 15 20 25 30 off axis distance (mm)absolute dose (cGy) (d) MapCHECK CC13 BM CC04 BM True BM Figure 3-8. Example of calculated planar dose di stributions using three different beam models and the comparison with MapCHECK m easurement using 2%/2mm criterion (a) BM6, 85% (b) BM4, 96% and (c) BM08, 98%. Hot spots and cold spots were indicated by stars and square s respectively. Also plotted in (d) were calculated inplane profiles compared with MapCHECK measurement. The profiles were taken from the same position of each calculated planar dose distribution and the position was indicated by arrows. 62

PAGE 63

CHAPTER 4 INDEPENDENT PLANAR DOSE CALCULATION FOR IMRT QA Introduction Patient-specific quality assurance in inte nsity-modulated radiation therapy (IMRT) involves the verification of treatment paramete rs generated from the treatment planning system2. In most places, IMRT plans are verified with independent measurement in phantom. This has been done with ionization chambers, radiographi c or radiochromic films, thermo-luminescent dosimeter (TLD) chips, electronic portal imagi ng devices or other two dimensional detector arrays. These time-consuming meas urements require actual beam delivery and thus are usually performed after hours in a busy clinic. Alterna tively, IMRT plan verification can be conducted through independent computationa l algorithms. Algorithms for both stereotactic radiosurgery and large size multi-leaf collimator (MLC) shaped fields have been reported73-75. In these planar dose calculation algorithms, dose output factor can be separated into two components: a head scatter factor and a phantom scatter factor. Head scatter factor accounts fo r radiation scattered by the machine head structure reaching the point of calculation in-air while phantom scatter factor accounts for the scattered dose deposition within the phantom. Typically, phantom scatter factors are calculated through the convolut ion between in-air fluence distribution and a dose deposition kernel which is either genera ted via Monte Carlo simulation75 or an analytic kernel with parameters fitted from measured profiles73, 74. Recently, Mihaylov et al compared the accuracy of dose calculation using both analytic methods and Monte Carlo method to predict fluence modulation76. Although using analytic methods for fluence modulation calculation was f ound to be not as accurate as when Monte Carlo method was employed, analytic methods are easy to implement and can potentially achieve the desired accuracy for independent MU a nd fluence verification for IMRT plans. Many 63

PAGE 64

different analytic photon source models for head scatter factor calcula tion have been proposed7782. Most of the source models consist of a primar y (focal) point source at the X-ray target and one or more extra-focal sub-sources representing scattered photons from the primary collimator, flattening filter and/or other h ead structures. Some source models explicitly model the radiation backscattered into monitor chamber from the collimator jaws79. The main focus of this paper is to compare three different source models regardi ng their ability to model head scatter factors using a comprehensive set of data measured on a linear accelerator (LINAC) as well as facilitate planar dose calculation. Methods and Materials Head Scatter Source Models Head scatter source models use empirical m odels to parameterize the head scatter in LINAC head structure. Head st ructure components in the beam line of a typical LINAC include the X-ray target, primary collimator, exit window, flattening filter, monitor chamber, mirror and collimator jaws. Results of Monte Carlo simulations77, 83, 84 showed that more than 90% of the scattered photons are produced in the primary collimator and the flattening filter. Dose contribution to the isocenter from other scattering components including the exit window, monitor chamber, mirror and collimator jaws is negligible84-87. Thus, when developing a head scatter source model, we could model the contribution from the X-ray target, primary collimator and flattening filter only and ignore other components. Ding et al also showed the backscatter from the collimator jaws to the monitor cham ber plays a significant role in the beam output factors84. So it is likely one has to model the backscat ter effect explicitly in order to achieve a very accurate head scatter source model. The three different source models compared in this work are the three-source model78, the dual-source model79 and a single source model75, all modeled for one linac manufacturer (Varian Medi cal Systems). The three-source model has three 64

PAGE 65

sub-source components which are based on the results of Monte Carlo simulation 77. The dualsource model has two sub-source co mponents and it models the back scatter radiation explicitly. The single-source model has only on e source component. Our objective is to achieve faster head scatter calculation while still maintaining accepta ble accuracy for IMRT dose calculation. In this section, we briefly describe thes e three different source models. Three-source model. The three-source model proposed by Yang et al 78 used one point source for the primary photons coming from the X -ray target, and two extra-focal photon sources for photons scattered from the primary collimator and the flattening filter, respectively. There are three contributing factors to the total fluence at a point of calculation: (1) an isotropic point source at the X-ray target with intensity Cp (for all source intensities in this paper, the unit 1/MU, i.e., per monitor unit, is assumed); (2) a planar annulus source about 4.0 cm downwards from the target with 0.2 cm inner radii and 1.4 cm outer radii, the intensity of which is constant (Csp) inside the annulus and zero elsewhere; (3) a pl anar disk source about 12.5 cm downwards from the target and the source intensity Isf could be expresse d analytically as )exp()(0kr r A rIsf (4-1) Where A0 and k are constants and r is the radial distance from a point on the source plane to the center of the source. The shape, location and size of each sub source component are motivated from Monte Carlo simulation77, 83. Backscatter radiati on from the upper surface of the jaws into the monitor chamber is not separated or modele d directly. Instead, the effect is taken into account when fitting model parameters using head scat ter factors of different square field sizes. The fluence at the point of calculation was deri ved by integrating the visible source portion of each sub-source plane from the detectors eye view. The output factor is simply the fluence 65

PAGE 66

normalized to the fluence under reference conditi on. When the calculation point is on the isocenter plane, the head sc atter factor is simply drIdrICSsf spsf sp pc )( )( (4-2) where sp and sf are the visible back pr ojected portions by the jaw openings on the annulus source and disk source, respectively. The fact that the fluence is derived by integrating each source distribution eliminates th e singularity problem posed by E quation 4-1 automatically since its integral is always finite. Dual-source model. The dual-source model proposed by Jiang et al79 consists of the primary source and one extra-focal radiation so urce. The primary photons are modeled by a point source located at the positi on of the X-ray target. The extra-fo cal radiation originates only from the flattening filter and it is approximated by the sum of a series of Gaussian functions at the bottom of the flattening filter. The backscatter radiation into the monitor chamber is modeled explicitly. A linear relationship be tween the backscatter radiation and the irradiated area on the jaws upper surface is assumed. So the in-air output factor Sc for a certain field size fs could be calculated as the product of two terms, ))(1())(1()( fsFfsFfsSmbs efs c (4-3) where the first term describes co ntribution from scattering source (Fefs) and second term accounts for decrease due to monito r backscatter effect (Fmbs). The scattering contribution is derived by back-projection through detectors eye view and integration in a similar fashion as in three source model. Single-source model. A numerical single source model was used by Chui et al75 for IMRT planar dose calculation. All the primary and scatte ring photons were considered to originate from the source plane at the X-ray target for the sa ke of simplicity. The rotationally symmetric source 66

PAGE 67

distribution peaks at the central axis and monotonically decreases with off-axis distance. The source distribution was represented by a numerical form and samples from the source distribution were obtained by fitting the head scatter factors of square fields with different field sizes. The full source distribution was interpolated from these samples. Backscattered radiation was not considered in this model. The head sca tter factor is simply cal culated by back-projection and integration on the visible source portion through detectors eye view. It is noticed that the numerical source is very sensitive to the error in output factor measurement. To make the source model more st able, we propose an analytic function to approximate the numerical one as follows, d r c r rIba 1 ) ( (4-4) where I(r) is the source intensity at a point with radial di stance r from the central axis, and a, b, c and d are free parameters. In the original implementation of the three-source and single-source m odels, the effect of monitor chamber back scatter was not modeled exp licitly. This effect could be incorporated for possibly improved fitting to the measurement data. Therefore, we have also attempted to modify the original three-source and single-source models by explicitly incorpor ating the effect of monitor chamber back scat ter using Equation 4-3. Planar Dose Calculation Algorithm The algorithm is based on a generalized dose calculation procedure. The flow chart is illustrated in Figure 4-1. Since planar dose distri bution at isocenter plane with 10 cm of waterequivalent buildup is used for IMRT quality assu rance at our center, we aim at calculating the 40x40 cm2 dose distribution with 1 mm resolution under the same setting. The parameters that defined the source models were determined from fits to measured Sc data. Back-projection was 67

PAGE 68

done for each point of calcul ation through the combination of jaw opening and MLC opening from detectors eye view. In-air fluence dist ribution was obtained by integrating the visible portion of each sub-source component. Correction to horn effect caused by the flattening filter was then applied to the fluence distribution. A pencil beam kernel, generated from Monte Carlo simulation but modified to fit the in-phantom cr oss-beam profiles for several regular fields, was convolved with the in-air fluen ce distribution. The resulting relative dose distribution was converted to the absolute dose distribution through the effective field size method. The algorithm is similar to the one described by Chui et al75. Several modifications were applied: (1) In-air fluence distribution is derived based not only on the MLC opening, but the combined openings of jaws and MLC. (2) Off-axis fluence variation due to flattening filter is correc ted by using cross-beam in-air scans with appropriate buildup. (3) Pencil beam kernel is expressed as the sum of three Gaussian functions and the free parameters were obtained by fitting the in-phantom beam profiles of several regular square field sizes at the same plane as the dose calculation plane. The input data for commissioning the algorithm in clude the following: in -air output factors, in-air scan along the diagonal for the largest fiel d size, square field beam profiles in water and output factors in water at the depth of interest. Calculation of fl uence distribution. In model-based dose calculation algorithms, the subsource components are considered iso-tropical within a certain solid angle79. Thus, the fluence at any point of calculation can be calculated as the integration over the visible portion of each subsource components from the calculat ion points eye view as descri bed before. In step and shoot 68

PAGE 69

mode, the total in-air fluence (x, y) at point (x, y) for an IMRT field is the sum of the contribution from each segment, and is given by K k kkyxfMUyx1),( ),( (4-5) where MU is the total monitor unit delivered for the field, K is the total number of segments within this field, fk is the fractional monitor unit for segment k and k(x, y) is the fluence delivered in segment k, given by dIyxS s s ks 1)(), ( (4-6) where S is the total number of sub-source co mponents in the head scatter source model, s is the visible back projected portion on su b-source s from point (x, y), and Is the source intensity distribution on sub source s. When determining the visible portion from the point of calculation, we have to consider the combined opening of the jaws and MLC when doing back-projection onto the source plane. One way of achieving this is to back project the MLC and jaw openings separately onto the source plane as jaw and MLC, respectively, and then find their intersection. Thus s can be expressed as MLC jaw s (4-7) This process is computationally time-consumi ng. Thus the more source planes (sub source components) there are in the s ource model, the more computer time is needed to calculate delivered fluence distribution. The inter-leaf an d mid-leaf transmission, tongue and groove effect, rounded leaf end of MLC and jaw transmission were taken into account while determining jaw and MLC, using the methods described by Chui et al.75 69

PAGE 70

Horn effect correction. It is well known that the flattening filter causes horn effect on the beam profile perpendicular to the beam centr al axis. The central portion of the beam suffers more attenuation which results in an increased nu mber of photons at lower energy with off axis distance. The combined effect is the horn effect on the cross beam profiles. To correct for this effect, an in-air scan with appropriate buil dup was done along the diagona l of the largest open field (40x40 cm2 for the Varian Trilogy). The horn effect correction factor, or the off-axis ratio (OAR), is the ratio between this in-air scan scan and the corresponding calculated calc from the source model (ref Figure 4-2) and is expressed as )( )( )( r r rOARcalc scan (4-8) where r is the radial distance from the point of calculation to the beam central axis. This correction factor was applied to in-air fluence calculated for any IMRT field. Pencil beam kernel generation. The dose distribution is calculated by pencil beam convolution75 '')','()','(),(),(),( dydxyyxxhyxyxhyxyxD (4-9) where (x, y) is the delivered fluenc e distribution and h(x, y) the pencil beam kernel. The pencil beam kernels are typically obtained from Monte Carlo simulation for different nominal energy beams. Monte Carlo simulation considering the full energy spectrum of the nominal energy beam will result in a polyenergetic pencil beam kernel. Usually, photon beams with the energy spectrum of the initial bremsstrahlung photons em itted from X-ray target are employed in the Monte Carlo simulation. However, the energy spectrum of treatment photon beams undergone multiple scattering differ from their initial stat us. Thus the pencil beam kernel derived from 70

PAGE 71

Monte Carlo simulation has to be adjusted or The sum of three Gaussian functions was us ed to approximate the pencil-beam kernel, customized to account for the spectrum change. 3 1 22222i iAwhere Ai and i are the amplitude and sta ndard deviation for the ith Gaussian, respectively. Monte Carlo generated pencil beam kernel was used to acquire the initial values for the free parameters, Ai and i, which were then fine-tuned by fitting the model to the in-plane and crossplane beam profiles of several square fields (e.g., 10x10, 20x20, 30x30 and 40x40 cm2) on isocenter plane. The beam profiles were measured with a 0.2 mm radius ion chamber (C deconvolved using the algorithm in Cha2/)(),(yx iie yxh (4-10) the C13) and pter 3.18 The free parameters in the formula were adjusAbsolute dose conversion. Using Equation 4-5 distribution Dc( ) could be expressed as (4-11) the iso dose m tput fact or with field opening Se by and respectively, the calculated relative dose distribution Dc( ) could be converted to absolute dose distribution Da( ) as following, ted such that the calculated profiles (via convolution) ma tched the corresponding measurements in a least-square sense. and 4-9, the calculated relative dose K k kk K k kk ch fMUh fMU D1 1))(( )( )( It is the summation of the contribution from each segment within one IMRT field. If we denote centereasured and calculated with field opening Se by Dm(se) and Dc(se) respectively, and denote measured and calculate d relative ou)(e m cpSS )(SSe c cp 71

PAGE 72

h fMU h sS sS f sD sD MU sSsD sSsD h fMU sD sD h fMU DK k kk k k e c cp k e m cp k ref ec ref em K k K k k e c cp ref ec k e m cp ref em kk K k k ec k em kk a 1 1 1 1)('' )( )( )( )( )( )()( )()( ))(( )( )( ))(( )( (4-12) where and are the effective square field sizes formed by the MLC shape k and the reference field respectively. They are calcu lated according to a method described by Dong et al.73 We use and to denote the modified total and fractional monitor units, respectively. The last step in Equation 4-12 shows the convers ion can be done by modifying monitor units and the contribution to in-air fluence from each segment can still be added up and only one convolution (i.e., the one between to tal in-air fluence and the penc il beam kernel) is needed for each IMRT field. k esref es'MU'kfComparison for Head Scatter Calculation The three different source models were implemented and were first applied to their original data to ensure correct implemen tation. The models were then applied to a comprehensive set of Sc measurement data with a 6 MV photon beam collected on a LINAC with the 120-leaf Millennium MLC (Trilogy, Varian Oncology Sy stems, Palo Alto, CA). A cylindrical ion chamber (IC-10, Wellhofer Dosimetrie, Germany) in a 4-cm diameter cylindrical mini-phantom at 10 cm depth was used for the measurement. The measurement geometries were very similar to those of Jiang et al .76 Two groups of data were measured fo r the 6 MV photon beams at isocenter. All calculated and measured in-air outputs were normalized to th e reference condition, i.e., at isocenter with 10x10 cm2 field size. The first group incl uded: (1) in-air output factor Sc for 72

PAGE 73

square fields from 4x4 cm2 to 40x40 cm2; (2) Sc for symmetric rectangula r fields with 10 cm Xjaw (lower jaw in the cross-plane direction) setting and Y-jaw settings from 4 to 40 cm; (3) Sc for symmetric rectangular fields with 10 cm Y-jaw setting and X-jaw settings from 4 to 40 cm. The entire group data was used to obtain the fitting coefficients for the dual-source model. For the other two source models, only square field Sc was needed during fitting and the rest of the Sc was used for verification purpose. The second group wa s used for additional verification for each of the source models and included two sets of asym metric rectangular fields, i.e., X jaws set to symmetric 10 cm and 40 cm and Y2 jaw set to 20 cm while changing Y1 jaw from 2 cm to 20 cm. Comparison for IMRT Dose Calculation The main goal of this work is to compare di fferent source models for the purpose of planar dose calculation for independent IMRT plan verification. Each of the source models was plugged in the planar dose calculation algorithm de scribed above. The calculated planar dose distributions were compared with measurements done with a 2-dimensional (2D) diode array (MapCHECK 1175, Sun Nuclear Corp., Melbourn e, FL). The MapCHECK device has been demonstrated to be an accurate and reliable tool for IMRT treatment plan verification38, 88. An intercomparison between film a nd MapCHECK showed that MapC HECK can effectively replace film dosimetry in routine IMRT QA40. Seven clinical cases (3 brai n and 4 H&N cases for a total of 53 treatment fields) were se lected for comparison, all deliv ered on the Trilogy with 6 MV nominal photon beams and with gantry angle set to 0o (IEC convention). Planar dose distributions were calculated w ith each of the different source models and compared to measurements done in a rectangul ar solid water phantom at 10 cm depth and 100 cm source-todetector distance (SDD). All the ca lculation was done using a 1.0x1.0 mm2 dose grid. The MapCHECK device was calibrate d for absolute dose distribution measurement based on the 73

