<%BANNER%>

Alternative Techniques of Backscatter Radiography

University of Florida Institutional Repository
Permanent Link: http://ufdc.ufl.edu/UFE0025092/00001

Material Information

Title: Alternative Techniques of Backscatter Radiography Snapshot Aperture Backscatter Radiography and Collimated Segmented Detector Scatter X-Ray Imaging
Physical Description: 1 online resource (111 p.)
Language: english
Creator: Bougeant, Olivier
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2009

Subjects

Subjects / Keywords: aperture, backscatter, collimated, compton, csd, csdsxi, detector, imaging, radiography, rays, sabr, scatter, segmented, snapshot, sxi, x, xrays
Nuclear and Radiological Engineering -- Dissertations, Academic -- UF
Genre: Nuclear Engineering Sciences thesis, M.S.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Unlike standard transmission radiography, Compton Backscatter Imaging (CBI) techniques are non-destructive examination methods that rely on the detection of X-ray photons backscattered in the target object. They have the advantage of being single-sided imaging techniques and can yield better images than transmission radiography for certain applications. The X-ray backscatter imaging system currently used at the University of Florida employs a method called Radiography by Selective Detection (RSD). It uses an X-ray pencil beam to illuminate the target object while scintillation detectors positioned around the X-ray source count backscattered photons. As the X-ray beam scans the target object, one real-time 2D image is created per detector based on the recorded counts. Lead collimation sleeves placed around the detector prevent particles scattered above a given depth from being detected, and help provide good depth information in RSD images. This system has been commercialized and can be used for detection of land mines, security inspections and detection of defects or foreign object debris. One of the drawbacks of this technique, however, is the image acquisition time, especially when detectors are highly collimated. The two X-ray backscatter imaging techniques presented in this thesis were originally designed to yield images with a shorter acquisition time. Snapshot Aperture Backscatter Radiography (SABR) is a single-sided Compton Backscatter Imaging technique that is based on a snapshot acquisition method, contrary to most other X-ray Backscatter Imaging systems which generate images by scanning. This characteristic of the SABR technique greatly reduces image acquisition time. The detector used with this technique is a CR plate shielded from direct X-ray radiations by a lattice of lead tiles: X-rays illuminate the target object through apertures between the lead tiles and are backscattered toward the CR plate. However, both Monte Carlo simulations and actual experiments have shown that this technique, with the employed aperture arrangements, yields images with poor depth information. Collimated Segmented Detector-based Scatter X-ray Imaging (CSD-SXI) is a new backscatter radiography technique. Its principle relies on the use of a pixelated detector, collimated by a fine grid of a strongly absorbing material. The X-ray source comes in the form of a fan beam, parallel to the segmented detector. Monte Carlo simulations and the first practical experimental tests have shown very promising results in both image quality and depth information
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 Olivier Bougeant.
Thesis: Thesis (M.S.)--University of Florida, 2009.
Local: Adviser: Dugan, Edward T.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2010-08-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2009
System ID: UFE0025092:00001

Permanent Link: http://ufdc.ufl.edu/UFE0025092/00001

Material Information

Title: Alternative Techniques of Backscatter Radiography Snapshot Aperture Backscatter Radiography and Collimated Segmented Detector Scatter X-Ray Imaging
Physical Description: 1 online resource (111 p.)
Language: english
Creator: Bougeant, Olivier
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2009

Subjects

Subjects / Keywords: aperture, backscatter, collimated, compton, csd, csdsxi, detector, imaging, radiography, rays, sabr, scatter, segmented, snapshot, sxi, x, xrays
Nuclear and Radiological Engineering -- Dissertations, Academic -- UF
Genre: Nuclear Engineering Sciences thesis, M.S.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Unlike standard transmission radiography, Compton Backscatter Imaging (CBI) techniques are non-destructive examination methods that rely on the detection of X-ray photons backscattered in the target object. They have the advantage of being single-sided imaging techniques and can yield better images than transmission radiography for certain applications. The X-ray backscatter imaging system currently used at the University of Florida employs a method called Radiography by Selective Detection (RSD). It uses an X-ray pencil beam to illuminate the target object while scintillation detectors positioned around the X-ray source count backscattered photons. As the X-ray beam scans the target object, one real-time 2D image is created per detector based on the recorded counts. Lead collimation sleeves placed around the detector prevent particles scattered above a given depth from being detected, and help provide good depth information in RSD images. This system has been commercialized and can be used for detection of land mines, security inspections and detection of defects or foreign object debris. One of the drawbacks of this technique, however, is the image acquisition time, especially when detectors are highly collimated. The two X-ray backscatter imaging techniques presented in this thesis were originally designed to yield images with a shorter acquisition time. Snapshot Aperture Backscatter Radiography (SABR) is a single-sided Compton Backscatter Imaging technique that is based on a snapshot acquisition method, contrary to most other X-ray Backscatter Imaging systems which generate images by scanning. This characteristic of the SABR technique greatly reduces image acquisition time. The detector used with this technique is a CR plate shielded from direct X-ray radiations by a lattice of lead tiles: X-rays illuminate the target object through apertures between the lead tiles and are backscattered toward the CR plate. However, both Monte Carlo simulations and actual experiments have shown that this technique, with the employed aperture arrangements, yields images with poor depth information. Collimated Segmented Detector-based Scatter X-ray Imaging (CSD-SXI) is a new backscatter radiography technique. Its principle relies on the use of a pixelated detector, collimated by a fine grid of a strongly absorbing material. The X-ray source comes in the form of a fan beam, parallel to the segmented detector. Monte Carlo simulations and the first practical experimental tests have shown very promising results in both image quality and depth information
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 Olivier Bougeant.
Thesis: Thesis (M.S.)--University of Florida, 2009.
Local: Adviser: Dugan, Edward T.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2010-08-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2009
System ID: UFE0025092:00001


This item has the following downloads:


Full Text





ALTERNATIVE TECHNIQUES OF BACKSCATTER RADIOGRAPHY:
SNAPSHOT APERTURE BACKSCATTER RADIOGRAPHY AND
COLLIMATED SEGMENTED DETECTOR SCATTER X-RAY IMAGING



















By

OLIVIER BOUGEANT


A THESIS PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
MASTER OF SCIENCE

UNIVERSITY OF FLORIDA

2009


































2009 Olivier Bougeant

































To my family: Alain, Christiane, Carole and Nathalie, and to Verena









ACKNOWLEDGMENTS

I would like to thank Dr. Edward Dugan, my advisor, for his precious guidance and for

sharing with me his knowledge and his wisdom during the past two years. I thank Dr. James

Baciak for being on the committee.

I would also like to thank Dan Shedlock who offered valuable help, Dan Ekdahl, Georgi

Gorgiev, and my colleagues and friends: Nissia Sabri, Christopher Meng and Kara Beharry. I

thank all the students, professors and staff members at the department of Nuclear and

Radiological Engineering for so warmly welcoming me and for making me feel at home in

Florida.

Finally, I would like to thank my family and all my friends for their unconditional support.

I especially want to thank Verena, my girlfriend, for her love and her patience.

Financial acknowledgment:

University of Florida, Department of Nuclear and Radiological Engineering
Nucsafe










TABLE OF CONTENTS

page

A C K N O W L ED G M EN T S .......................................................................... 4

LIST OF FIGURES......................................................... ........ 7

ABSTRACT ..................................................................................................................12

CHAPTER

1 IN T R O D U C T IO N .................................................................................................... .. ............ 14

Radiography by Selective Detection............................. .............. ............... 14
Limits to Radiography by Selective Detection................................ 15

2 SABR: INTRODUCTION.................... ....................... ......... 19

Snapshot B ackscatter R adiography ........................................................................ 19
Principle of SABR ..................................... ............... 20
Experimental Setup for SABR .......................................................................... 21

3 SABR: IMPORTANT FACTORS ON IMAGE QUALITY................. ...........26

Effects of the X-ray Generator Configuration........................................................... 26
Exposure ...................................... .......................................... 26
V o lta g e ....................... .. ... .. ......... .. .................................................. 2 7
Notes on MCNP Photon Source for SABR Simulations........................................... 28
E effects of the Sub state ...................................................................... 30
Effects of the Size of the L ead Tiles ............................................................................. ............. 32
Effect of a Gap Between the CR Plate and the Object .................................... 33
Effect of a Gap Between the Object and the Substrate......................................................34

4 SABR: VARIOUS OBJECTS THAT HAVE BEEN IMAGED ................ .... ...........45

SABR Images of a Spray-on-Foam Insulation Block........................................................46
SABR Images of a Corroded Piece of Aluminum ........................................... 47
Other Objects that have been Imaged with the SABR Technique ........................................49

5 SA B R : C O N C L U SIO N S ................................................................................................. 56

6 CSD-SXI: INTRODUCTION .......................................... 58

Computed Image Backscatter Radiography ................................................... 58
Backscatter Radiography using an Uncollimated Segmented Detector ...............................58









7 CSD-SXI: PRINCIPLE AND MONTE CARLO SIMULATIONS .................... .......... 65

Principle of Collimated Segmented Detector Scatter X-ray Imaging................ 65
R solution .................. ........................................................ ........ 68
Influence of Collimation ..................... ............. ........ 69
Depth information and Possibility of 3D Imaging............................ ......... ........ 71
2D Collimation Grid for Better Depth Resolution ..................... ............... 71
Possibility of 3D Backscatter Imaging................................................................... 72
CSD-SXI 3D Imaging: Example of a Ring of Air Inside Aluminum ..............................75

8 CSD-SXI: CONSTRUCTION AND TEST OF A FIRST PROTOTYPE ...............................92

E x p erim en ta l S etu p ..................................................................................................................... 9 2
Segm ented D etector.................................................................................. .................. 92
Collim action Grid ...................................... ......... ........ ...... 93
X-ray fan Beam Source .. ............................................ ............... 94
Test of the First CSD -SXI Prototype ............................................................. ... ....94
C o n tra st a n d D etails....................................................................................................... 9 4
Resolution and Modulation Transfer Function (MTF) for CSD-SXI .............................. 95
D epth Penetration ............. ............................... 96

9 CSD-SXI: CONCLUSIONS AND FUTURE WORK ...................... ............... 108

L IST O F R E F E R E N C E S ................................................................................................................. 1 10

BIOGRAPHICAL SKETCH ......................................... 111









LIST OF FIGURES


Figure page

1-1 Principle of backscatter Radiography by Selective Detection ................ ............... 17

1-2 Nal and YSO detectors mounted on a pencil beam RSD system................ 17

1-3 Scanning pattern of pencil beam RSD system s................................................ 18

2-1 Letters of lead on a nylon substrate............................................... 23

2-2 Principle of Shadow Aperture Backscatter Radiography .................................................23

2-3 Differential scattering cross section per unit solid angle at 1 keV, 100 keV and 2
M eV............... ................... ................... ................................ ......... 24

2-4 SR 115 portable X-ray generator ......................................................................... 24

2-5 Typical energy spectrum of a medical X-ray generator at 75 kVp with and without
aluminum filter .................. .......... ................... 25

2-6 Assemblies of 1", 1.5" and 2" lead tiles.................... ...... ..................... 25

3-1 Foreign O object D ebris on a nylon substrate ...................................... ............................. 36

3-2 Geometry of the MCNP simulation of the Shadow Aperture Backscatter
Radiography of a steel washer on a nylon substrate. .............. ................. ............36

3-3 Flux of photons received after a MCNP simulation by each point of the CR plate for
three photon energy ..................................................... 36

3-4 Flux of photons received after a MCNP simulation by each 0.5 pixel of the CR plate.....37

3-5 Steel Washer on nylon background.................................... ............... 37

3-6 Foreign Object Debris on a nylon substrate .............................................................. 38

3-7 MCNP simulated SABR images .......... .............................38

3-8 Foreign Object Debris on an aluminum substrate.......................... .............39

3-9 Profile plot of the previous SABR image on which can be seen two features (the
hole in the large washer and a small washer). .................................. 39

3-10 Foreign Object D ebris on a lead substrate....................................................... .... ....40

3-11 MCNP simulated SABR images of a steel washer .......................... ..........40









3-12 Foreign Object Debris on a nylon substrate ................ ...............41

3-13 MCNP simulated SABR images of a steel washer on a nylon background..................41

3-14 SABR images of Foreign Object Debris on a nylon background at 75 kVp, 60mAs,
at 25 inches .......... ............................. ........ 42

3-15 Possible paths of photons backscattered in the target object.............................................42

3-16 Geometry of the MCNP simulation of the Shadow Aperture Backscatter
Radiography of a steel washer on a nylon substrate, with a 4 mm gap between the
w asher and the CR plate ........................................ 43

3-17 MCNP simulated SABR images of a steel washer on a nylon background..................43

3-18 Nylon washer suspended by a thread over a lead background............. ................43

3-19 Geometry of the MCNP simulation of the Shadow Aperture Backscatter
Radiography of a steel washer located 3.5 inches over a nylon substrate..........................44

3-20 MCNP simulated SABR images of a steel washer on a nylon background..................44

4-1 SABR of Foreign Object Debris on a nylon background taken at 75 kVp, 60 mAs, 25
inches ......................................... ............... 50

4-2 MCNP simulated SABR images of a steel washer on a nylon background................. 50

4-3 SABR of Foreign Object Debris on a nylon background taken at 75 kVp, 60 mAs, 25
inches .. ............................................... ........ 51

4-4 Block of spray-on-foam insulation.................................... ............... 51

4-5 Five holes, 0.75 inch in diameter and of different depth in a spray-on-foam block..........52

4-6 Five dimes placed in holes drilled in foam, 0.75 inch in diameter and of different
depth .............. ................... ............................ ...................................... 52

4-7 Zoom of the area of the previous SABR image that contained the third coin. .................. 53

4-8 Corroded aluminum plate on a nylon background...... ......................53

4-9 Corroded aluminum plate suspended 3.5 inches over a nylon background ..................54

4-10 Corroded aluminum plate covered by a 1mm thick aluminum plate and suspended
3.5 inches over a nylon background......................... ...... ....... ........ 54

4-11 Image enhancement of the previous SABR image with features surrounded in red .........55

6-1 Backscatter radiography images with a 1 mm resolution of letters of lead on nylon........62









6-2 Geometry of the MCNP simulation of the backscatter radiography of a lead strip on a
nylon background using an uncollimated segmented detector................ ........... .... 62

6-3 Energy spectrum of the X-ray source at 65 kVp....................................63

6-4 Geometry of the MCNP simulation of the pencil beam backscatter radiography of a
lead strip on a nylon background ........................................ 63

6-5 Comparison between the normalized fluxes observed in the uncollimated array
detector with a fan beam source and with a pencil beam source. ..............................64

6-6 Possible paths of backscattered X-ray photons toward the uncollimated segmented
detector with a fan beam source. ......... ................... ................ 64

7-1 Geometry of the MCNP simulation of a Collimated Segmented Detector Scatter X-
ray Image of a lead strip on a nylon background.................................. 79

7-2 Possible paths of backscattered X-ray photons from a fan beam source toward the
segm ented detector................................................................................ 79

7-3 Comparison between the normalized fluxes observed in the collimated array detector
with a fan beam source and with a pencil beam source for a lead strip on the surface
of a nylon block ................. ................... ..................... ..........................80

7-4 Possible paths of X-ray photons eventually reaching an image pixel. ........................... 80

7-5 Measure of the resolution for Collimated Segmented Detector Scatter X-ray Imaging.... 81

7-6 Backscatter X-ray secondary source as seen from point y = t....................................... 81

7-7 Normalized contribution to a 0.6 mm wide pixel, with a 1 cm collimation.....................82

7-8 Geometry of the MCNP simulation of a Collimated Segmented Detector Scatter X-
ray Image of a lead strip 2 cm deep inside a nylon block......................... ..... 82

7-9 Comparison between the normalized fluxes observed in the collimated array detector
with a fan beam source and with a pencil beam source for a lead strip 2 cm deep in a
nylon block ........................... ................ ............... 83

7-10 Comparison between the normalized fluxes observed in the collimated array detector
with a fan beam source for four different collimation lengths for a lead strip 2 cm
deep in a nylon block .................................................... 83

7-11 Geometry of the MCNP simulation of a Collimated Segmented Detector Scatter X-
ray Image of a 1 cm deep air gap inside an aluminum block with a 2D grid
collimated detector at a 450 inclination ............... ....................... ............. 84

7-12 Geometries of the 3 MCNP simulations of an air gap inside an aluminum block,
view s of th e X Z -plan e.............................................................................................................. 84









7-13 Comparison between the normalized fluxes from MCNP simulations of a 1 cm deep
air gap inside an aluminum block with a fan beam source...................................... 85

7-14 Geometry of the 2D collimated array detector at a 450 inclination, view of the XZ-
plane ............................................ ................ 85

7-15 Backscatter X-ray secondary source as seen from point t..................................... 86