PAGE 74

manufacturer-recommended procedures. Quantitative comparison between the calculation and MapCHECK measurements was done using the MapCHECK analysis software (version 2.01) in the absolute dose comparison mode. The calculated planar dose distributi ons were imported into the MapCHECK analysis software. Each of th e MapCHECK points above 5% of the maximum dose level was compared with calculation using percent dose difference (%Diff) and distance-toagreement (DTA) acceptance criteria. Two sets of criteria, 2%/2mm and 3%/3mm, were employed, and the percentage of the points pa ssing the acceptance criteria was evaluated. Results Comparison of Head Scatter Factors The results of the three different source models in their fit to the head scatter factor data from a Varian Trilogy LINAC is presented in Fi gure 4-3, Figure 4-4 and Figure 4-5. The dualsource model achieved the best fitting with stan dard deviations of 0.05%, 0.08%, and 0.30% for the three data sets in Figure 4-3 (a)~(c), resp ectively. The original three-source and single-source models without explicit modeling of the effect of the monitor b ack scatter were less satisfactory in fitting the measured Sc data, with standard deviations of 0.14%, 0.23%, and 0.54% for the three data sets in Figure 4-4 (a)~(c), respectiv ely, for the three-source model and 0.10%, 0.22%, and 0.48% for the three data sets in Figure 4-5 (a)~(c), respectiv ely, for the single-source model. These two models underestimated the head scat ter factors for certain symmetric rectangular fields and asymmetric fields. The discrepancy is especially large for symm etric rectangular fields with variable upper jaws and fixed lower jaw openi ng at X= 4 cm, with deviations about 0.4% at large field sizes (Figure 4-4 (b) and Figure 4-5 (b)). For asymmetr ic fields, the calculation is systematically lower with average deviations of 0.5%. However, these two models were improved significantly when the effect of the mon itor back scatter were modeled explicitly as in the dual-source model, with standard deviations of 0.07%, 0.12%, and 0.33% for the three data 74

PAGE 75

sets in Figure 4-4 (d)~(f), respectively, for th e three-source model and 0.07%, 0.21%, and 0.30% for the three data sets in Figure 4-5 (d)~(f ), respectively, for the single-source model. Comparison of Dose Calculation Figure 4-6 shows the comparison between meas ured and calculated dose profiles for a 10x10 cm2 field. The calculation was done using the single-source model with either a Monte Carlo-generated kernel or an experimentally -determined kernel. Using the experimentallydetermined kernel noticeably im proved the agreement between calcu lated and measured profiles, especially around the shoulder and foot region. Sim ilar results were also achieved with the other two source models. The passing rates between the measured and cal culated planar dose distributions for the 7 clinical cases (53 treatment fi elds) using both the 2%/2mm and 3%/3mm criteria are summarized in Figure 4-7, where the passing rates were binned into 5% steps for the three different sourcemodels, and Table 4-1 lists the average passing rates for all th e cases. The single-source model gave better passing rates than the other two source models when evaluated with the 2%/2mm criterion, with an average passing rate of 96.2% compared with average passing rates of 90.8% and 89.7% for the dual-source and three-source models, respectively. For the single source model, almost all the cases ach ieved passing rates of > 90%, while about half of the cases had passing rates less than 90% for the other two models. However, comparable average passing rates (>97%, see Table 4-1) were achieved for al l the source models if the criterion of 3%/3mm was employed. Figure 4-8 shows an example of th e calculated planar dose distribution using the single-source model and a comparison between the measured and calculated profiles with different source models. In this case, the passing rates were 94.5%, 83.9%, and 83.5% using the 2%/2mm criterion and 97.5%, 96.5%, and 96.2% using the 3%/3mm criterion for the singlesource, dual-source, and threesource models, respectively. 75

PAGE 76

Computation Time Computation time is another concern for a ny dose calculation algorith m to be used in clinic. As described above, the energy depos ition process is isolated from the fluence determination in the dose calculation algorithm. For each of the source models, the energy deposition process takes the same amount of com putation time, i.e., the time to do a convolution between pencil beam kernel and a two dimensiona l fluence distribution. However the time spent on fluence determination through back-projection and integration varies from one source model to the other depending on how many sub source planes the source model has. The three-source model takes twice as much time as the other tw o source models (4 seconds vs. 2 seconds on a Pentium 1.66 GHz laptop computer with 2 GB of RAM) to calculate fluence distribution in air on average for one IMRT field due to its additi onal sub source component. The dose deposition, described by the pencil beam convolution, take s the same amount of time (2 seconds) while implemented using Fast Fourier Transform. The average total time for each IMRT field is about 6 seconds for the three-source model and 4 seconds for the other two models. Discussion Both single-source model and three-source mode l underestimate the head scatter factors in some cases as shown in Figure 4-4 b & c and Figu re 4-5 b & c, especially when the lower jaws (X jaws) are fixed and upper jaws increase to a maximum opening. This could be explained by the shape of the source distribu tion (see Figure 4-9 a). Each of th e three analytical source models uses either exponentially decaying functions or fast decaying reciprocal function to model the extra-focal source distribution. When the jaw ope ning keeps increasing above a certain field size (10 cm in Figure 4-4 b), the increase in contribu tion from integrating larger visible area around the tail region of the source di stribution becomes insignificant. This becomes worse when fixing the lower jaw opening and increasing the upper jaw opening, where the output factors increase 76

PAGE 77

much faster (see Figure 4-4 b) while large portion of the sour ce is blocked by the lower jaws. From this reasoning, all three source models would fail in predicting the head scatter factors in those cases. However, the major difference be tween the dual-source model and the other two source models is that the former isolates and models the monitor chambe r back scatter effect separately while the other two don t. The reduced back scatter cont ribution at large field sizes, which increases the linac output, makes up the fast decaying source. A linear relationship between the backscatter radi ation and the irradiated area on the jaws upper surface is appropriate if one notices the a pproximately linear increase in Sc curve with large upper jaw opening (see Figure 4-4 (b)). By modeling this e ffect explicitly, the dual-source model achieved better agreement between measured and calculated head scatter factors than the other two. On the other hand, significant improvement is observe d when one incorporates this relationship into the other two source models. These facts show it is desirable to isolate the monitor chamber back scatter effect when using analytical functions to model head scatter fact ors. This conclusion is consistent with the finding in Ding84. Fast decaying analytical functions can predict the fast increasing in Sc at small field size while nearly linear increase of Sc at large field size is better modeled by modeling the monitor chamber back scatter explicitly. Since the machine output varies from one LINAC model to another, even from the same manufacturer, the behavior of Sc at large field size differ significantly. For exam ple, when fixing the lower jaw to 4 cm and moving the upper jaw to an opening larger than 30 cm, the head scatter factor at isocenter from Varian high energy machine (Clinac 2100/2300 and Tril ogy) is about 1% larger than Varian low energy machine (Clinac 600 C/D).78 The Sc curve from the low energy machines tends to flatten out at large field sizes, in whic h case it can be fit equally well with or without the explicit modeling of the mon itor back scatter. 77

PAGE 78

IMRT fields are characterized by many small fi eld segments and sharp gradients. All three photon source models studied in this work have comparable performance in modeling output factors for small fields. Thus the performance in modeling penumbra regi on dictates the IMRT dose calculation accuracy. In IMRT patient-specific QA with either diode array or film, most of the failing points occur in the high dose gradient regions where the modeling of the penumbra is less accurate. In some dose calculation algorith ms, calculated fluence or dose profiles are convolved with a Gaussian function to account for the finite source size73, 74. The width of the Gaussian is independent of the off-axis posi tion. This causes problems since the contribution from the extra-focal source varies with the offaxis positions. In our algorithm, the contribution to the penumbra dose from scattering source is calculated from the extra-focal source based on the off-axis position. The calculated penumbra is fit to the corres ponding measurement at different off-axis positions by fine-tuning the pencil beam kernel. Figure 4-9 (b) shows the comparison between the measured dose profile for a 10x10 cm2 open field and the horn effectcorrected in-air fluence profiles calculated from each of the three analytic source models. The convolution between the fluence profiles and the pencil beam kernel resu lts in the calculated dose profiles. The faster decaying source distribu tions of the dual-source model and three-source model produce sharper corners in the penumbra regi on, while the calculated fluence profile from the single-source model has rounded shoulder and foot regions which, when convolved with the kernel, agree better with the m easured dose profile. This probab ly explains the better agreement between calculated and measured dose distributi ons with the single-source model than with the dualor three-source models. Patient-specific IMRT plan QA using indepe ndent MU calculation algorithms have been reported by many groups76, 82, 89-95 and are commercially available. Most of these algorithms 78

PAGE 79

verify dose to one or a few points. Due to the dy namic nature of IMRT delivery, the correctness of the dose at one spatial point dose not warrant the correctness of dose at other points. It is therefore necessary to verify fl uence/dose distribution for each IMRT field to ensure the derived time sequence of MLC from the TPS is correct75, 90, 95. Recently, patient-specific IMRT QA using 2D detector arrays, such as diode arrays or ion chamber arrays, has become increasingly popular due to their ease of use and immediate readout of the resu lts. Measurements are usually done in regular phantom with radiation beam orthogonal to phantom surface. Such measurements can potentially be replaced using independent calculation algorithms such as the one proposed in this work. Using computational algorithms to replace ph antom measurement as patient-specific IMRT QA has the drawback that it does not verify the functionality of the MLC delivery system for each field, and a supplementa l QA of the MLC delivery system needs to be in place. It has been proposed that the QA of the MLC leaf sequences (fluence maps) of a patient treatment and the dynamic MLC delivery system can be separated into two different tasks89, 95. A rigorous QA procedure to examine the mechanical and electronic reliability of the MLC delivery system must be established and their dosimetri c implications must be well understood. The frequency and extent of the QA of the MLC de livery system, however, ar e independent issues and should be determined by the need to maintain the normal operation of dynamic delivery system. This may not require an actual delivery ch eck for every IMRT treatment field, similar to that we do not check the dynamic jaw movement for every treatment field involved with a dynamic wedge in conventional radi otherapy treatment. The verifica tion of the dose distribution, on the other hand, should be done fo r every field in a patient treatme nt to ensure the accuracy of the planar dose distribution. This separation of QA tasks allows us to use computer simulation to 79

PAGE 80

examine the functionality of the MLC leaf seque nces and the delivered fluence/dose map based on a validated calculation algorithm. The other important aspect that needs to be verified if an indepe ndent MU verification algorithm is to be used in place of phantom measurem ent is the integrity of the data transfer from the TPS to the record-and-verify (R&V) system which controls the linac delivery. In our implementation, the algorithm read in the expor ted plan files from our R&V system (MOSAIQ version 1.2018, IMPAC Medical Systems, Inc., M ountain View, CA), rather than from the TPS. In this way, treatment parameters that are stor ed in the R&V system and are used directly for patient treatment are verified and any data tran sfer errors between the TPS and R&V could be detected. Conclusions The performance of three different source mode ls regarding their ability to model head scatter factor and facilitate planar dose cal culation for IMRT has been evaluated. The comparison was done on an extensive set of meas urement data from a 6 MV photon beam of a Varian Trilogy and showed that the dual-source model predicted the head scatter factor most accurately. The original three-source model and single-source model underestimated head scatter factors for asymmetric fields and certain symmetric rectangular fields, but similar good agreement could be achieved when monitor back scatter effect was incorporated explicitly. Each of the source models was used to calculate fluence distribution in-air, which when convolved with an experimentally-determi ned kernel, resulted in the planar dose distribution in a homogeneous phantom. The comparison on passi ng rates using percent dose difference and distance-to-agreement criteria showed the single source model achieved the best agreement with measurement when using the 2%/2mm criterion. When the criterion was relaxed to 3%/3mm, all source models achieved comparable passing rates (>97%). Due to the simplicity of the models, 80

PAGE 81

single-source and dual-source models required less computational time on in-air fluence distribution and overa ll calculation time. 81

PAGE 82

Table 4-1. The comparison of average passing rate s for the 53 IMRT fields while using different source models in the planar dose calculati on algorithms. The standard deviation is also given for each case. Criterion Single-source model (%) Dual-source model (%) Three-source model (%) 2 %/2 mm 96.2.8 90.8.4 89.7.7 3 %/3 mm 99.0.2 97.7.1 97.4.4 Horn correction Kernel customization Back-projection & integration Head scatter Source model Horn correction factor In-air fluence Monte Carlo generated kernel Customized Pencil beam kernel Horn corrected in-air fluence Convolution Absolute dose conversion Planar dose distribution Figure 4-1. Flowchart of the proposed IMRT planar dose calculation algorithm. 82

PAGE 83

0 10 20 30 0.4 0.6 0.8 1.0 1.2 off axis distance (cm)relative intensity In-air scan Calculated fluence OAR factor Figure 4-2. Comparison of the calculated fluence as a function of off-axis distance before correcting for the horn effect, the meas ured fluence from in-air scan, and the calculated OAR factor, which is the ratio between the in-air s can along the diagonal direction and the corresponding calculated fluence profile. 83

PAGE 84

0 5 10 15 20 25 30 35 40 0.95 0.97 0.99 1.01 1.03 1.05 side of field (cm)head scatter factor Square field X=10 cm Y=10 cm (a) Calculated Measured 0 5 10 15 20 25 30 35 40 0.94 0.95 0.96 0.97 0.98 0.99 1.00 side of field (cm)head scatter factor X=4 cm Y=4 cm (b) Calculated Measured 0 2 4 6 8 10 12 14 16 18 20 0.98 0.99 1.00 1.01 1.02 1.03 1.04 1.05 Y1 jaw position (cm)head scatter factor X=10 cm X=40 cm (c) Calculated Measured Figure 4-3. Comparison between the measured and calculated h ead scatter factors using the dual-source model for a Varian Trilogy 6 MV photon beam for three different sets of fields: (a) square fields from 4x4 cm2 to 40x40 cm2 (Square field), symmetric rectangular fields with lower jaw opening fixed to 10 cm while upper jaws changed from 4 cm to 40 cm (X=10 cm), and symmetric rectangular fi elds with upper jaw opening fixed to 10 cm while lower jaws moving from 4 cm to 40 cm (Y=10 cm). These data were used to fit the model pa rameters; (b) symmetric rectangular fields with lower jaw opening fixed to 4 cm while upper jaws changed from 4 cm to 40 cm (X=4 cm), and symmetric rectangular fiel ds with upper jaw opening fixed to 4 cm while lower jaws moving from 4 cm to 40 cm (Y=4 cm); (c) asymmetric rectangular fields with symme tric X jaws set to either 10 cm or 40 cm and Y2 jaw fixed at 20 cm while changing Y1 jaw fr om 2 cm to 20 cm. All the data were normalized to the results at 10x10 cm2 field size. 84

PAGE 85

0 5 10 15 20 25 30 35 40 0.95 0.97 0.99 1.01 1.03 1.05 side of field (cm)head scatter factor Square field X=10 cm Y=10 cm (a) Calculated Measured 0 5 10 15 20 25 30 35 40 0.94 0.95 0.96 0.97 0.98 0.99 1.00 1.01 side of field (cm)head scatter factor X=4 cm Y=4 cm (b) Calculated Measured 0 2 4 6 8 10 12 14 16 18 20 0.98 0.99 1.00 1.01 1.02 1.03 1.04 1.05 1.06 Y1 jaw position (cm)head scatter factor X=10 cm X=40 cm (c) Calculated Measured 0 5 10 15 20 25 30 35 40 0.95 0.97 0.99 1.01 1.03 1.05 side of field (cm)head scatter factor Square field X=10 cm Y=10 cm (d) Calculated Measured 0 5 10 15 20 25 30 35 40 0.94 0.95 0.96 0.97 0.98 0.99 1.00 1.01 side of field (cm)head scatter factor X=4 cm Y=4 cm (e) Calculated Measured 0 2 4 6 8 10 12 14 16 18 20 0.98 0.99 1.00 1.01 1.02 1.03 1.04 1.05 1.06 Y1 jaw position (cm)head scatter factor X=10 cm X=40 cm (f) Calculated Measured Figure 4-4. Comparison between the measured and calculated h ead scatter factors using the original three-source model (a-c), and the modified three-source model to incorporate the monitor back scatter effect explicitly (d -f). The data sets were the same as in Figure 3. In the original three-source model (a-c), only the square field data in (a) were used for the model fitting, while with the modified three-source model, both the square field data and the rect angular field data in (d) were used for the model fitting. 85