7-16 Normalized contribution of each z-level in the target object to all 6 rows of pixels
and to the sum of the rows............. .......... ..................... 87

7-17 Comparison between the fluxes recorded by rows 1, 2, 3, 4, 5 and 6 of the 2D
collimated array detector, with a 450 inclination from an MCNP simulation of CSD-
SX I of an air gap inside an alum inum block ............................................................. ..... 87

7-18 Geometry of the ring of air located 1 cm deep inside aluminum............................. .. 88

7-19 Geometries of the three different imaging systems..................................................... 88

7-20 MCNP generated X-ray backscatter images of a ring of air inside aluminum using a
Nal detector with a pencil beam source with the following collimation lengths............ 89

7-21 MCNP generated X-ray backscatter images of a ring of air inside aluminum using a
linearly collimated segmented detector with a fan beam source with the following
collimation lengths .. ............................................... 89

7-22 MCNP generated X-ray backscatter images of a ring of air inside aluminum using a
2D collimated segmented detector at a 450 inclination with a fan beam source............. 90

7-23 Z-levels corresponding to each of the six MCNP generated images of the ring of air......91

8-1 Photograph of the 12" Envision Product Design segmented detector ................................ 98

8-2 Simplified diagram of the segmented detector used for CSD-SXI............... ............... 98

8-3 Segmented detector mounted on the X-ray tube at a 400 angle .......................................99

8-4 Flux detected by the linearly segmented detector with and without the X-rays on..........99

8-5 Collimation grid made of 0.4 mm thick lead.............................. 100

8-6 Lead collimation grid mounted on the bottom surface of the segmented detector.......... 100

8-7 CSD-SXI image of letters of lead (1 mm thick) on nylon............................................... 101

8-8 Lead collimation grids made of 0.4 mm and 1.08 mm thick lead plates and spacers......101

8-9 Explanation of the apparition of artifacts in CSD-SXI images when the collimation
grid pitch does not match the size of pixel bins ............................................................ 102









8-10 CSD-SXI images of a disk of lead on nylon ................................... 102

8-11 Lead shield w ith a 1 m m slit.................................................... 103

8-12 Lead shield with variable aperture, in this case 0.2 mm.......................................... 103

8-13 Various objects on an aluminum background ........................................ 104

8-14 Letters of lead on a nylon background.............................................. 104

8-15 Lead test pattern on an aluminum background ....................................................... 105

8-16 Modulation Transfer Functions (MTF) for the CSD-SXI and RSD images shown in
Figure 8-15.. ............................................... 106

8-17 Various objects on an aluminum background.................................... 106

8-18 Various objects on an aluminum background.................................... 107

8-19 Various objects on an aluminum background.................................... 107









Abstract of Thesis Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Master of Science

ALTERNATIVE TECHNIQUES OF BACKSCATTER RADIOGRAPHY:
SNAPSHOT APERTURE BACKSCATTER RADIOGRAPHY AND
COLLIMATED SEGMENTED DETECTOR SCATTER X-RAY IMAGING

By

Olivier Bougeant

August 2009

Chair: Edward T. Dugan
Major: Nuclear Engineering Sciences

Unlike standard transmission radiography, Compton Backscatter Imaging (CBI) techniques

are non-destructive examination methods that rely on the detection of X-ray photons

backscattered in the target object. They have the advantage of being single-sided imaging

techniques and can yield better images than transmission radiography for certain applications.

The X-ray backscatter imaging system currently used at the University of Florida employs

a method called Radiography by Selective Detection (RSD). It uses an X-ray pencil beam to

illuminate the target object while scintillation detectors positioned around the X-ray source count

backscattered photons. As the X-ray beam scans the target object, one real-time 2D image is

created per detector based on the recorded counts. Lead collimation sleeves placed around the

detector prevent particles scattered above a given depth from being detected, and help provide

good depth information in RSD images. This system has been commercialized and can be used

for detection of land mines, security inspections and detection of defects or foreign object debris.

One of the drawbacks of this technique, however, is the image acquisition time, especially when

detectors are highly collimated. The two X-ray backscatter imaging techniques presented in this

thesis were originally designed to yield images with a shorter acquisition time.









Snapshot Aperture Backscatter Radiography (SABR) is a single-sided Compton

Backscatter Imaging technique that is based on a snapshot acquisition method, contrary to most

other X-ray Backscatter Imaging systems which generate images by scanning. This characteristic

of the SABR technique greatly reduces image acquisition time. The detector used with this

technique is a CR plate shielded from direct X-ray radiations by a lattice of lead tiles: X-rays

illuminate the target object through apertures between the lead tiles and are backscattered toward

the CR plate. However, both Monte Carlo simulations and actual experiments have shown that

this technique, with the employed aperture arrangements, yields images with poor depth

information.

Collimated Segmented Detector-based Scatter X-ray Imaging (CSD-SXI) is a new

backscatter radiography technique. Its principle relies on the use of a pixelated detector,

collimated by a fine grid of a strongly absorbing material. The X-ray source comes in the form of

a fan beam, parallel to the segmented detector. Monte Carlo simulations and the first practical

experimental tests have shown very promising results in both image quality and depth

information.









CHAPTER 1
INTRODUCTION

The purpose of this work is to present two new X-ray backscatter imaging techniques:

Snapshot Aperture Backscatter Radiography (SABR) and Collimated Segmented Detector

Scatter X-ray Imaging (CSD-SXI), both originally designed to yield faster images than existing

Compton Backscatter Imaging (CBI) systems currently used at the University of Florida, while

bringing additional information.

Radiography by Selective Detection

Backscatter Radiography by Selective Detection (RSD) is a single-sided imaging

technique developed by the Scatter X-Ray Imaging (SXI) group at the University of Florida. Its

principle is shown in Figure 1-1. X-rays are emitted in the form of a pencil beam from an X-ray

tube toward the object to be imaged. Sodium Iodide (Nal) and Yttrium Orthosilicate (YSO)

scintillation detectors (Figure 1-2) placed around the X-ray tube then record the number of

photons backscattered toward them for a particular pencil beam position. As the system scans

across the target object, a real-time image is formed based on the counts recorded by each

detector.

Lead collimation sleeves, positioned around the scintillation detectors prevent the detection

of photons backscattered above a certain depth, called the collimation plane. Thanks to this

feature, RSD can yield high quality images of features located at selected depths inside the target

object.1

Applications of backscatter Radiography by Selective Detection include, among other

things, detection of land mines and Homeland Security inspections. It is, however, currently used

for detection of flaws and defects, such as cracks, voids and corrosion in a wide variety of

materials including aluminum, steel, concrete, carbon-carbon composites and Spray-On-Foam-









Insulation (SOFI). In particular, since 2004, six RSD systems have been used by NASA and the

Lockheed Martin Space Systems Co. to detect flaws in the foam insulation on the external tank

of the space shuttle prior to each launch.2 Such flaws caused parts of this foam to strike the wing

of the Columbia Space Shuttle in 2003 shortly after lift-off, damaging the shuttle's heat shield.

RSD is also used for the detection of Foreign Object Debris (FOD).

Limits to Radiography by Selective Detection

Because RSD is based on a pencil beam X-ray source, it allows obtaining very accurate

backscatter images, with sub-millimeter resolution. In order to detect small differences in

contrast between features of a target object, the RSD technique needs to achieve a high count

rate to reduce the statistical uncertainty associated with the measurement. However, because this

technique requires a narrow pencil beam source, it is necessary to shield the majority of the

photons as they exit the X-ray tube to reduce the dispersion in the pencil beam. As only a

fraction of the photons produced by the X-ray tube is used, RSD systems must spend sufficient

time on every pixel of the image to ensure that the statistical uncertainties are limited. Total

scanning time is obviously an important factor for an imaging technique, and much of the

research work accomplished by the SXI group is aimed at reducing the image acquisition time.

Nucsafe, a company based in Oak Ridge, Tennessee, which is working with the University of

Florida, has mobile pencil beam systems capable of imaging over a square meter per minute and

portable systems capable of imaging several square meters per minute. However, these fast

systems do not use collimated detectors and as a result, there is only limited depth information.

The scanning pattern for RSD is shown in Figure 1-3. The X-ray source sweeps across the

target object to record the counts for one line of pixels, before going to the next line and so on

until the whole area has been imaged. To scan a square foot area with a 1 mm resolution,

assuming that the required illumination time per pixel is 0.1 second, the total image acquisition









time with the RSD system currently in place at the University of Florida, and constructed in

2004, would be two hours and thirty five minutes.

The two backscatter radiography systems presented in this thesis, Snapshot Aperture

Backscatter Radiography (SABR) and Collimated Segmented Detector Scatter X-ray Imaging

(CSD-SXI) were designed in an effort to reduce the acquisition time.
































Figure 1-1. Principle of backscatter Radiography by Selective Detection.


Nal detectors






Lead collimation -
sleeves


YSO detectors







X-ray pencil
L beam


Figure 1-2. Nal and YSO detectors mounted on a pencil beam RSD system.









-. . .
-= -


-^^E37


_ __ -


Figure 1-3. Scanning pattern of pencil beam RSD systems.


















18


Y

X









CHAPTER 2
SABR: INTRODUCTION

Snapshot Backscatter Radiography

The SABR technique is a method of X-ray backscatter imaging originally based on the

Snapshot Backscatter Radiography (SBR), a technique developed by the Scatter X-Ray Imaging

group at the University of Florida.3

The SBR technique obtains backscatter, single-sided images of an object without scanning,

thus reducing the image acquisition time. SBR images are obtained by placing a Computed

Radiography (CR) plate directly on the top the target object, which could, for example, be

composed of Foreign Object Debris (FOD) placed on a background, and then exposing this

target to an X-ray snapshot through the CR plate. As a result the CR plate is exposed a first time

by the X-ray photons which are then scattered in the background and can be absorbed in the

FOD placed under the CR plate. Depending on the scattering-to-absorption ratio of the

background and the FOD objects, a given fraction of these photons reach the Computed

Radiography plate. The background, or substrate, behaves like a secondary source of X-rays and

the Foreign Object Debris shadows the CR plate from this source. This is why the best images

were obtained with a highly scattering substrate and strongly absorbing objects.

A CR plate is a film-like plate that is made of photo-stimulable storage phosphors. As the

X-rays, emitted by the X-ray generator, strike the phosphor's atoms; the electrons are excited to a

higher energy level. Then a CR plate reader (in this case, a Kodak INDUSTREX ACR-2000

Digital System) scans the plate with a laser that causes the electrons to go back to their ground

state. In the process, they release visible light photons that the reader collects and counts in order

to compute a digital image.









However, the SBR technique often results in an overexposure of the CR plate and a very

low signal-to-noise ratio because the backscatter signal is superimposed onto the transmission

signal on the image. Figure 2-1 shows the photograph of a target object and the corresponding

SBR image. In this case, letters of lead were placed on a nylon background to obtain the highest

possible contrast as the nylon is a highly scattering material and the lead is a strong absorber.

This image was taken at 50 kVp (peak kilovoltage) with a 2.85 mAs exposure, with the X-ray

source placed 23 inches away from the target object.

Principle of SABR

Because of the low signal-to-noise ratio of SBR based images, the SXI group developed a

new technique, called Shadow Aperture Backscatter Radiography, which is based on the SBR

method but in which the backscatter signal is not superimposed onto the transmission signal.3

The principle of the SABR technique is shown in Figure 2-2. As for the SBR method, a CR

plate is placed on the object that is to be imaged. However, in the case of the SABR technique,

the CR plate is covered by tiles of lead to prevent it from being completely saturated by

illumination photons. Instead, the illumination X-rays can only reach the CR plate and the target

object through apertures between the lead tiles. These X-ray photons are then absorbed in the

object or backscattered toward the parts of the CR plate that are shadowed by the lead tiles.

As a result, the contrast in the shadowed parts of the CR plate is greatly enhanced when

compared to the images obtained with the SBR technique, but while the exposure of most parts

of the image is good, there are some saturation (white) lines on the image corresponding to the

illumination apertures between the lead tiles. Consequently, these parts of the images cannot be

used to detect features directly under the CR plate. This problem can be overcome by simply

acquiring one image of an object, and then by shifting the lead tiles, and by reacquiring another

image to make sure that all the missing parts from the first image can be seen on the second one.









However, because the intensity received by each point of the CR plate decreases

exponentially with its distance to the apertures, image reconstruction is complex and two images

cannot be simply overlapped to get rid of the white lines.

The energy range used in this experiment was roughly from 0 to 100 keV, which

corresponds to energies at which the Compton scattering is still relatively isotropic, as shown in

Figure 2-3, whereas higher energy photons experience strongly forward peaked scattering.4 This

range of energy allows a large proportion of X-ray photons to be backscattered toward the CR

plate, while still allowing them to travel through a moderately absorbing medium, such as

aluminum for instance.

Experimental Setup for SABR

The X-ray source used for SABR was a Source Ray SR 115 portable x-ray generator

(Figure 2-4) which allowed maximum photon spectrum energies from 40 to 100 keV. The

designation used for such spectra is e.g., 40 kVp or 100 kVp (meaning 40 or 100 keV peak). The

energy spectrum of such an X-ray source usually resembles a Maxwell-Boltzmann distribution

with average photon energy of about 40% the maximum energy. Figure 2-5 displays a typical

medical X-ray source energy spectrum at 75 kVp with a 2.7 mm aluminum filter, with average

photon energy of 39 keV. This graph does not represent real measurements done on the Source

Ray X-ray generator used for the SABR experiments but is an accurate energy spectrum for a

standard X-ray source.

The single shot exposure on this X-ray generator can vary between 0.15 mAs and 60 mAs,

but it was possible to take several shots to obtain even higher exposures (the cooling time

between two 60 mAs shots is about two minutes).

The CR plate used was a Kodak GP Digital Imaging Plate SO-170, which is about 0.6 mm

thick and which is mainly composed of a layer of bariumfluorobromoiodide (BaFBr) protected









by a thin polyester coat.5 In the MCNP calculations detailed later in this report, the CR plate was

assumed to be only made of BaFBr. The density of this type of CR plate is roughly 5.0 g/cm3.

Although a CR plate can theoretically be scanned thousands of times if handled with extra

caution, the plates used were not in perfect condition and this resulted in artifacts in some of the

images.

To obtain a SABR image, the target objects, which are generally composed of various

objects on a scattering substrate, is placed on a lead sheet that is laid on a steel table. Then, the

CR plate is put directly on the objects that are to be imaged. Finally, an assembly of tiles of lead

is placed over the CR plate. Three different assemblies of tiles were used to shade the CR plate

and create the apertures to allow illumination of the objects. Each assembly consisted of 1 mm

thick square tiles of lead glued next to each other on a sheet of paper with an average spacing of

1 mm between the tiles for the apertures. The square tiles of the three assemblies were about 1,

1.5 and 2 inches long (Figure 2-6).






















A B

Figure 2-1. Letters of lead on a nylon substrate: A) Photograph B) SBR image at 50 kVp, 2.85
mAs and 23 inches between the X-ray source and target.


X-ray source


CR plate


Tiles of lead


Figure 2-2. Principle of Shadow Aperture Backscatter Radiography.

















barn





obaMeV.e)
bam


Figure 2-3. Differential scattering cross section per unit solid angle at 1 keV, 100 keV and 2
MeV.4


Figure 2-4. SR 115 portable X-ray generator.6








4.00E-02
3.50E-02
3.00E-02
2.50E-02
2.00E-02
1.50E-02


5.00E-023
O.&0E+00


Relative intensity of X-rays











B


- wit ho ut fi lt ri r
with 2.7 mm Al filter


20 30 40 50
Photan energyin keV


60 70


Figure 2-5. Typical energy spectrum of a medical X-ray generator at 75 kVp with and without
aluminum filter.


Figure 2-6. Assemblies of I", 1.5" and 2" lead tiles.









CHAPTER 3
SABR: IMPORTANT FACTORS ON IMAGE QUALITY

Effects of the X-ray Generator Configuration

There are several factors that can affect the quality of images obtained with the SABR

technique. Perhaps the most important ones are the maximum energy of the X-ray photons and

the exposure. These factors can be modified directly on the SR-115 X-ray generator.

Exposure

The exposure, measured in mAs (on the SR-115, the current is fixed at 15 mA, and the

exposure time varies between 0.01 and 4.0 seconds6), is very important for obtaining a

reasonable amount of photons to avoid under or overexposure of the CR plate. This factor is also

strongly linked to the distance between the X-ray source and the CR plate. Indeed, because X-ray

photons are roughly emitted isotropically at the anode in the X-ray tube for energies below 100

kVp, the number of photons reaching the CR plate is proportional to 1/R2, where R is the

distance between the source and the CR plate. For example, by increasing the distance between

the X-ray source and the CR plate by a factor of two, the exposure needs to be increased by a

factor of four to obtain roughly the same image.