PAGE 86

0 5 10 15 20 25 30 35 40 0.95 0.97 0.99 1.01 1.03 1.05 side of field (cm)head scatter factor Square field X=10 cm Y=10 cm (a) Calculated Measured 0 5 10 15 20 25 30 35 40 0.94 0.95 0.96 0.97 0.98 0.99 1.00 1.01 side of field (cm)head scatter factor X=4 cm Y=4 cm (b) Calculated Measured 0 2 4 6 8 10 12 14 16 18 20 0.98 0.99 1.00 1.01 1.02 1.03 1.04 1.05 Y1 jaw position (cm)head scatter factor X=10 cm X=40 cm (c) Calculated Measured 0 5 10 15 20 25 30 35 40 0.95 0.97 0.99 1.01 1.03 1.05 side of field (cm)head scatter factor Square field X=10 cm Y=10 cm (d) Calculated Measured 0 5 10 15 20 25 30 35 40 0.94 0.95 0.96 0.97 0.98 0.99 1.00 1.01 side of field (cm)head scatter factor X=4 cm Y=4 cm (e) Calculated Measured 0 2 4 6 8 10 12 14 16 18 20 0.98 0.99 1.00 1.01 1.02 1.03 1.04 1.05 1.06 Y1 jaw position (cm)head scatter factor X=10 cm X=40 cm (f) Calculated Measured Figure 4-5. Comparison between the measured and calculated h ead scatter factors using the single-source model without explicit incorporation of the monitor back scatter effect (a-c), and the single-source model with the monitor back scatter effect incorporated explicitly (d-f). The data sets were the same as in Figure 4-3. In the first case (a-c), only the square field data in (a) were used for the model fitting, while in the second case (d-f), both the square fiel d data and the rectangular fiel d data in (d) were used for the model fitting. 86

PAGE 87

-10 -8 -6 -4 -2 0 2 4 6 8 10 0 0.2 0.4 0.6 0.8 1.0 off axis distance (cm)Relative intensity Measured Customized Monte Carto Figure 4-6. Comparison between the measured and calculated cross beam profiles at 100 cm SDD and 10 cm water-equivalent depth for a 10x10 cm2 field. The calculation was done using the single-source model combined with either a Monte Carlo generated kernel or an experimentally-fitted kernel. 70 80 90 100 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 passing rate (%) with 2%/2mm criterionfraction (a) SSM DSM TSM 70 80 90 100 0 0.2 0.4 0.6 0.8 1.0 passing rate (%) with 3%/3mm criterionfraction (b) SSM DSM TSM Figure 4-7. Comparison of the passing rates between the MapCHECK-measured and calculated planar dose distributions for 7 clinical cas es (53 treatment fiel ds) using the singlesource model (SSM), dual-source m odel (DSM) and three-source model (TSM). Comparison was done using the MapCHECK analysis software (version 2.01) in the AD mode w ith both the 2%/2mm (a) and 3%/3mm (b) DTA criteria. The passing rates were binned into 5% steps. 87

PAGE 88

Figure 4-8. Example of a calcu lated planar dose distribution us ing the single-source model (a) and its in-plane and cross-plane profiles compared with the profiles from MapCHECK-measured and calculated prof iles using the dual-source model (DSM) and the three-source model (TSM) (b). The positions where the profiles were taken are indicated by the arrows. The cross-plane (horizontal arrow) profile was scaled by a factor of 1.5 to make the tw o profiles more distinguishable. -8 -6 -4 -2 0 0 0.2 0.4 0.6 0.8 1.0 off axis distance (cm)relative profile intensity (%) (b) Dose SSM DSM TSM 0 2 4 6 8 10 12 14 16 18 20 10-10 10-8 10-6 10-4 10-2 100 off axis distance (cm)relative source intensity (%) (a) SSM DSM TSM-1 TSM-2 Figure 4-9. (a) Source intensit y distribution of the modified single-source model (SSM), the dual-source model (DSM) and the threesource model (TSM-1 and TSM-2) modeled for Varian Trilogy. TSM-1 and TS M-2 represent the annulus source for the primary collimator and the disk source for the flattening filter, respectively (b) Comparison between measured dose profile (Dose) in a water phantom for a 10x10 cm2 field at 10 cm depth and horn effect-corrected in-air fluence profiles calculated using the single-source model (SSM), the dual-source model (DSM) and the three-source model (TSM). In-air flue nce profiles for DSM and TSM overlap each other. The relatively slower decaying s ource of the single-source model results in rounded profile corner. 88

PAGE 89

CHAPTER 5 THE CALIBRATION OF A NOVEL 4D DIODE ARRAY Introduction It has become standard practice to use planar two dimensional (2-D) detectors (film, diode array, ion chamber array and elec tronic portal imager) for IMRT QA. Due to the planar design of those detectors, IMRT beams are usually delivered with fixed gantry angle (0o, IEC convention) and beam incident direction perp endicular to the detector plane38. One obvious limitation of this QA solution is that potential dosimetric and mu lti-leaf collimators (MLC) positioning errors during gantry rotation cant be detected. The emergence of innovative rotational radiotherapy delivery techniques such as dynamic arc ther apy, helical TomoTherapy and volumetric modulated arc therapy (VMAT) poses greater challenge for an appropriate QA solution. Rotational delivery modalities are characterized by many time-varying parameters, such as dynamic MLC, dynamic gantry speed and variable dos e rate. In the pre-treatment verification of these radiotherapy modalities, dosimetric measur ement would be ideally performed in arc mode. An ideal dosimeter for treatment verification wo uld provide absolute dos e distribution in a full three dimension (3-D) space96. Individual beam dosimetry shoul d be available at the end of measurement to identif y sources of error. Gel dosimetry would be an appropriate candidate to fulfill the requirement for a full 3-D dose distribution. Very encouraging dosimetri c measurements have been shown with Gel dosimetry97-100. However, the need of expensive magne tic resonance imaging system prevents the wide use of Gel dosimetry in practice. Film in combination with dedicated cylindrical phantom with spiral cavity has been used to sample and measure dose distribution from 3-D subspace101. Either radiographic or radiochromic film can be inserted in the spiral cavity. The design gives the system the advantage that beam always comes perpendicular to the film at any 89

PAGE 90

gantry angle which makes it suitable to validat e IMRT beams with planned gantry angle and rotational delivery modalities. On the other hand, film dosimetry is a very time-consuming procedure and great caution has to be taken to achieve accurate absolute dosimetry. Many user friendly on-line 2-D detector a rrays have become commercially available with which absolute dosimetry can be conveniently performed.38, 102, 103 These detectors are designed to sample and measure dose within a single plane perpendicular to beam incident direction with fixed gantry. Dedicated phantom has been made to accommodate these 2-D detectors for the use of rotational radiotherapy QA104. Usually, 2-D detector in combinat ion with dedicated phantom requires special corrections and results in limited accuracy with oblique beam incidence. In addition, since film and most on-line detect or arrays are integral dosimeters88 and only a composite dose distribution is available at the end of the measurement, correcti on for detector angular response difference is not possible with varying gantry angles. Individual beam dosimetry would be required to identify the sources of errors in the planning and delivery process especially for rotational delivery modalities with many time-varying parameters. When dosimetric error presents, integral dosimeters cant provide ad equate information for identifying the cause. Clearly, a desired dosimeter woul d be able to consistently sa mple dose distribution from 3-D subspace and capable of performing individual beam dosimetry to validate radiotherapy plans and separate sources of error. A novel four dimensional (4-D) diode array has been developed for this purpose (ArcCHECKTM, Sun Nuclear, Melbourne, FL). The prot otype of the 4-D diode array contains 124 diodes embedded in a cylindrical acrylic pha ntom. The diodes are arranged to sample and measure dose distribution in 3-D subspace. Di ode signal is updated every 50 ms (the 4th dimension) which makes both composite and pe r beam analysis possible for any complex 90

PAGE 91

rotational delivery. The unique ar rangement makes the diode array present a consistent detector image to beams eye view at arbitrary gantry an gle. At any beam incident direction, entrance dose and exit dose in the cylindrical phantom w ill be measured simultaneously. The objective of this work is to calibrate the 4-D diode array for IMRT QA with planned beam angle and VMAT QA. Diode sensitivity and angular response dependence will be accounted for to achieve rotational isotropicity. Algorithm s to derive instant beam angl es will be developed based on detector signal to interpolate corresponding angular correction f actors for each detector update. The calibration will be validated using water tank scanned beam profiles. Methods and Materials Description of the 4-D Diode Array A prototype of the novel 4-D diode array is shown in Figure 5-1(a). The diode array consists of 124 diodes embedded in a cylindrical acrylic phantom with 13.7 cm height, 25.4 cm outer diameter and 15 cm inner diameter. Th e cylindrical phantom can accommodate a 15 cm diameter inner core composed of various homogeneous or heterogeneous materials. In this work, an inner core made from acrylic was inserted to make a homogeneous phantom. The 124 diodes are arranged on four rings of the same diameter (19.6 cm), giving each diode 3 cm intrinsic acrylic build up. Each ring has 31 evenly distribu ted diodes as can be easily identified on the CT image (Figure 5-1 b). The four rings are placed in the center with 1 cm spacing along the axis of the cylinder. Positions of diodes on adjacent rings are slightly offset to form a spiral configuration. N-type diodes with 0.8x0.8mm2 active measuring area are used in the diode array. All the diodes are mounted on two flexible pr inted circuit boards (P CB). Diode signals are conducted from the two PCBs to two proprie tary designed 64-cha nnel analog-to-digital electrometers. Diode signal is updated every 50 ms. Collected charge in the capacitor is emptied during each signal update such that the detector will never saturate. The fact the diode array 91

PAGE 92

samples and measures dose distribution in 3-D subspace and its capability of real time signal update make it a real 4-D detector. Unless specified otherwise, the axis of the cylinder was always aligned with gantry rotation axis and center of the cylindrical phantom aligned with iso-center during all the measurements in this work. The source to surf ace distance (SSD) was 87.3 cm. Position of each diode was carefully marked on the outer surface of the cylinder and a plastic stand was made to support the phantom which provided setup reproducibility. A positive charge was applied to compensate the diode signal leakage during th e measurement. Background signal was measured and monitored during each measurement session and subtracted from collected diode signal. Diode Array Calibration N-type silicon diodes are used in the diode arra y. A silicon diode is a p-n junction diode. A p-n junction is formed by combining P-type and N-type semiconductor materials in close contact (Figure 5-2). Diodes are referred as n-type diode or ptype diode depending on the base material. The opposite thinner layer material is referred as doping material. The close contact between these two materials is called depletion region. Ra diation will create electr on hole pairs inside the body of the diode. Those pairs within diffusion re gion will diffuse into depletion region. An electric field will be applied to collect those ch arges. Diode response is very sensitive to its construction. A small difference in the position of p-n junction could cau se significant difference in its response. Due to the anisotropic desi gn of p-n junction diode, radiation coming from different direction experiences different amount of attenuation and back scatter caused by the high-Z material surrounding the diode and metal cont act used to collect charges. This effect is referred as diode directional re sponse dependence. Diode respons e also depends on temperature, dose rate and radiation energy spectrum. Howeve r, diode sensitivity and directional response dependence are most dominating factors. In this chapter, we will discuss diode calibration, 92

PAGE 93

focusing on correcting the effect caused by these tw o factors. A few notations will be introduced for this purpose. Relative Gantry Angleto diode m: angle formed by two lines, one connecting isocenter and diode m, the other connecting isocen ter and radiation source (Figure 5-3 a) R ( m,): response of diode m at relative gantry angle SD ( m ): intrinsic diode sensitivity of diode m ; SA ( m,): angular response se nsitivity of diode m at relative gantry angle It is normalized at = 0o 1)0,( mSA (5-1) D (): absolute dose at the position of the diode of interest at relative gantry angle Response of diode m with relative gantry angle can be expressed as: )(),()(),( DmSAmSDmR (5-2) CFD( m ): correction factor for diode intrinsic sensitivity for diode m; CFA( m,): correction factor for angular response sens itivity for diode m at relative gantry angle ; Diode sensitivity correction factors Minor difference in individua l diode construction and elec trometer gain could cause significant variation in diode intrinsic sensitivity. A sensitivity correction factor is needed for each diode to account for this effect. A straightfo rward way to calibrate the diode array is to expose each individual diode under same beam condition and compare their responses. With Equation 5-2, if diode m and a randomly selected reference diode r are irradiated at = 0o under same beam condition, their responses will be )0()0,()()0,( DmSAmSDmR (5-3) 93

PAGE 94

and )0()0,()()0,( DrSArSDrR (5-4) The ratio of responses of diode r and m will be, )0,()( )0,()( )0()0,()( )0()0,()( )0,( )0,( mSAmSD rSArSD DmSAmSD DrSArSD mR rR (5-5) Since angular correction factor is normalized at = 0o, )( )( )0,()( )0,()( )0,( )0,( mSD rSD mSAmSD rSArSD mR rR (5-6) Thus, diode sensitivity co rrection factor for diode m with respect to reference diode r is given by, )0,( )0,( )( )( )( mR rR mSD rSD mCFD (5-7) In calibration, the diode array was positioned on the plastic stand with its axis aligned with gantry rotation axis. Either the di ode array or the gantry could be rotated such that each time one different diode was irradiated on beam central axis Gantry rotation was chos en due to the lack of a reliable mechanism to rotate the diode array at the moment. Gantry rotation accuracy was checked before the calibration according to TG-40105. Radiation beams of field size of 8x30 cm2 and 100 MU were used in each radiation as show n in Figure 5-3 (b). The marker of each diode on the cylinder outer surface was used as reference to rotate gantry and position beam central axis right on top of each diode. CFD for all di odes was obtained with Equation 5-7 after a full gantry rotation. Diode directional response correction factors The different amount of backsca tter with respect to beam in cident direction caused by the metal contact and other high-Z packaging materi als surrounding the diode partly explains diode directional response dependence. For each diode, di rectional correction fact or is a function of 94

PAGE 95

beam incident angle. Ideally, all diodes shoul d exhibit same angular behavior and all diodes share the same set of angular co rrection factors. However, due to the difference in individual diode construction and their placement in the cylindrical phantom, di fferent diodes exhibit different angular behavior. As a consequence, individual angula r correction fact ors should be acquired for each diode. An intuitive way to evaluate diode direction response dependence is to irradiate the diode with the same beam condition and build-up materi al from different beam direction and observe its response difference. However, it is not pr actical to pull out each of the 124 diodes and conduct such experiments. The 4-D diode array is designed to present an isotropic detector image in beams eye view. Thus the goal of the calibration is to remove response difference from different directions and achieve isotropic dosimet er response. In other words, the measured dose distribution should be independent of beam incident direction. In previous section, a full gantry rotation is performed to obtain CFD for each diode. The beam opening (8x30 cm2) is large enough to cove r all diodes during each irradiation. Response of diode m as a function of relative gantry angle is )(),()(),( DmSAmSDmR (5-8) as shown in Figure 5-3 (c). The normalized response ),(mRwould be, )0( )( )0,( ),( )0()0,()( )(),()( )0,( ),( ),( D D mSA mSA DmSAmSD DmSAmSD mR mR mR (5-9) Diode sensitivity SD ( m ) cancels out since it is independent of gantry angle. Assume the true dose profile as a function of relati ve gantry angle is given by D (). The normalized true dose profile would be, 95

PAGE 96

)0( )( )( D D D (5-10) Obviously, the difference between diode measured pr ofile and true dose profile is attributable to diode angular response dependence as shown in Figure 5-3 (c). Thus the angular response sensitivity correction factor for diode m is defined as the ratio of true beam profile to the beam profile measured by diode m, ),( )( ),( mR D mCFA (5-11) Plug Equation 5-9 and Equation 5-10 into Equation 5-11, ),( )0,( )( )0( ),( )0,( )0( )( ),( mSA mSA D D mSA mSA D D mCFA (5-12) which says angular response sensitivity correctio n factor is the invers e of angular response sensitivity when normalized at = 0o. This makes sense since diode response will be corrected by multiplying with CFA to remove angular response dependence. In this work, the true dose profile will be approximated with treatment planning system (TPS) calculation. The accuracy of TPS calculation will be validated with ion chamber scanning in water tank. The normalized profile calculated with TPS is given by, )0( )( )(c c cD D D (5-13) Plug Equation 5-9 and Equation 5-13 into Equation 5-11, )( ),( )0,( )( )0( ),( )0,( )0( )( ),( )( ),( K mSA mSA D D mSA mSA D D mR D mCFAc c c (5-14) where K () is defined as 96