Affirmation of this behavior is demonstrated in Figure 3-1. A number of metallic objects (a

lead wedge, some steel washers and a penny shown in Figure 3-1A) were placed on a nylon

substrate which is a very good scattering material. A first SABR image (Figure 3-1B) was

obtained at 75 kVp with an exposure of 60 mAs and with the CR plate located 25 inches away

from the X-ray source. The exposure of this image is, given the very high sensitivity of CR

plates, very close to the exposure of the second image (Figure 3-1C) obtained with the SABR

method at 240 mAs and with a 47 inch distance between the source and the CR plate (47 inches

is the maximum possible distance between the source and the top of the table that was used), still









at 75 kVp. The exposure and voltage were chosen to obtain the best possible images. The vertical

and horizontal white lines that can be seen on the two SABR images correspond to the

illumination apertures.

Voltage

The effects of voltage variations are even more important. For instance, at 50 kVp, with a

source-to-target distance of 25 inches, the objects on a nylon surface are only visible for

exposures above 240 mAs, and with an aluminum substrate, only the white lines are visible at

exposures as high as 360 mAs; the rest of the CR plate remains completely dark. This can be

explained by the fact that at lower energies (below 20 keV), the dominant collision type of

photons in the nylon and the aluminum is photoelectric absorption, and as a result a lower

fraction of X-rays can be backscattered in the substrate through the various objects and toward

the CR plate.

On the other hand, at energies greater than 80 kVp, for tiles of lead 1 mm thick, there is too

much transmission through the 1 mm thick lead shadow shields and the signal-to-noise ratio is

decreased because the backscatter signal is superimposed onto the transmission signal. This is

because, at energies higher than 80 keV, a significant fraction of photons have a mean free path

in lead on the order of 1 mm. For instance, for a photon of 30 keV, which is roughly the average

energy of X-ray photons for a source voltage of 70 kVp, the mean free path in lead is about 0.03

mm7; so, at this energy, very few photons actually go through the lead. However, at 90 kVp, a

non-negligible fraction of photons have energies higher than 80 keV for which the mean free

path in lead is about 0.3 mm, which is of the same order as the thickness of the lead tiles.

MCNP calculations were performed to confirm the impact of photon energy on the SABR

image contrast. MCNP is a Monte Carlo particle simulation code developed by the Los Alamos

National Lab. Figure 3-2 shows the MCNP geometry used to simulate the SABR image of a steel









washer on a nylon substrate. The air is represented in purple, the lead in dark blue, the CR plate

in light blue, the nylon in yellow and the steel washer in green. The source for this geometry was

a circular surface source placed above the lead tiles and which emitted photons at a given energy

directed downward (Because the point source for the SABR experiment was 25 inches away, and

the total size of the nylon background was 6 inches, it can be safely assumed that all photons had

the same direction).

Figure 3 -3 shows the flux received by each point of the CR plate when the source emitted

500 million photons at 10 keV, 50 keV and 100 keV. The average relative error per pixel of the

background for the simulation at 50 keV is close to 50%, whereas, on a SABR image at 75 kVp,

60 mAs and 25 inches, the average statistical uncertainty per pixel is 3%. Therefore, it can be

estimated that 500 millions photons in the MCNP simulation correspond to a SABR image with

an exposure of about 0.2 mAs only. However, due to the lack of sensitivity of the CR plate,

SABR images taken at 0.2 mAs appear dark. The parts of the image in white have not received

any photon during the MCNP calculations, each of which lasted roughly 90 minutes, and would

be represented by very dark pixels on the corresponding SABR image. As explained before, the

image with the best contrast is by far the one at 50 keV, which is close to the mean energy used

with the SABR technique. The pixels size for each image was 0.5 mm, and each lead tile

measures 2 inches. Because the circular source of the MCNP simulation did not cover the whole

area, the comers are underexposed compared to the rest of the image.

Notes on MCNP Photon Source for SABR Simulations

Although, the image quality at 50 keV is better than at 10 or 100 keV, some parts of the

CR plate, corresponding to white pixels were not crossed by any photon during the MCNP

calculation. Moreover, the relative error of the flux for pixels that received very few photons is

close to 100 %, even after hours of MCNP calculation. In fact, this is due to the very low









probability of photons going through the lead tile assembly. Indeed, the aperture holes only take

less than 0.5 % of the total area of the assembly, and the mean free path of 50 keV photons in

lead is about 0.1 mm, which is less than the tenth of the thickness of the lead tiles, so the

probability of a photon to pass through the lead is roughly 4 x 10-5. As a result, more than

99.5% of photons are wasted if a continuous surface source is used.

Consequently, a better source was introduced in all the other MCNP input files. The

particles were then emitted over the aperture holes between the lead tiles only. Figure 3-4 shows

the flux received at each point of the CR plate for the two different source types. The image

shown in Figure 3 -4A was created with the continuous source, with 500 million photons emitted

and the calculation lasted for more than 90 minutes. The other image, shown in Figure 3-4B was

made using the other source, for which only 5 million particles were emitted in less than 50

minutes (the inefficiency of the source sampling is responsible for such a small number of

particles emitted per unit time).

The difference in image quality is explained by the fact that on average, the CR plate

pixels received ten times more particles with the modified source than with the regular one (this

can be seen on the scales). Also the largest uncertainty of flux value for a 0.5 mm wide pixel is

only about 20% with the modified source. Figure 3-5 shows the proximity between the image

observed experimentally with the SABR of a steel washer (Figure 3-5A), and the image created

after a MCNP simulation of a SABR of the same steel washer (Figure 3-5B). This similitude

validates the MCNP geometry and source models used.

Finally, it was observed that typically, variations of voltage between 60 and 80 kVp do not

seem to dramatically change the contrast between the objects and the substrate they are placed

on. In general, for most objects and backgrounds, 75 kVp, 60 mAs and 25 inches are settings that









offer fairly good SABR images and the contrast is rarely improved significantly by changing the

energy or the voltage, as long as the image exposure is acceptable.

Effects of the Substrate

The substrate, or background, is a plate on which the target objects are placed. Its

composition can greatly modify the SABR image. Three types of substrates were used and

compared: a nylon substrate, an aluminum substrate and a lead substrate. The first two offered

fairly good images, because their scattering-to-absorption ratios are high enough. At 30 keV, the

approximate average energy of photons for voltages of about 70 kVp, the scattering-to-

absorption ratio is 0.3 for the aluminum, and 3.0 for the nylon.7 However, the SABR images of

objects placed directly on a lead substrate were very dark, except for the white lines

corresponding to the apertures. This is due to the very low scattering-to-absorption ratio of lead

(0.05 for 30 keV photons). Indeed, the vast majority of photons that are allowed through the

apertures are absorbed in the lead substrate instead of being scattered back toward the CR plate.

In Figure 3-6, can be seen a variety of objects on a 0.5 inch thick nylon substrate and the

resulting SABR image taken at 75kVp with an exposure of 60 mAs, the CR plate being 25 inches

away from the X-ray source. The tiles used to obtain this image were medium sized tiles, of

about 1.5 inches. All the different objects are visible on this SABR image, except the nylon

washer, as could be expected, because the average mean free path of photons in that energy

range in the nylon is about 4 cm, so a very small fraction of X-rays collided in the 2 mm thick

nylon washer. As a result the CR plate was not shaded by this object enough to detect the nylon

washer. It can be noticed that the holes in some of the washers cannot be detected, probably

because of their relative position to the aperture grid (objects that are too close or too far from

the aperture lines tend to appear with less contrast).









MCNP calculations confirm that on nylon or aluminum background, only a relatively

strong absorbing material can be detected. Figure 3-7 shows MCNP simulations of SABR

images of a steel washer (Figure 3-7A) and a nylon washer (Figure 3-7B) laid on a nylon

substrate. The steel washer is clearly visible on the first image, whereas the nylon washer is

invisible on the other image. These MCNP calculations also used five millions 50 keV photons.

Figure 3-8 shows a similar SABR image but this time, the various objects were placed on

an aluminum substrate. The X-ray generator setup was the same: 75 kVp, 60 mAs and at 25

inches. Also, the same medium sized lead tiles were used. It can be clearly noticed that the

quality of this image is not as good as the one with the nylon substrate. This is due to the low

scattering-to-absorption ratio of X-rays in aluminum at this energy range compared to nylon. As

a result, fewer photons are backscattered toward the CR plate and can therefore be detected.

However, because most of the objects laid on the aluminum substrate are strong absorbers they

still appear on the SABR image but with a much lower contrast. It becomes extremely difficult to

see the hole in the large steel washer or even to see the smallest washers. However, with a profile

plot, those features can be easily detected (Figure 3-9).

Contrary to more conventional X-ray backscatter imaging methods, with which the

contrast between objects in aluminum and nylon can be easily detected (the nylon being a strong

scattering material and the aluminum a relatively good absorber), SABR cannot detect a thin

nylon washer on an aluminum plate. What appears very surprising at first can be easily explained

by the fact that objects placed on the substrate are merely shading the CR plate from the photons

backscattered in the substrate. Because the average mean free path for photons between 20 and

50 keV in the nylon is on the order of 4.0 cm, very few photons are absorbed in the nylon washer

which does not provide sufficient shade for it to be detected on the CR plate. In fact, the SABR









technique was never able to detect scattering material with a very small absorption cross section,

which reduces the field of possible applications for the SABR technique to the detection of

strong absorbers, like metals.

Figure 3-10 shows another SABR image of various objects on a 1.5 mm thick lead

background. The setup for this experiment was exactly the same as previously: 75 kVp, 60 mAs

of exposure and a distance between the X-ray source and the CR plate of 25 inches. As can be

seen in the SABR image, not a single object can be detected at all. The fact that these objects do

not appear on the SABR image, is not due to an underexposure of the CR plate for this particular

X-ray generator setting, because gradually increasing the exposure did not help to detect any

object, and the only effect was the broadening of the white lines on the image. In fact, the

impossibility to see these objects is due to the very low scattering-to-absorption ratio of photons

in lead in the 10 keV to 75 keV range (about 0.05).7 Indeed, a large fraction of the photons which

pass through the aperture grid are absorbed in the lead and very few are scattered back toward

the objects, which prevent these objects from being visible.

This SABR experiment was also simulated by MCNP calculations, which gave results very

close to what was observed experimentally. Figure 3-11 shows the flux received by each pixel of

the CR plate after MCNP simulation of Shadow Aperture Backscatter Radiography of a steel

washer on nylon, aluminum or lead substrate. The white pixels in the third image represent areas

that were not crossed by any photon.

Effects of the Size of the Lead Tiles

The size of the lead tiles is also an important factor for SABR image quality. First of all,

the amount of exposure should be increased with the size of the tiles because, for large lead tiles,

a very small fraction of the X-ray photons that are going through the aperture holes backscatter

toward the center part of the tiles. For a smaller tile size, on the other hand, a large part of the









resulting SABR image is overexposed because the surface taken by the aperture grid (white

lines) is increased, and the surface taken by the shadowed areas is decreased (Figure 3-12). The

overall trend is that image quality tends to be better for medium-sized or large tiles.

The effects of the size of the lead tiles observed on the SABR images are in accordance

with MCNP simulations too. On Figure 3-13 can be seen the two simulated images of a steel

washer on a nylon substrate, with 1 and 2 inches lead tiles. For both images, the contrast, defined

as the ratio of the average flux detected over the steel washer to the average flux detected over

the nylon background, is equivalent (steel-to-nylon ratio of about 0.5%). However, a much larger

area of the CR plate is overexposed (red lines) with the small lead and this can lead to more

hidden features in SABR images. 50 keV photons were emitted by the enhanced source for both

MCNP runs.

Effect of a Gap Between the CR Plate and the Object

Another interesting and surprising factor affecting the image quality is the distance

between the CR plate and the objects to be imaged. Indeed, even a small gap of a few millimeters

can hurt the image quality a lot. This is shown in Figure 3-14, where the first SABR image

represents metallic objects placed on a nylon substrate without any gap between the CR plate and

the objects (the CR plate was placed directly on the objects), at 75 kVp, 60 mAs, and 25 inches.

The second SABR image shows the same objects but with a 4 mm gap between the CR plate and

the objects, the X-ray generator setup being the same (75 kVp, 60mAs and 25 inches).

It is obvious in these images that introducing even a small gap between the CR plate and

the target objects dramatically reduces the image quality. Even by modifying the X-ray generator

setup it was not possible to increase the contrast for the SABR image with the 4 mm gap. This

implies that the SABR technique, using the configurational setups examined in this work, can

only be used to detect near-surface defects or Foreign Object Debris that are very close to a









reachable thin surface. (It was possible for instance to detect large metallic objects just behind

aluminum or carbon-carbon composite plates, as seen later in this report.)

The explanation for this phenomenon is shown in Figure 3-15. It is at first surprising that

for the SABR technique, a small gap between the CR plate and the target object can hurt the

image quality so dramatically whereas for a pencil-beam scanning backscatter imaging system,

such as the RSD (Radiography by Selective Detection) system, the image quality remains good

even with a large gap between the detectors and the target object. In fact, with the pencil beam

scanning technique, if multiple-scattered photons are ignored, then, when the beam is over the

target object, all the X-ray photons reaching the detectors were previously backscattered in the

object (Figure 3-15A). On the other hand, with the SABR technique the entire background is

illuminated directly or indirectly through the aperture holes. Consequently, as the gap between

the CR plate and the object is increased, the solid angle with which the area of the CR plate

directly over the object sees distant parts of the background is increased, and X-ray photons

backscattered in another part of the background reach that area of the CR plate (Figure 3-15B).

As a result, the contrast is rapidly nullified.

Once again this experimental observation was validated by a MCNP calculation for which

the geometry can be seen in Figure 3-16. Figure 3-17 shows the difference between the images

obtain with MCNP simulation of a SABR image of a steel washer on a nylon background with

the CR plate directly laid on the object, or with a 4 mm gap. The steel washer appears much

sharper, and with more contrast between the objects and the background when there is no gap

between the object and the CR plate.

Effect of a Gap Between the Object and the Substrate

The last noticeable effect is caused by the introduction of a gap between the object and the

substrate, with the CR plate directly on the object. Shadow Aperture Backscatter Radiography of









objects suspended with the help of threads 3.5 inches above the substrate gave images with an

extremely good contrast. This phenomenon is explained by the fact that the suspended object is

crossed by photons which have been backscattered from every single area of the substrate.

Therefore from the point of view of both the object and the CR plate just above it, the substrate

acts as a much more uniform secondary source. Moreover, the areas of the CR plate that are

close to the illumination apertures are not crossed by a larger number of particles than the areas

under the center of the tiles. This causes less area of the SABR image to be overexposed. In fact,

suspending objects a few centimeters over the substrate allows detecting objects that could not be

detected when the object was in contact with the substrate. The nylon washer could even be

detected when suspended over aluminum or lead background, despite the fact that those

materials are strong absorbers. Indeed, Figure 3-18 shows the SABR image of a nylon washer

suspended 3.5 inches over a lead background, at 70 kVp, 60 mAs and at 25 inches, and proves

that this object can be detected when suspended, whereas it could not be seen in Figure 3-6 for

instance. This feature of the SABR method could have some application, for instance to help

detect objects located inside of a plane, in contact with the external hull and with an aluminum

background a few centimeters behind them.

A MCNP simulation of a steel washer suspended 3.5 inches over a nylon substrate also

proves that the contrast of SABR images is improved when the object is suspended. The

geometry of this MCNP calculation can be seen in Figure 3-19 and the comparison between the

MCNP simulated SABR images of a steel washer laid or suspended 3.5 inches over the nylon

substrate is shown in Figure 3-20. The background for the suspended washer (Figure 3-20B) is

more uniform than in Figure 3 -20A, and therefore, the steel washer is more easily detectable.












I


~L.


IMI


Figure 3-1. Foreign Object Debris on a nylon substrate: A) photograph, B) SABR images at 75
kVp, for 60 mAs at 25 inches, C) SABR images at 75 kVp, for 240 mAs at 47 inches.


Figure 3-2. Geometry of the MCNP simulation of the Shadow Aperture Backscatter
Radiography of a steel washer on a nylon substrate.


H -

--- t- 4---


Th___"


Figure 3-3. Flux of photons received after a MCNP simulation by each point of the CR plate for
three photon energy: A) at 10 keV, B) at 50 keV and C) at 100 keV.


A


*..
111
.;*..


B-OOOBi







.0013M


Figure 3-4. Flux of photons received after a MCNP simulation by each 0.5 pixel of the CR plate:
A) with the regular source and B) with the modified, more efficient source.


U


Figure 3-5. Steel Washer on nylon background: A) SABR image obtained at 75 kVp, 60mAs, at
25 inches and B) MCNP simulated SABR image based on the flux of photons
detected by the CR plate.


__ __ I I __ _


t


"01
.00032
2.9-7


m a











-nw
p..


I

I


Figure 3-6. Foreign Object Debris on a nylon substrate: A) Photograph and B) SABR image
obtained at 75 kVp, 60mAs, at 25 inches.