PAGE 97

)( )( )0( )( )0( )( )( D D D D D D Kc c c (5-15) which accounts for the uncertainty in TPS calculation. Ideally, K( )=1. Equation 5-14 will be used to estimate directional response sensitiv ity correction factor, CFA. Response of each diode as a function of gantry angle can be extracte d from the measurement conducted in previous section when determining diode intrinsic sensitiv ity correction factor. In other words, both CFD and CFA could be obtained duri ng one full gantry rotation. Independence of Calibration Factors on Field Sizes In this work, a beam of field size 8x30 cm2 was used to irradiate diodes to obtain diode sensitivity. Diode response is known to depend on beam field size and beam energy spectrum. If individual diode exhibits di fferent response dependence on fi eld sizes, the obtained diode correction factors would depend on the selected fi eld size. Experiments were designed to check diode response dependence on field sizes for the diode array. Four diodes were randomly selected from the diode array. In the first expe riment, each diode was irradiated by beams of 100 MU and 4 different field sizes, 4x4 cm2, 8x8 cm2, 8x16 cm2 and 8x30 cm2 with beam central axis normal to diode surface (Figure 5-4 a). The seco nd experiment was the same as the first one except the beam central axis entered the diode at an oblique angle of about 45o (Figure 5-4 b). The response of each diode was normalized to the response at field size 8x8 cm2. The variation among relative diode response of the four diodes was examined at both normal and oblique beam incidence. Gantry Angle Derivation Algorithm There is one more step in appl ying directional correc tion factors. The instant beam incident direction has to be determined to interpolate the corresponding directiona l correction factors. A 97

PAGE 98

digital inclinometer may be used to monitor the beam angle. However, since gantry will keep moving during rotational therapy and diode signal will be updated every 50 ms, it would be difficult to register measured ga ntry angle with each signal update. An ideal approach would be to derive gantry angle from instant diode array signal. This section describes such an algorithm for gantry angle derivation. The algorithm is depicted in Figure 5-5. In ge neral, radiation beam on field edge intersects the ring of diodes twice as beam enters and exits the cylinder. Nei ghboring diodes around the intersection points supply information to determine exact locations of the intersection points (50% on beam edge). The radiation source shoul d reside on the line conn ecting beam exit point and entrance point on the ring. The intersection of two beam lines from two sides of field edge was used to locate radiation source and hence beam incident direction. The detailed pseudo algorithm is listed in Table 5-1. Once a measurement is made, the first step is to correct for diode sensitivity which doesnt depend on beam angle. The unfolded diode profile is shown in Figure 5-5 (left panel) Then Matlab code is used to look for two peaks, one locates on the beam entrance side ( EntMax ) and the other on the beam exit side ( ExtMax ) as in Figure 5-5. Starting from EntMax, looking for the first DownSlope. Th is DownSlope will include the point the beam edge intersects the ring of diodes. A polynomial fit is used to fit a curve along the DownSlope and the 50% point is interpolated on the curve, which is point B on Figure 5-5 (left panel). The search will continue to look for an UpSlope and its 50% point C. Once the search passes ExtMax, another DownSlope and its 50% point D will be located. When the search reaches the end, it circles around to the beginning of the ring and locates another UpSlope and its 50% point A. A straight line is connected from poi nt D to point A. Another line is connected from C to B. The intersection of these two lines will be considered to be the radiation source position as in Figure 98

PAGE 99

5-5 (right panel). The algorithm can be applied to each of the four rings simultaneously to derive four gantry angles and take the average to improve the accuracy. Diode Array Calibration Validation An independently determined (true) dose profile is needed to derive diode array directional correction factors. It could come from independent m easurement with film or other dosimeters, Monte Carlo simulati on or calculation using TPS. In this work, Pinnacle calculated beam profile is used. There are two things we need to validate, (1) Pinnacle calculated beam profile has to be validated independently; (2) Diode array correcti on factors have to be validated to make sure that isotropicity is achieved. This section describes th e validation procedures. The idea to validate Pinnacle dose calculati on is to compare Pinnacle calculated dose profile to dose profile converted from ion chamber scanning in water tank. For this purpose, a rectangular acrylic phantom with dimensions of 45x35x10 cm3 and a 25.4 cm diameter hole was machined to accommodate the diode array. Measurement was performed with the diode array inside the rectangular phantom. Radiation beam of field size 8x30 cm2 and SSD 87.3 cm was chosen to cover all diodes. Dose calculation was performed by Pinnacle on the CT image of the whole phantom with same geometry. A 48x48x48 cm3 water tank (Wellhofer Dosimetrie, Shwarzenbruck, Germany) was used to collect dose profile with the same setup. Cross beam profiles were scanned every half a centim eter along depth direction with a 0.04 cm3 volume and 4 mm diameter ion chamber (CC04, Scanditroni x Wellhofer). The cross beam profiles were interpolated to obtain a fu ll 2-D dose plane in 1x1 mm2 resolution. The 2-D dose distribution in water was then converted to dose distri bution in acrylic according to TG-21106 by accounting for (1) density scaling from acrylic to water (2) inverse square law correction and (3) excessive scattering condition in acry lic (Figure 5-6). Suppose ),,( SSDrdDacrylic is the dose we want to 99

PAGE 100

determine in acrylic where d` is the radiological path length in the medium, r is the field size at the surface, and SSD is the source-surface di stance. The dose can be converted from dose measured in water using the following equation106: ),( ),( ),,(),,'(' 2drTAR drTAR dST dST SSDrdDSSDrdDd d acrylic water en water acrylic (5-16) where ST is the radiological path length in air and d is given by water acrylicdd (5-17) and e d d dr r ST dST rr and (5-18) is the mean linear attenuation coefficient of the beam and e is the electron density of acrylic relative to that of water. The last term in Equation 5-16 is the ratio of the tissue air ratio (TAR) of unscaled field size to that of scaled field size accounting for excess phantom scatter in acrylic relative to water. The converted dose distribution was used to validate the calculated dose distribution by Pinnacle and the diode array measurement corrected with both CFD and CFA. Another experiment was designed to validate isotropicity has been achieved with the calibration. Two open beams with field sizes 8x8 cm2 and 8x30 cm2 were measured with three different beam angles (0o, 81.4o and 278.6o). Each time one different diode was right under beam center. The measured beam profiles were compared to each other to verify the isotropic detector response. Results Relative diode response as a function of field si zes for four randomly se lected diodes with normal beam incidence is shown in Figure 5-7. Re sponse of each diode is normalized to that of 100

PAGE 101

field size 8x8 cm2. Variation among four diodes is within 0.3% at both large field size (8x30 cm2) and small field size (4x4 cm2). Diode response with oblique beam incidence is shown in Figure 5-8 (normalized at 8x8 cm2). Variation at all field sizes is within 0.4%. Figure 5-9 shows relative diode sensitivity. Re lative diode sensitivity exhibits variation ranging from 0.7 to 1.55. Same experiment was repeated five times. It was found variation of diode sensitivity is within 0.6% which agrees w ith reproducibility results reported in Chapter 6. Directional responses of four randomly picked diodes are shown in Figure 5-10 (a). Diode responses are normalized to that of normal beam incidence. All diodes exhibit similar pattern with higher sensitivity when irradiated at 90o and 270o and lower sensitivity when irradiated at 180o. Variation in directional response among diodes is on the order of 2%. A 13% variation can be observed from 90o to 180o which gives 0.14% change per degree. Average directional response of all 124 diodes as a function of beam in cident angle is shown in Figure 5-10 (b). The error bars show stand deviation. The averag e and maximum standard de viations are 1.3% and 1.8%, respectively. Results to validate diode array calibration and TPS dose calculation using a water tank scanning system are shown in Figure 5-11. Acrylic dose distribution converted from water tank scan agrees with TPS calculation within 0.5% which shows TPS can adequately model the cylindrical diode array. In general, diode arra y measurement after applying calibration factors and directional correctio n factors also agrees with converted acrylic dose from water tank measurement within 0.5%. The discrepancy in the low dose region is slightly higher (1%). Open beam profiles measured at three diffe rent gantry angles for field size 8x8 cm2 and 8x30 cm2 are shown in Figure 5-12. Excellent agreement could be observed among three measurements except in the high gr adient region. All three measur ed profiles agree with Pinnacle 101

PAGE 102

calculation with all points passing 1% di fference and 1 mm DTA criteria. Same good comparison can be found for field size 8x30 cm2. Table 5-2 shows the results of beam angle derivation based on detector signal for 4 step and shoot IMRT beams delivered with nominal ga ntry angles. First column of Table 5-2 shows nominal gantry angle. Second column gives th e number of valid signa l updates (beam on time divided by 50 ms). A beam angle is derived for each signal update of each IMRT beam. The average and standard deviation of derived beam angles are shown in the table. Also shown is the difference between nominal beam angle and averag e of derived beam angle. The difference is within 0.9o with a standard deviation of ~1.4o. Discussion It has been realized the current IMRT QA pr actice is insufficient by checking all beams at fixed gantry angle. Dosimetric effect cause d by gantry rotation erro r, gantry sagging and potential MLC movement error dur ing gantry rotation has to be evaluated. It becomes more critical for rotational therapy which has mo re time varying parameters. Current IMRT QA dosimeters are limited by their planar design. When not mounted on the gantry, the planar dosimeters have to be placed parallel to the surfac e of the couch. Thus when gantry rotates, the dosimeter presents an inconsistent image to b eams eye view which could potentially bias the QA results. The planar dosimeters could be mounted on the gantry and rotated with the gantry. In this way, the beam always sees the same detect or image. However, the biggest problem of this solution is that it has no check on gantry rotation. In an extreme case, when the gantry gets stuck and doesnt move at all, no alarm wo uld be generated by the QA device. The 4D diode array, ArcCHECK, provided by Sun Nuclear, is mainly designed for rotational therapy QA. The main idea is to maintain a consistent detector image to beams eye view. The goal is physically achieved by using a cy lindrical phantom and a spiral arrangement of 102

PAGE 103

diodes. Dosimetrically, the device ha s to be calibrated to present isotropic response. In this work, the calibration procedure consists of two parts, diode sensitiv ity correction a nd directional response dependence correction. Di ode sensitivity correc tion factors are obtained by irradiating different diodes under same beam condition and comparing their responses. Direction response dependence correction factors are ob tained based on the idea of isot ropic response, i.e., measured dose distribution should be independ ent of beam incident direction. When an accurate and reliable rotating mechanis m is available, the diode array could be rotated such that each time one diode is placed u nder beam center and irradiated. It is equivalent to rotate the gantry and fix the diode array. In this work, due to the lack of a reliable rotating mechanism, gantry was rotated during the calibra tion. It is assumed th e gantry rotation has minimum error and machine output is independent of gantry angle. The beam chosen to irradiate the diodes has a field size of 8x30 cm2 which has a relatively large un iform region in beam center. Thus the calibration procedure is not sensitive to minor gantry rotation uncertainty. The span of the four rings of diodes is 3 cm in the beam center. The four diodes facing the gantry from each of the four rings are considered to receive the sa me irradiation. Thus all four rings of diodes are calibrated during one rotation. The experiment designed to study diode response as a function of beam field sizes revealed that all four diodes responded in the same way when beam central axis was normal to diode surface. The variation of relative diode response was within 0.3%. This is within experiment uncertainty when considering machine output vari ation. This result shows diode sensitivity obtained in this work is indepe ndent of the beam field size us ed in calibration. In another experiment where beam entered di ode at an obliq ue angle of 45o, relative diode response varied 103

PAGE 104

within 0.4% which is also within experiment uncertainty. Thus selected beam field size during calibration shouldnt aff ect diode directional correction factors. Results in Figure 5-9 shows diode sensitivity changes about two folds. The difference in individual diode design partially accounts for the large variation. The exact geometry (shift and orientation) with which each diode is placed in its position (a small air space in the detector phantom) might cause difference also. The air space is about twice the size of the diode which is a big dosimetric concern for both TPS modeling and diode measurement. The directional response depe ndence correction factors are de rived based on the concept of a consistent detector image. When irradiated from different beam angles, the diode array is supposed to measure the same dose distribution. After diode sensitivity difference is removed, the difference left in measured dose distribution fro m different beam directions is attributed to diode directional response dependence. The measurement collected for the purpose of correcting diode sensitivity can be reused. All the corr ection factors are derived from same set of measurements. All diodes show similar pattern in directional response. It over-responses when irradiated from the side and under-responses when irradiated from back and front surfaces. The variation of diode directional responses is well within 2% as s hown in Figure 5-10 (b). It might be reasonable to use the averag e directional response for each si ngle diode. The feasibility will be evaluated in next chapter when disc ussing diode array clinical applications. Pinnacle dose calculation, diode array measur ement and dose profile converted from water tank scan come to very good agreement (<1%) as shown in Figure 5-11. This finding shows that Pinnacle can model the cylindrical diode array adequately. The corrected diode reading falls right on Pinnacle calculated dose profile which proves the accuracy of the calibration procedure. The measurements from three different beam angles for two differe nt field sizes (8x8 cm2 and 104

PAGE 105

8x30 cm2) all fall on the Pinnacle calculation. They all agree each other except around the high gradient region where measurement is extremely sensitive to diode position. This shows the calibration has successfully achieved isotropicity. The algorithm to derive instant gantry angle can achieve accuracy of within 1o with standard deviation of 1.4o. The purpose of this algorithm is to derive instant gantry angle to interpolate corresponding angular correction factors, not to check accuracy in gantry rotation. The diode directional response changes by only 0.14% per degree. Thus the uncertainty of this algorithm would cause at most 0.2% dosimetric change which is negligible. The full sized ArcCHECK would have 62 diodes on each ring whic h doubles the density of diodes. There will be more diodes on both the upslope and the downs lope. The polynomial fit used to determine the 50% point would be more accurate. Thus it is ex pected the accuracy of the algorithm would be further increased with the full sized ArcCHECK. Its applicability for ga ntry rotation QA would be discussed when a full sized ArcCHECK is available. This algorithm relies on finding two beam edges intersecting the ring of diodes. When the beam is too large and go outside of the cylinder, the algorithm will fail to find the radiation source and hence the gantry angle. This would be a limitation of the algorithm. However, for practical IMRT beams and full sized ArcCHECK with more than 20 rings of diodes, it is expected at least one ring could capture both sides of the beam edges and provide enough inform ation to determine beam incident direction. The current calibration procedure takes about 30~40 mins. Recommended calibration procedure period is once a year. A more effici ent calibration algorithm is still in search. Conclusions A 4D diode array provided by Sun Nuclear Corpor ation is described and calibrated in this chapter. Calibration procedure taking into account diode sens itivity and directional response dependence is developed. The ca libration factors are validated ag ainst Pinnacle calculation as 105

PAGE 106

well as water tank scan. The diode array presents isotropic res ponse after applying calibration factors. A real time algorithm to derive inst ant beam angles is developed and validated. Corresponding directional correction factors could be interpolated based on derived beam angle and applied to correct instant dose response. The 4D diode arra y is calibrated and ready for clinical application. Next chapter will examine its dosimetric characteristics and demonstrate its clinical application for both IMRT and VMAT QA. 106

PAGE 107

Table 5-1. Pseudo algorithm to derive instant beam angle based on diode array signal. Refer to Figure 5-5 for an illustration of the algorithm. Apply diode sensitivity correction factors Find diode with maximum reading, EntMax; Start from EntMax look for another peak which is exit peak, ExtMax ; Start from EntMax go through diode ring, l ook for diodes on downslope Polynomial fit downslope and in terpolate to get 50% point, TopDown; Start from TopDown look for upslope in between TopDown and ExtMax ; Polynomial fit upslope and in terpolate to get 50% point, BottomUp; Start from ExitMax look for downslope ; Polynomial fit downslope and in terpolate to get 50% point, BottomDown; Start from BottomDown look for upslope in between BottomDown and Entmax; Polynomial fit upslope and in terpolate to get 50% point, TopUp; Connect a line from BottomDown to TopUp; Connect a line from BottomUp to TopDown; Find the intersection of the two lin es which is the source position; 107

PAGE 108

Table 5-2. Beam incident angle derivation for f our Step and shoot IMRT beams irradiated with nominal beam angles. Diode signal is updated every 50 ms. Beam angle is derived for each valid signal update during the course of beam irradiation. Average and standard deviation of derived beam angles are shown in the table. Also shown is the difference between nominal and derived beam angles. Nominal Beam Angle Numb er of Signal Update Derived Beam Angle Mean Difference Standard Deviation 0 159 0.1 0.1 1.4 70 164 70.0 0.0 1.4 220 220 219.6 0.4 1.3 290 133 289.1 0.9 1.2 108