01
.00032

2.9-7
B.9-9


_ I 1 I "_ _


0


coooas
H I
oooosa B


Figure 3-7. MCNP simulated SABR images of: A) a steel washer and B) a nylon washer, on a
nylon substrate.


*U


4I









A
Figure 3-8. Foreign Object Debris on an aluminum substrate:
image obtained at 75 kVp, 60mAs, at 25 inches.


I
I
I


** I
- I-
I'll

Eul


I
I
I


I
I
I
-Bii


A) Photograph and B) SABR


50 100 150 200 A
Large washer Distance (pixels) Small washer


Figure 3-9. Profile plot of the previous SABR image on which can be seen two features (the hole
in the large washer and a small washer).

















Figure 3-10. Foreign Object Debris on a lead substrate: A) Photograph and B) SABR image
obtained at 75 kVp, 60mAs, at 25 inches.


B-I
j ..j
,rB


p- C


Figure 3-11. MCNP simulated SABR images of a steel washer: A) on a nylon background, B) on
an aluminum background and C) on a lead background.


I 4


_ I [ _


4 6*


#-










i WB
)IM
bag


B -


U
a
U
H


W -


U'

3 ,

- C


Figure 3-12. Foreign Object Debris on a nylon substrate: A) Photograph, B) SABR image with
small lead tiles and C) with large lead tiles, obtained at 75 kVp, 60mAs, at 25 inches.


17?-I
S S 3
a.1


I I


.00032

2.9-7
a.-P


Figure 3-13. MCNP simulated SABR images of a steel washer on a nylon background: A) with 1
inch lead tiles and B) with 2 inches lead tiles.


*6*I


I I


1 r I 1 I i


________________*_________________*________________









I.,


1

I
U


i.E

'RI
-H


Figure 3-14. SABR images of Foreign Object Debris on a nylon background at 75 kVp, 60mAs,
at 25 inches: A) with no gap between the CR plate and the objects and B) with a 4
mm gap.


Figure 3-15. Possible paths of photons backscattered in the target object: A) for the RSD
technique and B) for the SABR technique with a gap between the CR plate and the
target obj ect.


I

mn


Irce



N_


Lead tiles
Target object
Background


'Backscattered
X-rays


Target object
Background











Figure 3-16. Geometry of the MCNP simulation of the Shadow Aperture Backscatter
Radiography of a steel washer on a nylon substrate, with a 4 mm gap between the
washer and the CR plate.


.01
.00032

2.9-7
a.9-9


oo069,
00099

00002
a-29 B


Figure 3-17. MCNP simulated SABR images of a steel washer on a nylon background: A)
without a gap and B) with a 4 mm gap between the washer and the CR plate.






UMI




EMI
NMI
Figure 3-18. Nylon washer suspended by a thread over a lead background: A) Photograph and B)
SABR image taken at 75 kVp, 60mAs, at 25 inches.















































Figure 3-19. Geometry of the MCNP simulation of the Shadow Aperture Backscatter

Radiography of a steel washer located 3.5 inches over a nylon substrate.


r V


01


.00032




2.9-7


8.9-9


o-.046 ,








4 1- B


Figure 3-20. MCNP simulated SABR images of a steel washer on a nylon background: A)

without a gap and B) with a 3.5 inches gap between the background and the washer.


I I


3%
-
u..


^^^^_^^^^^-^^^^^_^^^^^_^^^^^_^^^^^


----------------*-----------------*----------------









CHAPTER 4
SABR: VARIOUS OBJECTS THAT HAVE BEEN IMAGED

Foreign Object Debris behind a thin aluminum or carbon-carbon composite plate

One of the conditions for the SABR technique to be used to detect Foreign Object Debris

is the ability to take images though thin layers of common materials, such as aluminum or

carbon-carbon composite. The three SABR images of lead and steel pieces on a nylon substrate,

which can be seen in Figure 4-1, were all taken at 75 kVp, 60 mAs, and 25 inches. The SABR

image shown in Figure 4-1A was obtained with the CR plate directly on the objects, while the

SABR image in Figure 4-1B shows the same objects but with a 1 mm thick aluminum plate

placed between the objects and the CR plate. Even though most of the metallic objects can still

be detected, the image quality is greatly reduced when the image is taken through 1 mm of

aluminum. However, this loss in image quality is not necessarily due only to x-ray attenuation in

the aluminum, but also to the 1 mm gap between the CR plate and the metallic objects induced

by introducing the aluminum plate. It was stated earlier in Chapter 3 that a gap between the CR

plate and the objects dramatically reduces the contrast in the SABR image.

The SABR image of the same objects in Figure 4-1C was also taken through a 1 mm thick

aluminum plate, but with a 5 mm stand-off distance between the objects and the aluminum plate.

The CR plate was placed on the aluminum. In this image, it becomes very difficult to detect even

the largest metallic objects. This is again due to the effect of a gap between the CR plate and the

objects on the SABR image contrast. In order to understand the effect of introducing an

aluminum plate and a stand-off distance, all three images were taken with the same experimental

setup. As a result, the last two images of Figure 4-1 are both slightly underexposed. However,

even by increasing the exposure for these two images, the contrast remained very low, and no

other object could be detected.









Each of the three previous SABR images was simulated by MCNP calculations, with

results displayed in Figure 4-2. The various metallic objects were replaced by a single steel

washer. This object can be detected on the first two simulated SABR images (Figure 4-2A and

Figure 4-2B) but not in the one with the 5mm gap between the object and the aluminum plate

(Figure 4-2C), as was observed experimentally on the SABR images.

The same three SABR images were taken, but this time, with a 1.5 mm thick carbon-

carbon composite plate, consisting of carbon fibers reinforcing a carbon matrix, instead of the

aluminum plate to test for depth penetration capabilities for other materials. These SABR images

can be seen in Figure 4-3. The first two (Figure 4-3A and Figure 4-3B), which are images

without and with the carbon-carbon plate, were taken at 75 kVp, 60 mAs and at 25 inches. The

SABR image in Figure 4-3C was taken with the carbon-carbon composite plate and a 5 mm

stand-off at 75 kVp and at 25 inches, with an exposure of 90 mAs, to avoid underexposure.

Again, the image quality was greatly decreased when a carbon-carbon plate was introduced

between the objects to be imaged and the CR plate. And with a 5 mm stand-off between the

objects and the 1.5 mm carbon-carbon composite plate, even the largest objects were almost

invisible.

SABR Images of a Spray-on-Foam Insulation Block

The next object to be imaged was a block of foam (12 inches long, 12 inches wide, 2

inches thick), with 5x5 regularly spaced holes in it, all different in diameter (0.25", 0.38", 0.5",

0.65" and 0.75") or in depth (0.125", 0.25", 0.375", 0.5" and 0.65"). This block of foam, which

was placed on an aluminum plate, can be seen in Figure 4-4. Radiography by Selective Detection

is able to detect flaws in this kind spray-on-foam insulation that covers the Space Shuttle

external tank.2 As stated before, flaws inside the foam were responsible for parts of the external









tank's insulation striking the Colombia's heat shield in the 2003 accident. Detecting flaws in the

foam could help prevent such an accident from happening again.

At first, the 5 holes with the largest diameter (0.75 inches) were imaged with the SABR

technique. However, contrary to what can be obtained with the Radiography by Selective

Detection, the holes are invisible on the SABR image. This can be seen in Figure 4-5. This

SABR image was taken at 70 kVp, 60 mAs, and 25 inches with the large tiles (2 inches).

Different energy levels and exposures were tried, but the holes could never be detected in the

SABR images.

In order to detect those holes, a dime cupronickell, 18 mm in diameter) was placed in each

of them. This is shown in Figure 4-6. The SABR image was taken at 70 kVp, 60 mAs, and 25

inches.

Only the coins placed in the two shallowest holes were clearly visible. This can again be

explained by the dramatic loss of contrast in the SABR images when a gap between the objects

and the CR plate is introduced. Because a dime is about 1.35 mm thick, the true gap between the

CR plate and the last clearly visible dime measures only roughly 0.5 mm. A surface plot shows

that the third dime is also visible with some image processing as shown in Figure 4-7. However,

it is very unlikely that this third coin would have been detected if the operator did not know

where to look. Because of the very low density of this foam, it is likely that the image contrast

would be the same if the coins were introduced inside of the foam at the equivalent depths

instead of just being placed in holes.