PAGE 109

Figure 5-1. (a) prototype of a 4D diode array. A plastic stand is machined to support the cylindrical phantom. An acrylic insert is made and put in place during all measurements. (b) A CT slice of the 4D diode array showing one ring of diodes. p-type n-type Cathode Anode metal contact Figure 5-2. Diagram of a p-n junction diode. The close contact between n-type silicon and ptype silicon is called depletion region. Elect ron-hole pairs create d inside the diode body diffuse into the depletion region. Electr onic field can be applied to collect the charge. Electron backscatte r from metal contact and high-Z packaging materials surrounding the diode cause diode di rectional response dependence. 109

PAGE 110

Radiation Source (A) (B) (C) Figure 5-3. (A) Relative gantry angle is the angle between the norm passing a diode and the norm passing radiation source. is the angle between Y-axis and the norm passing the diode. Nominal gantry angle = + (B) Set up to calibrate the diode array. Diode array was rotated by 360o with step size of 5.6o. Each time one diode was irradiated under beam center with a beam of field size 8x30 cm2 and 100 MU. (C) Normalized beam profile measured by di ode as a function of gantry angle and normalized true dose profile. The differen ce between diode profile and true profile is caused by diode angular response depe ndence. Ratio of true dose profile to measured dose profile is used to corr ect diode angular re sponse dependence. -200 -100 0 100 200 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 gantry anglediode response R(m, ) True Profile diode m Gantry Y-Axis N orm passing diode m 110

PAGE 111

Radiation Source Radiation Source (a) Beam CAX normal to diode surface (b) Beam CAX oblique to diode surface Figure 5-4. (a) Four randomly selected diodes were irradiated with b eam central axis (CAX) normal to diode surface. Each diode were irra diated with four beams of different field sizes (4x4 cm2, 8x8 cm2, 8x16 cm2 and 8x30 cm2). (b) Four randomly selected diodes were irradiated with beam CA X oblique to diode surface (45o). Each diode was irradiated with five beams of different field sizes (4x4 cm2, 8x8 cm2, 8x16 cm2, 8x24 cm2 and 8x30 cm2). 111

PAGE 112

Radiation Source D C B A D A EntMa x sta r tin g p o int C B Up S l op e Do wn S l o p e Ex tMax Figure 5-5. Algorithm for deriving instant beam in cident direction. Left panel shows the typical unfolded dose profile. T h e algorithm searches for four 50% points using polynom ial fitting and interpo l ation. Right panel: Radiation s ource position is determined th rough back projection from two beam edge s. r T S Radiatio n Source Radiatio n Source Water r S d` T Acrylic A A d B Figure 5-6. Convert dose m easured in water to d ose in acrylic accord ing to TG-21. A diode was assum e d to be at pos itio n A in acrylic pha ntom Density sca l ing was perf orm ed f r om radio l ogica l path TA in a crylic to TB in water acc ording to the rela tive ele c t ron concentration of acrylic to water. Inve rse square law correc tion was applied along radiological path length (from SB bac k to SA). Excessive scattering in acry lic was also taken into account during conversion. 112

PAGE 113

0 5 10 15 20 25 30 0.92 0.94 0.96 0.98 1.00 1.02 1.04 1.06 field size (cm)relative diode response diode 1 diode 2 diode 3 diode 4 Figure 5-7. Relative diode respons e as a function of field size fo r four randomly selected diodes. Each diode was irradiated with four beams of different field sizes (4x4 cm2, 8x8 cm2, 8x16 cm2 and 8x30 cm2) with beam central axis nor mal to diode surface. Diode response was normalized to field size 8x8 cm2. 0 5 10 15 20 25 30 0.92 0.94 0.96 0.98 1.00 1.02 1.04 1.06 field size (cm)relative diode response diode 1 diode 2 diode 3 diode 4 Figure 5-8. Relative diode respons e as a function of field size fo r four randomly selected diodes with beam central axis oblique to diode surface (45o). Five beams of different field sizes (4x4 cm2, 8x8 cm2, 8x16 cm2, 8x16 cm2 and 8x30 cm2) were used for this purpose. Diode response was normalized to 8x8 cm2. 113

PAGE 114

0 20 40 60 80 100 120 140 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 diode numberrelativ diode sensitivity Figure 5-9. Relative diode sensitivit y of all diodes on the diode array. 114

PAGE 115

0 50 100 150 200 250 300 350 0.94 0.96 0.98 1.00 1.02 1.04 1.06 1.08 1.10 1.12 angle between beam central axis and diode surface normal(o)diode directional response diode #1 diode #2 diode #3 diode #4 0 50 100 150 200 250 300 350 0.92 0.94 0.96 0.98 1.00 1.02 1.04 1.06 1.08 1.10 1.12 angle between beam central axis and diode surface normal(o)average diode directional response Figure 5-10. (A) Directional response as a func tion of beam incident angle for four randomly selected diodes. It is normalized at 0o. (B) Average directiona l response of all 124 diodes as a function of beam incident angle. The error bars show stand deviation. -40 -30 -20 -10 0 10 20 30 40 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 arc distance along ring of diodes (cm)relative dose Pinnacle Diode array Measured Figure 5-11. Normalized beam profile calculated by Pinnacle (8.0m), measured with diode array and converted from a water tank measurement for an 8x30 cm2 beam. Diode array measurement was corrected for diode se nsitivity and directional response dependence. Dose distribution measured in water was converted to dose in acrylic according to TG-21. 115

PAGE 116

116 -40 -30 -20 -10 0 10 20 30 40 0 0.2 0.4 0.6 0.8 1.0 1.2 arc distance along ring of diodes (cm)relative dose profile Pinnacle Diode #1 Diode #2 Diode #3 (a) -40 -30 -20 -10 0 10 20 30 40 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 arc distance along ring of diodes (cm)relative dose profile Pinnacle Diode #1 Diode #2 Diode #3 (b) Figure 5-12. Relative beam prof iles calculated by Pinnacle (8.0 m) and measured with diode array for a beam of field size (a) 8x8 cm2 and (b) 8x30 cm2. Three measurements were performed with different beam angles (0o, 81.4o and 278.6o) with diode #3, #1 and #2 under beam center respectively.

PAGE 117

CHAPTER 6 CLINICAL APPLICATION OF A NOVEL 4D DIODE ARRAY FOR IMRT AND VMAT QA Introduction Two-dimensional (2D) dosimeters have been widely used for routine pre-treatment verification of intensity-modulated radiation therapy (IMRT) due to their ease of use and immediate readout. The purpose of patient specific IMRT QA is to detect potential errors in treatment planning and delivery. So urces of errors in treatment pl anning stage include inadequate beam modeling and dose calculation algorithm errors. Gantry rotation errors and multi-leaf collimator positioning errors with gantry rotation are among the sources of errors in the treatment delivery stage. It is all these complex dynamic co mponents that make patie nt specific individual IMRT beam QA necessary. Due to the planar desi gn of existing 2-D detectors, radiation beams are delivered with fixed gantry and beam central axis perpendicular to the plane of detectors during patient specific IMRT QA. The IMRT QA pr ocedure is very likely to capture potential errors in the planning and plan transfer. However, errors in gantry rotation, gantry sagging and MLC movement errors with gantry rotation woul d remain undetected since IMRT beams are not delivered in planned beam angles during QA. Rotational therapy delivery techniques such as RapidArc and VMAT utilizes dynamic gantry angle, dynamic MLC and variable dose ra te to achieve optimal dose distribution and shorter treatment time. Patient specific QA for these rotational therapy modalities has to be changed significantly from step and shoot IMRT. In rotational therapy delivery, patient specific QA has to be performed with gantry rotation. Ex isting 2D dosimeters are insufficient to meet the demand of rotational therapy QA. A novel 4-D diode array (Sun Nucl ear, Melbourne, FL) was deve loped to meet the need of appropriate and efficient QA solu tion for both segmental IMRT a nd rotational radiotherapy. The 117

PAGE 118

isotropic arrangement of diodes in a cylindrical phantom presents a consistent detector image in beams eye view at arbitrary gantry angles. The 50 ms update rate of th e detector array makes per-beam analysis possible with any rotational delivery techniques. Ca libration procedure was developed in Chapter 5 to correct for individual diode sensitivity di fference as well as directional response dependence. An algorithm was developed to derive instant beam incident angle from diode array signal to apply correspondi ng directional corr ection factors. In this chapter, dosimetric char acteristics of the 4D diode ar ray such as reproducibility of detector response, linearity as a function of delivered monitor units (MU), detector response dependence on pulse rate and dose rate will be eval uated. Its clinical application for MLC based IMRT and VMAT QA will be demonstrated. The sens itivity of the diode array to gantry rotation errors and MLC positioning errors at different gantry angles is also evaluated by introducing artificial errors. Materials and Methods Detector characterization An Elekta PreciseTM was used to characterize the 4-D diode array. Radiation beams with nominal energy of 6 MV were used. When comm issioning beam model for this machine, a 0.04 cm3 volume with a diameter of 4 mm ion chamber (CC04, Scanditronix Wellhofer) was used to collect beam profiles. Absolute dose calibration of this machine was performed according to TG51107 giving 1 cGy/MU at dmax=1.5 cm in water with 100 cm SAD and 10x10 cm2 field size. Leaf positions of both leaf banks were calibrated to have accuracy better than 0.3 mm with an inhouse procedure33. The 4-D diode array was supported by th e plastic stand with the axis of the cylinder aligned with gant ry rotation axis and its center co incident with machine isocenter. The reproducibility of the detector was evaluated by irradiating the detector 20 times over a two hour period with the same setup and beam condition (100 MU and 8x30 cm2 field size). 118

PAGE 119

The field size was selected to cover all four ri ngs of diodes and avoid i rradiating electronic part of the detector. A 30 second background measur ement was performed before each measurement and subtracted from the following measurement. A cylindrical ion chamber (IC-15, Wellhofer, Schwarzenbruck, Germany) was used to monitor machine output constancy. Ten diodes were randomly selected from the top, the bottom and off-axis positions from the cylinder. The diode array reproducibility was evaluated by calculating th e relative standard deviations of the selected diodes during the 20 measurements. Linearity of diode array response was evaluated by irradiating the detector with beams of 8x8 cm2 field size and 1~1000 MU. Larger MU could be used since the diode array will never saturate but this range of MU will cover all clinical applications. Response of the top diode on beam central axis was studied. Other diodes were calibrated against the top diode to have same response using procedure described in Chapter 5. Diodes are known to have instant dose rate dependence38, 66, 108-110. As we know, photon beams are generated when high speed electrons hit target. In medical LINAC, high speed electrons come in the form of pulses as in Figure 6-1 (a). Instant pulse rate is defined as number of pulses delivered per unit time. It is also calle d pulse repetition rate. Pulse rate is inversely related to the gap between two c ontinuous pulses. The pulse rate c ould be changed directly from Elekta console. Available pulse rate for Elekta Precise is [25 50 100 200 400] MU/min. Higher pulse rate means less time to deliver the same amount of monitor units or dose. Dose rate is a different concept. Its the rate of dose deposited in the medium. Pulse rate is directly related to dose rate. Higher pulse rate result s in higher dose rate. Another wa y to change dose rate is to change the distance from the point of interest to the radiation sour ce. Dose rate changes according to inverse square law with the distance from radiation source to the medium. In this 119

PAGE 120

way dose rate changes while pulse rate remains the same. In clinic, pulse rate and dose rate are used interchangeably. Pulse rate and dose rate dependence of the diode array needs to be investigated because of two reasons: (1) Diodes are distributed on a 3-D cylinder of 20 cm diameter and bottom diodes on the cylinder are ~ 20 cm farther away from the radiation source than diodes on the top. Dose rate at the top will be 1.5 times highe r than that at the bottom. (2) Rotational radiotherapy techniques such as RapidARC and VMAT could be delivered with variable dose rate during rotati on by changing pulse rate. In this study, beam of 100 MU and 8x8 cm2 field size was delivered with pulse rate changing from 50 MU/min to 400 MU/min from Elekta console to investigate diode pulse rate dependence. To evaluate diode dose rate dependence, dose rate was changed by varying sour ce to detector distance (SDD) from 70 cm to 130 cm as in Figure 6-1 (b). A farmer Ion Chamber was used as referen ce in both cases with 5 cm water equivalent build up. Response of the di ode on the top of the cylinder was compared to the farmer ion chamber to evaluate it s pulse rate and dose rate dependence. Clinical Application The purpose of radiotherapy QA is to verify that the delivered dose distribution agrees with prescribed dose distribution. Clinical appl ication of the 4-D diode array follows the same principles as in the application of 2-D detector arrays and software38. In this study, three head and neck IMRT plans (totaling 17 beams) and one pr ostate conformal arc plan were verified with the diode array on an Elekta PreciseTM machine capable of volumetr ic modulated arc therapy. For step and shoot IMRT, treatment plans were designed with Pinnacle3 and transferred onto the CT images of the diode array for dos e calculation with nominal beam angles. A 2x2x2 mm3 dose grid was used in dose calculation. Three dimensional dose distribution for each IMRT beam was exported from Pinnacle3 using Radiation Therapy On cology Group (RTOG) protocol. All IMRT beams were delivered with planned beam angles and measured with the diode array. 120

PAGE 121

Instant gantry angles were deri ved from detector signa ls using gantry angl e derivation algorithms developed in Chapter 5. Diode sensitivity correction factors and corresponding angular correction factors were applied to diode raw reading. The reference diode was irradiated with an open beam of 8x8 cm2 field size, 100 MU to obtain an absolute dose conversion factor which would convert relative diode response into absolu te dose. Calculated 3D dose distributions were compared to the corrected diode array measurement in 3D space using percent difference (%diff) and distance to agreement (DTA ) for each individual beam. A pa ss/fail test was performed for each diode. In this study, a 5% threshold of th e maximum dose was employed to filter out low dose value which could bias the pass/fail analys is. Criteria of 1%/2mm and 1%/3mm were both used in the analysis. Percent difference was calcu lated based on local dose value which is more stringent than based on maximum dose in the w hole field. The DTA search was performed in 3D space. For the prostate conformal arc pl an, gantry went clockwise from 270o to 90o. Pinnacle treats conformal arc plan as a collection of static beams (cont rol points) with certain angle separation. Dose calculation is performed for each control point and final dose distribution will be interpolated from the super imposed dose distribution. A smaller angular separation between control points leads to smoother final dose distribu tion. In this work, angular separations of 2o and 5o between control points were used when creati ng the conformal arc plan which resulted in 90 and 36 total control points, respectively. In delivery and measurement with the diode array, real time gantry angle was derived to interpol ate corresponding angular correction factors to apply on raw diode reading. Relative diode respons e was converted into absolute dose in the same way as mentioned in IMRT measuremen t. Measured and calculated composite dose distributions were compar ed using %diff and DTA. 121

PAGE 122

In Chapter 5, it was found all diodes exhibit si milar patterns in directional response. The directional correction factors vary by less than 2% on average. The feasibility to use the averaged directional response correction factors for all di odes was investigated. The raw measurement for IMRT beams was corrected by individual directio nal correction factors as well as averaged directional correction factors. Both corrected m easurements were compared with TPS calculation to examine the pass rate change. For a comparis on, measurement without directional correction factors was also compared to TPS calculation. Sensitivity to errors To evaluate the diode array sensitivity to ga ntry rotation errors, three IMRT beams (with beam angles 20 o, 70o and 290o) were selected to deliver without gantry rotation errors and with o and o errors in gantry rotation which resulte d in five measurements for each beam. Positive and negative errors in gantry rotation cause gantry to over rotate and under rotate, respectively. These measurements were compared to TPS calculation and the percentage change of diodes passing the 1%/1mm and 1%/2mm criterion was used to evaluate the diode array sensitivity to minor erro rs in gantry rotation. Three IMRT plans were selected to estimate diode array sensitivity to random as well as systematic MLC position errors. Wrong leaf bank ca libration is among the causes of systematic MLC errors111. Systematic errors were simulated by m odifying both leaf banks by mm or mm as in Chapter 2. Positive and negative errors caused MLC apertures to contract and expand, respectively. A random leaf positioning error sampled from [0mm, mm, mm] was applied to change each leaf position to simulate random MLC position errors. IMRT plans were exported from Pinnacle using Pinnacle scripting language. Corresponding errors were applied to each leaf while avoiding leaf collisions. Modified plans were then imported back for delivery. Together 122

PAGE 123

with the original error free plan, five variations (one with random error and four with systematic errors) were delivered and resulted in six meas urements. These measurements were compared to the TPS calculation with error free plan to ev aluate diode array sensitivity to MLC position errors. Results The short term reproducibility was evaluated by repeating the same measurement 20 times over 2 hour period. The relative standard deviation of 10 randoml y selected diodes is 0.7% on average with a maximum of 0.9%. This result includes the machine output variation during the course of measurement. Machine output fluctuat ion is found to be 0.3% with a monitoring ion chamber. Short term diode variation is estimate d to be ~0.6% after subtracting machine output fluctuation in quadratic form. Diode array is found to have excellent linearity with a lin ear regression coefficient of 0.9997. Linear diode response is shown in Figure 6-2. Figure 6-3(a) shows diode pulse rate dependence of the diode array. A nearly linear decrease is noticed in diode sensitivity with decreased pulse rate. Relative diode sensitivity decreases by 1.7% when pulse rate decreases from 400 MU/min to 50 MU/min. In comparison, the Farmer ion chamber shows a decrease of less than 0.5%. This result agrees with the finding of Ltourneau et al .88 In their work, the behavior of 2-D diode array (MapCHECKTM, Sun Nuclear, Melbourne, FL) was evaluated which uses the same type of diode according to the manufacturer. Diode dose rate dependence is shown in Figure 6-3(b). SDD was changed from 70 cm to 130 cm to change dose rate. A Farmer ion chamber was used to measure dose rate and normalized at 100 cm SDD. A nominal repeti tion rate of 400 MU/min was used during measurement. Dose rate was estimated to cha nge from 2.4 Gy/min to 7.8 Gy/min when SSD changed from 130 cm to 70 cm. Within this dose rate range, diode sensitiv ity relative to response 123

PAGE 124

of Farmer ion chamber varies from -0.5% to 1.1% This result agrees with the findings of other works where the behavior of same type of diode was evaluated38, 88. A slight diode over response could be observed with increasing dose rate66, 88. For the measurement of the 17 IMRT beams, 65 out of 124 diodes are above the 5% threshold on average. Two beams with beam angle 135o and 220o which crossed table top before entering detector have significantly lower aver age passing rates than the other 15 beams. The average passing rates with one standard devi ation for the other 15 beams are 94.3.1% and 99.6.8% with 1%/2mm and 1%/3mm criteria resp ectively. While the average passing rates for the two beams crossing table top before the di ode array are only 82.2% (1%/2mm) and 88.8% (1%/3mm). Further experiment and analysis were conducted to investigate the cause of lower passing rate for these two beams. Both the detect or and beam incident direction were rotated 180o to avoid table top and to ir radiate the same part of the diode array. The average passing rates for the two beams jumped to 95.9% with 1%/2mm and 98.9% with 1%/3mm. It could be concluded that table top attenu ation was the cause. It was also noticed a 3.5% boost in MU would increase the passing rate to 93.0% with 1%/2mm and 98.2% with 1%/3mm when the couch was in the beams way. Figure 6-4 (a~d) shows diode measurements on all 4 rings for a typical IMRT beam with beam angle 290o. Also shown is the TPS calculation at corresponding positions. Dose profiles are unfolded for displaying purpose with origin set at the top of the cylinder. This particular beam enters the detector at arc distance of 12 cm and exits at -18 cm. Two peaks (entrance and exit) could be clearly observed from measured dose profiles on all 4 rings. Diodes passing the analysis criteria (1%/2mm) are shown as plus signs while cold and hot diodes are shown as stars and diamonds respectively. 124

PAGE 125

Table 6-1 shows the passing rate changes wh en using different directional correction factors. The average passing rates when using in dividual directional corr ection factors are 93.3% and 99.1% with 1%/2mm and 1%/3mm, respectivel y. The use of average directional correction factors caused negligible drop (1%) with both 1%/2mm and 1%/3mm. However, when no directional correction factors were applied, drops of average passing rate were noticed to be 13% with 1%/2mm and 9% with 1%/3mm. Results for the prostate conformal arc plan are shown in Figure 6-5. For the diode array measurement, all 124 diode measurements are above the 5% threshold with 120 diodes passing the 1%/2mm criterion, giving a 97% passing rate. When using 1%/3mm criterion, the passing rate increases to 100%. The plan is calculated with both 36 control poin ts (solid line) and 90 control points (dotted line). Dose profile calculated with 90 control points is much smoother than the other one. However, when compared to diode array measurement, they result in same agreement. Figure 6-6 shows average passing rate change with gantry rotation error using both 1%/1mm and 1%/2mm. The standa rd deviation is shown as error bar. When using 1%/2mm, average passing rate drops from 95.9% (no error) to 88.8% (+1o gantry rotation error), 86.6% (1o), 75.6% (+2o) and 70.6% (-2o), respectively. With tighter 1%/1mm, average passing rate decreases from 69.4% (no error) to 60.7% (+1o), 46.7% (-1o), 37.6% (+2o) and 33.6% (-2o). A slightly asymmetric behavior is noticed with la rger passing rate drop fo r negative gantry rotation errors. Sensitivity of the detector to MLC position er rors is shown in Figure 6-7. When using 1%/2mm, average passing rate stays around 94.6% and shows neg ligible change with random MLC errors or +1 mm systematic errors. Around 9% decrease in passing rate is observed with 125

PAGE 126

both -1 mm and +2 mm errors. Largest drop (15%) o ccurs with -2 mm error. The detector shows higher sensitivity with a tighter 1%/1mm criterion. The passing rate drops by more than 20% with mm error, ~6% with random error or +1 mm error and ~13% with -1 mm error. The detector is highly sensitive to systematic errors of 2 mm with either 1%/2mm or 1%/1mm. However, a tighter criterion is favor able to detect errors of 1 mm. Discussion In order to be useful as a radiation dosi meter, the diode array must possess several desirable characteristics. In this chapter, the diode array is characterized by its reproducibility, linearity, pulse rate dependence and dose rate de pendence. Its application for IMRT and VMAT QA is also demonstrated. The diode array was fo und to have excellent linearity with delivered dose. A linear regression coefficient of 0.9997 wa s obtained for the MU range of 1 to 1000. The diode array uses a proprietary designed ASIC chip which empties the capacitor every 50 ms. The diode wont saturate with higher MU. Excellent linearity can be expected from the diode array with MU higher than 1000 though it is beyond typical clinical range. The relative standard deviation during reproducibility study was found to be 0.6%. The same uncertainty has been reported for its counterpart, MapCHECK, which uses the same type of diodes39. Since the diode array uses N-type diodes, its dependence on instant pulse rate and dose rate is a concern. Figure 6-3 (a) shows 1.7% variation wh en instant pulse rate changes from 50 MU/min to 400 MU/min and 1.6% variation when dose rate changes from 2.4 Gy/Min to 7. 8 Gy/Min. This variation is in agreement with the same type of diodes reported from MapCHECK38, 39. The clinical effect of diode dose rate dependence on rotational deli very needs to be further evaluated. For IMRT beams, the measurement after corr ecting for diode sensi tivity and directional response dependence agrees with Pinnacle calculat ion pretty well as can be seen from Figure 6126

PAGE 127

4. An entrance peak and an exit peak can be clearly identified. The exit peak is much shorter than the entrance peak due to the attenua tion of the diode array itself as well as the plastic insert. For the measurement in Figure 6-4, several cold spot s (measurement lower than calculation) could be observed on the exit peak. This might be explai ned by the results shown in Figure 6-3 (b) that diodes under response at lower dose ra te. The dose rate at entrance peak is about 1.5 times higher than that at the bottom according to inverse square law. The pass/fail analysis was conducted in 3D space which is different from QA analysis based on 2D dosimeter. The 2D dosimeter samples a plane perpendicular to beam in cident direction. Dose change along beam incident direction would be very slow. High gradient informati on would be only contained in the sampled 2D plane. However, the 4D diode array samples a 3D sub space. The distance to agreement has to be searched in all directions to capture gradient in formation. This can be seen from Figure 6-4 (d). No agreement for the hot spot (failed 1%/2 mm) would be found in the displayed plane. However, when using 1%/3mm as the criterion, agreement would be found in the direction perpendicular to the displayed plane. It was also found the passing rate was dominated by the DTA criteria while percent difference had negl igible effect. The diodes passing 1%/2mm and 3%/2mm were almost the same for the majority of the IMRT beams. Thus the positioning accuracy is extremely important for accurate measur ement. It is conjectured the internal dose can be accurately reconstructed when entrance and ex it dose could be measured accurately with the diode array. The diode array provides flexibil ity to accommodate an ion chamber inside for verification purpose. The diode array is capable of both composite and individual beam dosimetry. Several researchers have shown composite dosimetry tends to smear out the difference between measurement and calculation in individual dosimetry. Excellent agreement has been achieved 127

PAGE 128

with individual dosimetry between the diode array measurement and TPS calculation. The agreement is even better for the prostate conf ormal arc plan when comparison is conducted on composite dose distribution. When comparing measurement to calculation, the full 3D calculated dose grid could be interpolated from 2x2x2 mm3 to 1x1x1 mm3 resolution in advance. However, it was found the system became out of memory quickly in this way. In this work, a sub dose grid of 12x12x12 cm3 was extracted on the fly for each diode. Inter polation and comparison we re conducted within the sub dose grid. We had no problem with computer memory and average time to conduct the comparison for each IMRT field was only around 10 seconds. Results in Table 6-1 show the feasibility to use average directional correction factors for each individual diode. This is in agreement with the data in Figure 5-10 (b) where directional correction factors change within 1.8% from diode to diode and th e factor that the passing rates are dominated by DTA. However, due to the large variation of directional correction factors with beam incident direction (largest change of 13%), the directional response dependence has to be taken into account. It was found the couch has significant influe nce on the agreement between measurement and calculation. The discrepancy is largely due to the couch atte nuation. QA with 2D dosimeters would not reveal the problem since radiation be ams always pass the couch after the dosimeter. To solve this problem, the couch has to be taken into account by the planning system for planning as well as QA purpose112, 113. In contrast to per beam analysis in fixe d gantry IMRT delivery QA, a composite dose distribution is calculated and compared to meas urement for VMAT QA. This is due to the way Pinnacle calculates the dose distribution. It samp les a static beam every a few degrees (e.g., 2o or 128

PAGE 129

5o in this work) which results in a large number of control points (36 and 90 control points for 2o and 5o spacing, respectively). The final dose distribu tion is interpolated fro m the superposition of the dose distribution calculated from each static beam. It is impractical to perform individual dose comparison for each beam due to (1) the larg e number of static beams (2) the difficulty in correlating dose calculation for each control point and diode array measurement. As pointed out before, composite dosimetry can help detect errors in the comprehensive system but not identify sources of errors. How to balan ce the necessity of individual b eam dosimetry and its associated computation time and memory space would be a future topic. The diode array shows potential to catch both minor gantry rotation errors and MLC position errors with results in Fi gure 6-5 and Figure 6-6. In both cases, tighter criterion shows stronger sensitivity. Gantry rotation error of 2o would be easily capture d with average passing rate drop of more than 20% and 30% with 1%/2mm and 1%/3mm, respectively. However, for 1o gantry rotation error, it is easier to make c onclusive decision using 1%/1mm, with which the passing rate drops about 10%~20% while it only drops 5%~7% w ith 1%/2mm. Similarly, the diode array shows stronger sensitivity to MLC position errors of 2 mm than 1 mm. A tighter criterion should be always used to increase th e ability to detect minor errors in MLC position.17 Conclusions A 4D diode array was characterized and its clinical application for IMRT QA has been demonstrated in this work. The 4D diode array was found to have excellent response linearity and reproducibility. Instant pulse rate and dose rate dependence of the diodes were quantified Maximum variation was found to be around 1.7% at extreme conditions. The 4D diode array was used to measure 3 head and neck IMRT pl ans and a prostate conformal arc plan. The measurement was corrected using correction fact ors obtained in Chapter 5. Algorithms have been developed for the comparison of the measurement and TPS calculation in 3D space. 129

PAGE 130

130 Excellent agreement was observed between diode array measurement and TPS calculation (99.6% passing rate with 1%/3 mm criterion) which shows the di ode array has been correctly calibrated and the TPS (Pinnacle 8.0 m) could adequately model the di ode array and perform dose calculation. The sensitivity of this new dosimeter to gantry rotation errors and MLC position errors was evaluated. It was found the di ode array could detect gantry errors of 2o and MLC position errors of 2 mm. The 4D diode array has been proven to be an effective tool for IMRT QA with planned beam angle and it has great potential for rotational delivery QA.

PAGE 131

Table 6-1. Effect of using av eraged directional correction fact ors on passing rate changes for three IMRT beams delivered with planne d beam angles. For comparison, passing rates of the measurement without directional correction factors was also shown in the table. Individual correction Average correction No correction 1%/2mm 1%/3mm 1%/2mm 1%/3mm 1%/2mm 1%/3mm G020 91.8% 97.3% 90.4% 98.6% 82.2% 91.8% G070 95.6% 100% 93.3% 97.8% 77.8% 91.1% G290 92.5% 100% 92.5% 100% 82.7% 88.5% average 93.3% 99.1% 92.1% 98.8% 80.9% 90.5% 131

PAGE 132

e-e-e-e-e-e-t (A) dt SDD Radiation Source (B) Figure 6-1. (a) Photon beams are created by electr on pulses. Radiation puls e rate is defined as 1/dt, where dt is the gap between continuous electron pul ses. Pulse rate could be changed from Elekta console directly. (b) Se tup to change dose ra te at the position of the diode on the top by varying source to detector distance (SDD) from 70 cm to 130 cm. 132

PAGE 133

0 200 400 600 800 1000 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 MUdiode response Figure 6-2. Linear detector re sponse as a function of delivered monitor units (MU) ranging from 1 to 1000. A beam of 8x8 cm2 field size was delivered with center of the diode array aligned with machine isocente r in this study. A linear regression coefficient of 0.9997 is obtained in this study. 133

PAGE 134

50 100 150 200 250 300 350 400 0.980 0.985 0.990 0.995 1.000 1.005 repetition rate(MU/min)relative diode response Farmer IC Diode Array (A) 0.75 1.00 1.25 1.50 1.75 2.00 0.990 0.995 1.000 1.005 1.010 1.015 dose-per-pulsediode response/dose-per-pulse(B) Figure 6-3. (a) Relative detector response as a function of pulse rate for diode and a Farmer ion chamber. Pulse rate was changed by changing the repetition rate from Elekta Precise console. The response is normalized at re petition rate of 400 MU/min. (b) Relative diode response as a function of dose per pulse. Dose rate was changed by varying source to detector distance (SDD) from 70 cm to 130 cm. Dose rate was recorded using a Farmer ion chamber and normali zed at SDD = 100cm and estimated to change from 2.4 Gy/min to 7.8 Gy/m in approximately. Diode response was normalized at SDD=100 cm. 134

PAGE 135

-30 -20 -10 0 10 20 30 0 5 10 15 20 25 30 35 40 45 50 arc distance along ring of diodes (cm)dose (cGy)(a) -30 -20 -10 0 10 20 30 0 5 10 15 20 25 30 35 40 45 50 arc distance along ring of diodes (cm)dose (cGy)(b) 135

PAGE 136

-30 -20 -10 0 10 20 30 0 5 10 15 20 25 30 35 40 arc distance along ring of diodes (cm)dose (cGy)(c) -30 -20 -10 0 10 20 30 0 2 4 6 8 10 12 14 16 18 arc distance along ring of diodes (cm)dose (cGy)(d) Figure 6-4. Dose profile calcula ted by Pinnacle (8.0m) and measured by all 4 rings of diodes on the diode array. Dose profiles were unfolded with origin at the top of the cylinder. Diodes passing 1%/2mm criterion are shown as plus signs. Diodes with measurement under and above calculation are shown as stars and diamond s respectively. 136

PAGE 137

-30 -20 -10 0 10 20 30 100 150 200 250 300 arc distance along ring of diodes (cm)dose (cGy) (a) TPS (90 CP) TPS (36 CP) ArcCHECK -30 -20 -10 0 10 20 30 100 150 200 250 300 350 arc distance along ring of diodes (cm)dose (cGy) (b) TPS (90 CP) TPS (36 CP) ArcCHECK 137

PAGE 138

-30 -20 -10 0 10 20 30 100 150 200 250 300 350 arc distance along ring of diodes (cm)dose (cGy) (c) TPS (90 CP) TPS (36 CP) ArcCHECK -30 -20 -10 0 10 20 30 100 150 200 250 300 350 arc distance along ring of diodes (cm)dose (cGy) (d) TPS (90 CP) TPS (36 CP) ArcCHECK Figure 6-5. Dose profile calcula ted by Pinnacle (8.0m) and measur ed by all 4 rings of diodes for a conformal arc plan going clockwise from 270o to 90o. The plan was calculated with angular separations of 5o (TPS 36 CP) and 2o (TPS 90 CP) between control points. 138

PAGE 139

-2 -1 0 +1 +2 20 30 40 50 60 70 80 90 100 gantry rotation error (o)average passing rate (%) DTA 1%/2mm DTA 1%/1mm Figure 6-6. Average passing rate change of 3 IMRT beams with gantry rotation errors. Nominal beam angles of the three IMRT beams are 20o, 70o and 290o. Positive and negative rotation errors cause gantry over rotate and under rotate, re spectively. Error bars show 1 standard deviation. 139

PAGE 140

140 -2 -1 0 R +1 +2 30 40 50 60 70 80 90 100 MLC positioning error (mm)average passing rate (%) DTA 1%/2mm DTA 3%/1mm DTA 1%/1mm Figure 6-7. Average passing rate change of three IMRT beams with random and systematic MLC errors of mm and mm. Positive and negative MLC errors cause beam aperture (both leaf banks) to expand and contract respectively. For random errors ("R"), a random error was sampled from [0, mm, mm] and applied to each leaf.

PAGE 141

CHAPTER 7 CONCLUSIONS The goal of patient specific IMRT QA is to detect potential sources of errors in the planning and delivery of IMRT treatment plans. The focus of this work is to evaluate and improve the current practice for patient specific IMRT QA. Existing patient specific IMRT QA approaches evaluate the agreement between dose distributions measured in a surrogate phantom with planar dosimeters and calculated in the same phantom by the treatment planning system. Discrepancy beyond a pre-specified tolerance level will be used as an indicator of potential errors in either planning stage or delivery stage. It is expected the current patient specific IMRT QA procedure could catch both gros s errors and subtle errors. The sensitivity of current typical patient-spec ific IMRT QA procedures to systematic MLC positioning errors on the order of 1 mm and 2 mm is evaluated in Chapter 2. Results show these MLC positioning errors cant be detected with the widely adopted 3%/3mm criterion. When using 3%/3mm, the average passing rate of 53 IMRT fields with artificially introduced systematic MLC positioning errors of 2 mm stays around 90% for both EBT films and MapCHECK as shown in Figure 2-2 and Figure 2-3, respectively. Figure 2-5 and Figure 2-6 indicate, only with a tighter 2%/2mm criterion, 2 mm systematic MLC positioning errors would cause significant drop in averag e passing rate. This study shows that a tighter criterion is desirable for IMRT QA to achieve reasonable sensitivity. However, the tighter 2%/2mm criterion would result in low average passing rate even wh en MLC is free of systematic positioning errors due to uncertainty of other components in the sy stem. As a consequence, large amount of false alarms would be created. This is supported by the survey conducted by Nelms et al11 that only few centers are adopting the 2%/2 mm criterion. This study shows gr oss errors would be captured 141

PAGE 142

with the current IMRT QA procedures, however, systematic MLC positioning errors of 2 mm which has significant dosimetric impact would likely be missed. Study in Chapter 3 reveals large discrepanc y between measured and calculated dose distribution of MLC error free IMRT plans is caused by the detector size effect which occurs at the very beginning of IMRT chain, i.e., the T PS commissioning stage. The finite size of the commonly used detectors (2 mm or 3 mm radi us) in collecting beam commissioning profiles deteriorates the high gradient na ture of IMRT beams. This phenomenon has been recognized for a long time. However, its impact on IMRT implem entation has not been documented. A practical and reliable solution is needed to remove or minimize the detect or size effect. Chapter 3 studies the clinical impact of detect or size effect on IMRT via an analytical fitting/deconvolution approach. Photon beam profiles collected with fi nite size detectors are deconvolved to remove detector size effect to obtain a set of true photon beam profiles free of detector size effect. Beam profiles with and without detector size effect are empl oyed to commission Pinnacle to obtain different beam models. Dose calculations with both beam models are compared with measurement performed with planar diode ar rays. It is noticed true beam models (commissioned with true beam profiles) signifi cantly increases the average passing rate (15% with 2%/2mm). The average passing rate of 53 IMRT beams with the true beam models is ~97% with 2%/2mm. The clinical significance of this study is (1) T PS could calculate dose distribution more accurately with the true b eam models which is more important for IMRT beams characterized by many small high gradient beam segments (2) A tighter criterion (2%/2mm) could be employed due to the increas ed agreement between TPS dose prediction and dose measurement which leads to higher sensit ivity of IMRT QA to subtle MLC positioning errors according to the sensitivity study in Chapter 2. 142

PAGE 143

Current patient specific IMRT QA procedure is being blamed for its low efficiency. Each IMRT beam needs to be delivered and measured usually after clinic hours. Average time spent on IMRT QA with online detectors for one patient is about 30 min from de tector setup to final data analysis. With the improved mechanic accu racy, few errors are caught and reported during patient specific IMRT QA. It is argued that physicists valuable time should be spent in other activities such as tumor contouring, patient se tup and motion management which leads to large dosimetric errors and could not be caught by time consuming patient specific IMRT QA. Another argument against indi vidual patient IMRT QA is that MLC positioning errors are machine specific instead of patient specific. MLC positioning errors should be caught during dedicated MLC QA. The solution proposed in this work (Chapter 4) in improving the efficiency of current IMRT QA is to combine an independent planar dose calcula tion with a periodic MLC QA. Instead of delivering and measuring each IM RT beam, dose distribution is calculated with an independent planar dose calculation algorithm. The algorith m is commissioned with photon beam profiles collected from individual machines Detector size effect is minimized with the deconvolution algorithm proposed in Chapter 3. Accu racy of the algorithm is validated with the 2D diode array measurement (gold standard). T PS dose calculation is compared to planar dose distribution calculated with this independent algorithm. The algorithm utilizes plan parameters from the record and verify system which contro ls the delivery system, thus errors generated during plan transfer will be caught. Delivery system accuracy will be checked via a periodic MLC QA procedure. Individual measurement for IMRT QA could be replaced by the efficient (10 seconds per beam) independent planar dose calculation algorithm. However, when seeing discrepancy between TPS dose calculation and the independent dose calculat ion, the plan should be delivered and measured as a gold standard for judgment. 143

PAGE 144

Planned IMRT beams are delivered during IMRT QA but not with planned beam angle due to the limit of current 2D planar dosimeters. Th e consequence is that e rrors related to dynamic gantry or dynamic MLC could remain undetect ed. Current practice of IMRT QA is acceptable for fixed gantry IMRT. However, significant change has to be made for dynamic gantry rotational therapy. In this work, a novel 4D diode array provided by Sun Nuclear Corp. is calibrated in Chapter 5 and its clinical application for IMRT QA is demonstrated in Chapter 6. The calibration of the 4D diode array is based on the concept of a coherent dosimeter, i.e., isotropic response. Diodes are isotropically arranged in a cy lindrical phantom, presenting a consistent detector image to beams eye view at any beam angle. The calibration must eliminate any non-isotropic factors caused by diode respon se. Based on this basic idea, two sets of calibration factors are obta ined: (1) diode sens itivity correc tion factors, to remove diode response difference under same beam conditions and (2) diode directional response correction factors, to remove diode angular response difference. The ca libration factors are validated with water tank scan and found to be very accurate. Same beams ar e irradiated with different gantry angles and excellent agreement among diode array responses are observed at each gantry angle, which shows the calibration has achieved isotropicity. Instant gantry a ngle during rotational therapy has to be determined to apply the corresponding angula r correction factors. A real time algorithm to derive gantry angle based on diode array signal is developed for th is purpose. The algorithm tries to identify beam edges on both sides of the field to determine their intersection as the radiation source. The accuracy of the algo rithm is found to be within 1o which is adequate to interpolate the corresponding angular correction factors. Chapter 6 focuses on the characterization of the 4D diode array and its clinical application for both IMRT and VMAT QA. The diode array is found to have excellent reproducibility and 144

PAGE 145

linearity. Its dependency on pulse repetition rate and diode rate is within 1.5%. For IMRT QA, 3D dose calculation is performed in the CT s can of the diode array and exported using RTOG format. The same beams are delivered with plan ned beam angles and measured with the 4D diode array. Calibration factors ob tained in Chapter 5 are applied to correct diode sensitivity and angular response dependence. The calculated and measured dose distributions are compared with percent difference and distance to agreement crite ria in 3D space. Excelle nt agreement is found between Pinnacle calculation and diode array measurement, which shows Pinnacle can adequately model the diode ar ray and the diode array has be en properly calibrated. The sensitivity of this device to gantry angle errors and MLC positioning errors is also evaluated. Its clinical application for VMAT QA is demonstrated with a pros tate conformal arc plan. Diode array measurement is compared to Pinnacle calcul ated dose distribution in 3D space. Excellent agreement is found between measurement and calculation. This study proves ArcCHCEK is an efficient and valuable tool for both IMRT and VMAT QA. 145

PAGE 146

LIST OF REFERENCES 1G. Ezzell, J. Galvin, D. Low, J. Palta, I. Rosen, M. Sharpe, P. Xia, Y. Xiao, L. Xing, C. Yu and "Guidance document on delivery, treatment pla nning, and clinical implementation of IMRT: report of the IMRT Subcommittee of the AAPM Radiation Therapy Committee," Med Phys 30, 2089-2115 (2003). 2J. Galvin, G. Ezzell, A. Eisbrauch, C. Yu, B. Bu tler, Y. Xiao, I. Rosen, J. Rosenman, M. Sharpe, L. Xing, P. Xia, T. Lomax, D. Low, J. Palta a nd "Implementing IMRT in clinical practice: a joint document of the American Society for Therapeu tic Radiology and Oncology and the American Association of Physicists in Medicine," Int J Radiat Oncol Biol Phys 58, 1616-1634 (2004). 3J. Palta, C. Liu and J. Li, "Quality assurance of intensity-modulated radiation therapy," Int J Radiat Oncol Biol Phys 71, S108-112 (2008). 4K. Otto, "Volumetric modulated arc therapy: IMRT in a single ga ntry arc," Med Phys 35, 310317 (2008). 5S. Bentzen, "Radiation therapy: intensity modul ated, image guided, biologically optimized and evidence based," Radiother Oncol 77, 227-230 (2005). 6C. Yu, C. Amies and M. Svatos, "Planning a nd delivery of intensit y-modulated radiation therapy," Med Phys 35, 5233-5241 (2008). 7L. Mell, A. Mehrotra and A. Mundt, "Intensity -modulated radiation therapy use in the U.S., 2004," Cancer 104, 1296-1303 (2005). 8I. Das, C. Cheng, K. Chopra, R. Mitra, S. Sr ivastava and E. Glatstein, "Intensity-modulated radiation therapy dose prescrip tion, recording, and delivery: patterns of va riability among institutions and treatment planni ng systems," J Natl Cancer Inst 100, 300-307 (2008). 9J. Van Dyk, R. Barnett, J. Cygler and P. Shragge, "Commissioning and quality assurance of treatment planning computers," In t J Radiat Oncol Biol Phys 26, 261-273 (1993). 10D. Low, W. Harms, S. Mutic and J. Purdy, "A technique for the quant itative evaluation of dose distributions," Med Phys 25, 656-661 (1998). 11B. Nelms and J. Simon, "A survey on planar IMRT QA analysis," J Appl Clin Med Phys 8, 7690 (2007). 12P. Basran and M. Woo, "An analysis of tolerance levels in IMRT quality assurance procedures," Med Phys 35, 2300-2307 (2008). 13J. Palta, S. Kim, J. Li and C. Liu, Tolerance limits and action levels for planning and delivery of IMRT QA, Proceedings of the AAPM 2003 Summer School: Intensity modulated radiation 146

PAGE 147

therapy: the state of the art. Colorado Co llege, Colorado Springs CO: Medical Physics Publishing (Madison, WI); 593-612 (2003). 14G. Ibbott, D. Followill, H. Molineu, J. Lowens tein, P. Alvarez and J. Roll, "Challenges in credentialing institutions and participants in advanced technology multi-institutional clinical trials," Int J Radiat Oncol Biol Phys 71, S71-75 (2008). 15O. Zeidan, J. Li, M. Ranade, A. Stell and J. Dempsey, "Verification of step-and-shoot IMRT delivery using a fast video-based electronic portal imaging device," Med Phys 31, 463-476 (2004). 16M. K. Woo and A. Nico, "Impact of multileaf collimator leaf positioning accuracy on intensity modulation radiation therapy quality assurance ion chamber measurements," Medical Physics 32, 1440-1445 (2005). 17G. Yan, C. Liu, T. Simon, L. Peng, C. Fox a nd J. Li, "On the sensitivity of patient-specific IMRT QA to MLC positioning errors," J Appl Clin Med Phys 10, 2915 (2009). 18G. Yan, C. Fox, C. Liu and J. Li, "The extraction of true profiles for TPS commissioning and its impact on IMRT patient-specific QA," Med Phys 35, 3661-3670 (2008). 19G. Budgell, B. Perrin, J. Mott, J. Fairfoul a nd R. Mackay, "Quantitative analysis of patientspecific dosimetric IMRT veri fication," Phys Med Biol 50, 103-119 (2005). 20G. Yan, C. Liu, B. Lu, J. Palta and J. Li, "Comparison of analytic source models for head scatter factor calculation and planar dos e calculation for IMRT," Phys Med Biol 53, 2051-2067 (2008). 21J. Bedford and A. Warrington, "Commissioni ng of volumetric modulated arc therapy (VMAT)," Int J Radiat Oncol Biol Phys 73, 537-545 (2009). 22C. Ling, P. Zhang, Y. Archambault, J. Bocanek, G. Tang and T. Losasso, "Commissioning and quality assurance of RapidArc radiotherapy deliv ery system," Int J Radiat Oncol Biol Phys 72, 575-581 (2008). 23A. Boyer, "Basic applications of a multileaf co llimator," in Teletherapy: Present and Future, Proceedings of the 1996 Summer School of the Amer ican Association of Physicists in Medicine, edited by J. Palta and R. Mackie (M edical Physics, Madison, WI, 1996). 24L. Wei, L. Jinsheng, Robert A. Price, Jr., C. Li li, Y. Jie, F. Jiajin, C. Zuoqun, M. Shawn, X. Xiu and M. Chang-Ming, "Monte Carlo based IM RT dose verification using MLC log files and R/V outputs," Medical Physics 33, 2557-2564 (2006). 25L. Thomas, Quality assurance of IMRT A prac tical guide to intensity-modulated radiation therapy, Med. Phys. Publishing, Madison, WI, PP 83-101 (2003). 147

PAGE 148

26G. Mu, E. Ludlum and P. Xia, "Impact of ML C leaf position errors on simple and complex IMRT plans for head and neck cancer," Phys Med Biol 53, 77-88 (2008). 27P. Zygmanski, J. Kung, S. Jiang and L. Chin, "Dependence of fluence errors in dynamic IMRT on leaf-positional errors varying with time and leaf number," Med Phys 30, 2736-2749 (2003). 28T. LoSasso, C. Chui and C. Ling, "Comprehensive quality assurance for the delivery of intensity modulated radiotherapy with a multileaf collimator used in the dynamic mode," Med Phys 28, 2209-2219 (2001). 29A. Stell, J. Li, O. Zeidan and J. Dempsey, "A n extensive log-file anal ysis of step-and-shoot intensity modulated radiation therapy segment delivery errors," Med Phys 31, 1593-1602 (2004). 30S. Samant, W. Zheng, N. Parra, J. Chandler, A. Gopal, J. Wu, J. Jain, Y. Zhu and M. Sontag, "Verification of multileaf collimator leaf positions using an electronic portal imaging device," Med Phys 29, 2900-2912 (2002). 31J. Bayouth, D. Wendt and S. Morrill, "M LC quality assurance techniques for IMRT applications," Med Phys 30, 743-750 (2003). 32M. Sastre-Padro, U. van der Heide and H. Welle weerd, "An accurate cali bration method of the multileaf collimator valid for c onformal and intensity modulated radiation treatments," Phys Med Biol 49, 2631-2643 (2004). 33C. Liu, T. Simon, C. Fox, J. Li and J. Palta, "M ultileaf collimator characteristics and reliability requirements for IMRT Elekta system," Int J Radiat Oncol Biol Phys 71, S89-92 (2008). 34N. Childress and I. Rosen, "Effect of pro cessing time delay on the dose response of Kodak EDR2 film," Med Phys 31, 2284-2288 (2004). 35M. Sastre-Padro, J. Welleweerd, E. Malinen, K. Eilertsen, D. Olsen and U. van der Heide, "Consequences of leaf ca libration errors on IMRT delivery," Phys Med Biol 52, 1147-1156 (2007). 36A. Niroomand-Rad, C. Blackwell, B. Course y, K. Gall, J. Galvin, W. McLaughlin, A. Meigooni, R. Nath, J. Rodgers and C. 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). 37O. Zeidan, S. Stephenson, S. Meeks, T. Wagne r, T. Willoughby, P. Kupelian and K. Langen, "Characterization and use of EBT radiochromic film for IMRT dose verification," Med Phys 33, 4064-4072 (2006). 38P. Jursinic and B. Nelms, "A 2-D diode array an d analysis software for verification of intensity modulated radiation therapy delivery," Med Phys 30, 870-879 (2003). 148

PAGE 149

39D. Ltourneau, M. Gulam, D. Yan, M. Oldham and J. Wong, "E valuation of a 2D diode array for IMRT quality assurance," Radiother Oncol 70, 199-206 (2004). 40F. Buonamici, A. Compagnucci, L. Marra zzo, S. Russo and M. Bucciolini, "An intercomparison between film dosimetry and diode matrix for IMRT quality assurance," Med Phys 34, 1372-1379 (2007). 41W. Frank, "Individual comparisons by ranking methods," Biometrics Bulletin 1, 4 (1945). 42D. Low and J. Dempsey, "Evaluation of th e gamma dose distribution comparison method," Med Phys 30, 2455-2464 (2003). 43M. Stock, B. Kroupa and D. Ge org, "Interpretation and evalua tion of the gamma index and the gamma index angle for the verification of IMRT hybrid plans," Phys Med Biol 50, 399-411 (2005). 44M. Arnfield, K. Otto, V. Aroumougame and R. Alkins, "The use of film dosimetry of the penumbra region to improve the accuracy of intensity modulated radiotherapy," Med Phys 32, 12-18 (2005). 45A. Van Esch, J. Bohsung, P. Sorvari, M. Tenhu nen, M. Paiusco, M. Iori, P. Engstrm, H. Nystrm and D. Huyskens, "Acceptance tests and quality control (QC) procedures for the clinical implementation of intensity modulated radiotherapy (IMRT) using inverse planning and the sliding window technique: experience from fi ve radiotherapy departments," Radiother Oncol 65, 53-70 (2002). 46F. Garca-Vicente, J. Delgado and C. Peraza, "Experimental determination of the convolution kernel for the study of the spatial response of a detector," Med Phys 25, 202-207 (1998). 47D. Dawson, J. Harper and A. Akinradewo, "Ana lysis of physical parame ters associated with the measurement of high-energy x-ray penumbra," Med Phys 11, 491-497. 48W. Laub and T. Wong, "The volume effect of detect ors in the dosimetry of small fields used in IMRT," Med Phys 30, 341-347 (2003). 49D. Dawson, N. Schroeder and J. Hoya, "Penumbral measurements in water for high-energy x rays," Med Phys 13, 101-104. 50R. Rice, J. Hansen, G. Svensson and R. Siddon, "Measurements of dose distributions in small beams of 6 MV x-rays," Phys Med Biol 32, 1087-1099 (1987). 51F. Garca-Vicente, J. Delgado and C. Rodrguez, "Exact analytical so lution of the convolution integral equation for a general profile fitting function and Gaussian detector kernel," Phys Med Biol 45, 645-650 (2000). 149

PAGE 150

52D. Herrup, J. Chu, H. Cheung and M. Pankuch, "Determination of penumbral widths from ion chamber measurements," Med Phys 32, 3636-3640 (2005). 53D. Low, P. Parikh, J. Dempsey, S. Wahab a nd S. Huq, "Ionization chamber volume averaging effects in dynamic intensity modulated radiation therapy beams," Med Phys 30, 1706-1711 (2003). 54G. Bednarz, M. Saiful Huq and U. Rosenow, "D econvolution of detector size effect for output factor measurement for narrow Gamma Knif e radiosurgery beams," Phys Med Biol 47, 36433649 (2002). 55P. Higgins, C. Sibata, L. Siskind and J. Sohn, "Deconvolution of detector size effect for small field measurement," Med Phys 22, 1663-1666 (1995). 56C. Chui and R. Mohan, "Extraction of pencil beam kernels by the deconvolution method," Med Phys 15, 138-144. 57E. Pappas, T. Maris, A. Papadakis, F. Zachar opoulou, J. Damilakis, N. Papanikolaou and N. Gourtsoyiannis, "Experimental determination of the effect of detector size on profile measurements in narrow photon beams," Med Phys 33, 3700-3710 (2006). 58C. Yang, F. Guo, C. Wu, T. Liu, J. Perks, R. Stern and J. Purdy, "WE-D-224A-09: Proposed Pass/fail Criteria for IMRT Patient Specific QA," Med Phys 33, 2248-2249 (2006). 59G. Starkschall, R. J. Steadham, R. Popple, S. Ahmad and I. Rosen, "Beam-commissioning methodology for a three-dimensional convolution/ superposition photon dose algorithm," J Appl Clin Med Phys 1, 8-27 (2000). 60W. Tom, "Beam modeling for a convolution/supe rposition-based treatment planning system," Med Dosim 27, 11-19 (2002). 61J. Bedford, P. Childs, V. Nordmark Hansen M. Mosleh-Shirazi, F. Verhaegen and A. Warrington, "Commissioning and qual ity assurance of the Pinnacle3 radiotherapy treatment planning system for external beam photons," Br J Radiol 76, 163-176 (2003). 62W. Simon, J. Kozelka, M. Rose, C. Liu, J. Palta, J. Dempsey, B. Lynch, J. Bayouth, D. Pavord, G. Ibbott and D. Followill, "SU-FF-T-311: LINAC Dosimetry: Benchmark Data Set Uncertainty," Med Phys 33, 2118 (2006). 63P. William, T. Saul, V. William and F. Brian, Numerical Recipes 2nd ed. (Cambridge University Press, Cambridge, 1992). 64G. Rikner and E. Grusell, "General specifica tions for silicon semiconductors for use in radiation dosimetry," Phys Med Biol 32, 1109-1117 (1987). 150

PAGE 151

65A. Beddar, D. Mason and P. O'Brien, "Absor bed dose perturbation caused by diodes for small field photon dosimetry," Med Phys 21, 1075-1079 (1994). 66M. Westermark, J. Arndt, B. Nilsson and A. Brahme, "Comparative dosimetry in narrow highenergy photon beams," Phys Med Biol 45, 685-702 (2000). 67P. Cho, H. Kuterdem and R. n. Marks, "A spherical dose model for radiosurgery plan optimization," Phys Med Biol 43, 3145-3148 (1998). 68K. Chang, F. Yin and K. Nie, "The effect of detector size to the broadening of the penumbra--a computer simulated study," Med Phys 23, 1407-1411 (1996). 69B. Fraass, K. Doppke, M. Hunt, G. Kutcher, G. Starkschall, R. Stern and J. Van Dyke, "American Association of Physic ists in Medicine Radiation Therapy Committee Task Group 53: quality assurance for clinical radiot herapy treatment planning," Med Phys 25, 1773-1829 (1998). 70S. Rustgi and D. Frye, "Dosimetric characte rization of radiosurgica l beams with a diamond detector," Med Phys 22, 2117-2121 (1995). 71P. Metcalfe, T. Kron, A. Elliott, T. Wong and P. Hoban, "Dosimetry of 6-MV x-ray beam penumbra," Med Phys 20, 1439-1445. 72P. Bloch and R. Wallace, "Computer genera ted scatter dose distributions for 6-MV radiotherapy photon beams," Med Phys 6, 149-152. 73L. Dong, A. Shiu, S. Tung and K. Hogstr om, "A pencil-beam photon dose algorithm for stereotactic radiosurgery using a miniature multileaf collimator," Med Phys 25, 841-850 (1998). 74F. Lorenz, J. Killoran, F. Wenz and P. Zygm anski, "An independent dose calculation algorithm for MLC-based stereotactic radiotherapy," Med Phys 34, 1605-1614 (2007). 75C. Chui, T. LoSasso and A. Palm, Computationa l algorithms for indepe ndent verification of IMRT, in A Practical Guide to Intensity-Modu lated Radiation Therapy (Madison, WI: Med. Phys. Publishing) pp 83 (2003). 76I. Mihaylov, F. Lerma, Y. Wu and J. Sieber s, "Analytic IMRT dose calculations utilizing Monte Carlo to predict MLC fl uence modulation," Med Phys 33, 828-839 (2006). 77H. Liu, T. Mackie and E. McCullough, "A dual source photon beam model used in convolution/superposition dose cal culations for clinical megavo ltage x-ray beams," Med Phys 24, 1960-1974 (1997). 78Y. Yang, L. Xing, A. Boyer, Y. Song and Y. Hu "A three-source model for the calculation of head scatter fact ors," Med Phys 29, 2024-2033 (2002). 151

PAGE 152

79S. Jiang, A. Boyer and C. Ma, "Modeling th e extrafocal radiati on and monitor chamber backscatter for photon beam dose calculation," Med Phys 28, 55-66 (2001). 80D. Jaffray, J. Battista, A. Fenster and P. Munro, "X-ray sources of medical linear accelerators: focal and extra-focal radiation," Med Phys 20, 1417-1427. 81M. Sharpe, D. Jaffray, J. Battista and P. Munro, "Extrafocal radiation: a unified approach to the prediction of beam penumbra and output factor s for megavoltage x-ray beams," Med Phys 22, 2065-2074 (1995). 82A. Boyer, L. Xing, C. Ma, B. Curran, R. Hill, A. Kania and A. Bleier, "Theoretical considerations of monitor unit calculations for intensity modulat ed beam treatment planning," Med Phys 26, 187-195 (1999). 83J. Deng, S. Jiang, A. Kapur, J. Li, T. Pawlic ki and C. Ma, "Photon be am characterization and modelling for Monte Carlo treatm ent planning," Phys Med Biol 45, 411-427 (2000). 84G. Ding, "An investigation of accelerator head scatter and output factor in air," Med Phys 31, 2527-2533 (2004). 85A. Ahnesj, "Collimator scatter in photon therapy beams," Med Phys 22, 267-278 (1995). 86E. Chaney, T. Cullip and T. Gabriel, "A Mont e Carlo study of accelerator head scatter," Med Phys 21, 1383-1390 (1994). 87M. Fix, M. Stampanoni, P. Manser, E. Born, R. Mini and P. Regsegger, "A multiple source model for 6 MV photon beam dose calculations using Monte Carlo," Phys Med Biol 46, 14071427 (2001). 88D. Ltourneau, H. Keller, M. Sharpe and D. Jaffray, "Integral test phantom for dosimetric quality assurance of image guide d and intensity modulated stereo tactic radiotherapy," Med Phys 34, 1842-1849 (2007). 89L. Xing, Y. Chen, G. Luxton, J. Li and A. B oyer, "Monitor unit calcul ation for an intensity modulated photon field by a simple scatte r-summation algorithm," Phys Med Biol 45, N1-7 (2000). 90L. Xing and J. Li, "Computer verification of fluence map for intensity modulated radiation therapy," Med Phys 27, 2084-2092 (2000). 91J. Kung, G. Chen and F. Kuchnir, "A mon itor unit verification cal culation in intensity modulated radiotherapy as a dosimet ry quality assurance," Med Phys 27, 2226-2230 (2000). 92Y. Yang, L. Xing, J. Li, J. Palta, Y. Chen, G. Luxton and A. Boyer, "Independent dosimetric calculation with inclusion of head scatter and MLC transmission for IMRT," Med Phys 30, 2937-2947 (2003). 152

PAGE 153

93Z. Chen, L. Xing and R. Nath, "Independent m onitor unit calculation for intensity modulated radiotherapy using the MIMiC multileaf collimator," Med Phys 29, 2041-2051 (2002). 94Y. Watanabe, "Point dose calcula tions using an analytical pencil beam kernel for IMRT plan checking," Phys Med Biol 46, 1031-1038 (2001). 95L. Xing, Y. Yong, L. Jonathan and C. Yan, "Monitor unit calculation and plan validation for intensity-modulated radiation ther apy," in Intensity-modulated radia tion therapy: the state of the art, edited by P. Jatinder (Med. Phys., Wisconsin, 2003), pp. 16. 96C. D. Wagter, "The ideal dosimeter for intensity modulated radiation therapy (IMRT): What is required?," Journal of Physics: Conference Series 3, 4 (2004). 97D. Low, J. Dempsey, R. Venkatesan, S. Mutic J. Markman, E. Mark Haacke and J. Purdy, "Evaluation of polymer gels and MRI as a 3D dosimeter for intensity-modulated radiation therapy," Med Phys 26, 1542-1551 (1999). 98F. Gum, J. Scherer, L. Bogner, M. Solleder, B. Rhein and M. Bock, "Preliminary study on the use of an inhomogeneous anthropomorphic Fricke gel phantom and 3D magnetic resonance dosimetry for verification of IMRT treatment plans," Phys Med Biol 47, N67-77 (2002). 99H. Gustavsson, A. Karlsson, S. Bck, L. Olsson, P. Haraldsson, P. Engstrm and H. Nystrm, "MAGIC-type polymer gel for three-dimensiona l dosimetry: intensity-modulated radiation therapy verification," Med Phys 30, 1264-1271 (2003). 100K. Vergote, Y. De Deene, W. Duthoy, W. De Gersem, W. De Neve, E. Achten and C. De Wagter, "Validation and applica tion of polymer gel dosimetry fo r the dose verification of an intensity-modulated arc therapy (I MAT) treatment," Phys Med Biol 49, 287-305 (2004). 101B. Paliwal, W. Tom, S. Richardson and T. Makie, "A spiral phantom for IMRT and tomotherapy treatment delivery verification," Med Phys 27, 2503-2507 (2000). 102B. Poppe, A. Blechschmidt, A. Djouguela, R. Kollhoff, A. Rubach, K. Willborn and D. Harder, "Two-dimensional ionization chamber arra ys for IMRT plan verification," Med Phys 33, 1005-1015 (2006). 103T. Wiezorek, N. Banz, M. Schwedas, M. Sche ithauer, H. Salz, D. Georg and T. Wendt, "Dosimetric quality assurance for intensity-modulated radiot herapy feasibility study for a filmless approach," Strahlenther Onkol 181, 468-474 (2005). 104A. Van Esch, C. Clermont, M. Devillers, M. Io ri and D. 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). 153

PAGE 154

154 105G. 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, M. Weller and L. Wingfield, "Comprehensive QA for radiation oncology: Report of AAPM Radiatio n Therapy Committee Task Group 40," Med Phys 21, 581-618 (1994). 106AAPM TG-21, "A protocol for the determina tion of absorbed dose from high-energy photon and electron beams," Med. Phys. 10, 741 (1983). 107P. R. Almond, P. J. Biggs, B. M. Coursey, W. F. Hanson, M. S. Huq, R. Nath, and D. W. O. Rogers, "AAPM's TG-51 protocol for clinical reference dosimetry of high-energy photon and electron beams," Med. Phys. 26, 1847 (1999). 108A. Saini and T. Zhu, "Dose rate and SDD dependence of commercially available diode detectors," Med Phys 31, 914-924 (2004). 109P. Jursinic, "Implementation of an in vivo diode dosimetry program and changes in diode characteristics over a 4-year clinical history," Med Phys 28, 1718-1726 (2001). 110J. Shi, W. Simon and T. Zhu, "Modeling the in stantaneous dose rate dependence of radiation diode detectors," Med Phys 30, 2509-2519 (2003). 111C. Ramsey, S. Dube and W. R. Hendee, "It is necessary to validate each individual IMRT treatment plan before delivery," Med Phys 30, 2271-2273 (2003). 112E. Spezi, A. Angelini, F. Romani, A. Guido, F. Bunkheila, M. Ntreta and A. Ferri, "Evaluating the influence of the Siemens IGRT carbon fibre tabletop in head and neck IMRT," Radiother Oncol 89, 114-122 (2008). 113W. Myint, M. Niedbala, D. Wilkins and L. Gerig, "Investigating treatment dose error due to beam attenuation by a carbon fiber ta bletop," J Appl Clin Med Phys 7, 21-27 (2006).

PAGE 155

BIOGRAPHICAL SKETCH Guanghua Yan was born in Xiantao, China. He received his bachelors degree in computer science in July 1998 from Huazhong University of Science and Technology and his masters degree in computer scie nce in July 2001 from the same university. In August 2002, he enrolled in the Department of Computer and Information Science and Engineering at the University of Florida. He jo ined the Department of Nuclear and Radiological Engineering at the same university for a comb ined degree program majoring in medical physics in January 2006. He received masters degr ee in computer engineering in December 2007, masters degree in nuclear engineering sciences in May 2008 and the degree of Doctor of Philosophy in nuclear engineering sciences in May 2009. His research in medical physics was conducted under the guidance of Drs. Jonathan Li and Chihray Liu. His research interests are in intensity-modul ated radiation therapy dose calculation and quality assurance.