SABR Images of a Corroded Piece of Aluminum

A small, extremely corroded plate of aluminum, shown in Figure 4-8A, was also imaged

with the SABR technique. The ability to see corrosion with an X-ray backscatter technique can

be used for instance during the maintenance of airplanes to check the general state of the inside









of the plane, without the need of tearing it apart. The SABR image on Figure 4-8B was taken at

70 kVp, 120 mAs, at 25 inches. The exposure was increased compared to most of the previous

scans because almost the entire nylon substrate was covered by the 1 mm thick corroded

aluminum plate which has a high absorption cross-section. Even by increasing the exposure, the

image quality remained very poor, and some of the holes could not even be detected. Also the

corrosion is invisible on this SABR image.

A similar SABR image was taken, with the aluminum plate suspended between two 5 cm

thick blocks of lead, about 3.5 inches above the nylon substrate, as shown in Figure 4-9A. The

corresponding SABR image (Figure 4-9B) was also taken at 70 kVp, 120 mAs and 25 inches. As

shown earlier, it is not surprising to observe that the image quality was greatly improved

compared to when the corroded part of aluminum was laid on the nylon substrate.

For this setup, lead and aluminum can also be used for the background, with no noticeable

loss in relative contrast. This was surprising because much brighter images had been expected

for the nylon background than for the lead background. Also, despite the relatively good quality

of this SABR image, the corrosion itself could not be detected, mainly because the contrast could

not be improved enough for it to be visible.

Another SABR image of this corroded aluminum part was taken at 75 kVp, 120 mAs and

at 25 inches, through a 1 mm thick aluminum plate. This image, along with the experimental

setup can be seen in Figure 4-10. As mentioned earlier, the introduction of an aluminum plate

decreased the image quality by a large factor.

However, some of the holes in the corroded aluminum plate are still slightly visible with

the help of some image enhancement, as can be seen in Figure 4-11. The corrosion can obviously

not be detected through the aluminum plate.









Other Objects that have been Imaged with the SABR Technique

SABR images have been taken of many different objects, including metallic objects inside

a block of foam, bones covered by tissue simulant material, objects made of plastic (a

screwdriver for example) and other highly scattering materials. None of them showed a

reasonable image of the objects, mainly because of the inability of the SABR technique to detect

materials in which the mean free path of X-ray photons is too long, and because this technique,

using the configurational setups examined in this work, can only image flat, near surface objects,

due to the dramatic loss in image quality when the CR plate is not directly in contact with the

objects.






I-m
I,,
oI.
Ia,


I
U
U
a


I
I
I
BI


,I'
U,'


I


I


IH


CI,


Figure 4-1. SABR of Foreign Object Debris on a nylon background taken at 75 kVp, 60 mAs, 25
inches: A) without an aluminum plate, B) with a 1 mm thick aluminum plate between
the CR plate and the objects and C) with the 1 mm thick aluminum plate with a 5 mm
stand-off distance.


0 B


Figure 4-2. MCNP simulated SABR images of a steel washer on a nylon background: A) without
an aluminum plate, B) with a 1 mm thick aluminum plate between the CR plate and
the washer and C) with the 1 mm thick aluminum plate with a 5 mm stand-off


U


I
I
I
I


I. I.


4. 4.


.11


4 4


. C





I -


i
I
-B


low
I,.
I.. 5


r
II
I.
C hai


lI
U'
- -


Figure 4-3. SABR of Foreign Object Debris on a nylon background taken at 75 kVp, 60 mAs, 25
inches: A) without a carbon-carbon plate, B) with a 1.5 mm thick carbon-carbon plate
between the CR plate and the objects and C) with the 1.5 mm thick carbon-carbon
plate with a 5 mm stand-off (taken at 90 mAs).


Figure 4-4. Block of spray-on-foam insulation.















Figure 4-5. Five holes, 0.75 inch in diameter and of different depth in a spray-on-foam block: A)
photograph and B) SABR image at 70 kVp, 60 mAs, and 25 inches.


I


Figure 4-6. Five dimes placed in holes drilled in foam, 0.75 inch in diameter and of different
depth: A) photograph and B) SABR image at 70 kVp, 60 mAs, and 25 inches.









-N_


Figure 4-7. Zoom of the area of the previous SABR image that contained the third coin.


1,1

I.'

A B
Figure 4-8. Corroded aluminum plate on a nylon background: A) photograph and B) SABR
image at 70 kVp, 120 mAs, and 25 inches.











A =- B
Figure 4-9. Corroded aluminum plate suspended 3.5 inches over a nylon background: A)
photograph and B) SABR image at 70 kVp, 120 mAs, and 25 inches.


UI

I.I

II
A S B
Figure 4-10. Corroded aluminum plate covered by a 1mm thick aluminum plate and suspended
3.5 inches over a nylon background: A) photograph and B) SABR image at 70 kVp,
120 mAs, and 25 inches.




























Figure 4-11. Image enhancement of the previous SABR image with features surrounded in red.









CHAPTER 5
SABR: CONCLUSIONS

The Shadow Aperture Backscatter Radiography technique is an X-ray backscatter imaging

technique that is based on a snapshot acquisition method instead of the scanning acquisition

method which is used for X-ray backscatter imaging. It uses assemblies of lead tiles to block the

transmission signal. This prevents overexposure of the CR plate as the X-ray photons cross it to

reach the target objects. By limiting the superimposition of irradiation photons on the backscatter

signal, the signal-to-noise ratio is dramatically improved compared to the SBR technique. The

snapshot mode of acquisition allows the operator to obtain images much faster than with systems

that use scanning to acquire images (two images can be obtained in a couple of minutes with the

SABR technique while the acquisition time with the RSD system currently in use at UF would be

of the order of hours). But on the other hand, the image quality is not nearly as good, and fewer

objects can be detected in a SABR image. Also, because the areas just under the apertures,

between the tiles, are overexposed (lines about 1 mm wide), small objects can be completely

hidden on the image and therefore, at least two images of a given target need to be taken in order

to be sure to see all the areas in examined target.

The SABR technique offers fairly accurate X-ray backscatter images of near surface highly

absorbing objects (metallic objects for instance) that are placed on a substrate made of a good

scattering material such as nylon. Images acquired on more absorbing substrates, such as

aluminum, can be good too despite a loss in contrast. However, it was not possible to detect

objects made of highly scattering material. For example, a nylon washer on an aluminum

substrate remained invisible in the SABR images when a steel washer would have appeared, and

this was true no matter what X-ray energy level was used. Also, introducing a small gap between

the CR plate and the objects reduces the image contrast, and gaps as small as 5 mm can prevent









any object from being detected. However, a gap between the object and the background can

greatly increase the SABR image quality, to the point where objects that did not appear when

laid directly on the substrate were very clear in the image when suspended a few inches over the

background.

Some large metallic objects could also be seen through a plate of carbon-carbon composite

or aluminum of about 1 mm in thickness which was placed directly over them. However, in this

case the image quality was greatly decreased, and most of the smallest objects (less than 1 cm in

size) were invisible. When a stand-off was introduced between the plate and the objects, nothing

could be seen in the SABR image because of the gap between the CR plate and the objects. This

problem reduces the field of possible applications of the SABR technique because in most cases,

the objects that need to be detected are not directly in contact with an external surface. A lead or

tungsten collimation grid placed between the target and the CR plate could prevent the averaging

of contrast.









CHAPTER 6
CSD-SXI: INTRODUCTION

Computed Image Backscatter Radiography

Computed Image Backscatter Radiography (CIBR) is another X-ray backscatter imaging

technique developed by the SXI group. It employs a fan beam source and rotational motion

rather than a pencil beam source and rastering motion; it uses the same collimated scintillation

detectors as RSD. It was originally designed to acquire X-ray backscatter images faster than

RSD. This is because a larger fraction of the X-ray photons produced in the tube is used to obtain

the image, the count rate is higher than with RSD, and therefore a scan of the same object and

with the same statistical uncertainty tolerance is faster with the CIBR technique. Computed

Image Backscatter Radiography requires scanning across the target objects at different angles

with relatively small increments in order to obtain acceptable images, and it also needs image

reconstruction, much like Computed Radiography.8 However, further research is required to

improve resolution and reduce reconstruction artifacts through algorithm improvement. Figure 6-

1 shows a comparison between the RSD and CIBR images of letters of lead on nylon. Some

artifacts are visible on Figure 6-1B.

Backscatter Radiography using an Uncollimated Segmented Detector

The possibility of adding a linear pixelated detector, parallel to the fan beam source, to the

CIBR technique in order to acquire more information on the target object was considered. This

case, with geometry shown in Figure 6-2, was simulated using MCNP. In this simulation, a lead

strip (6.0 cm x 2.0 cm x 0.1 cm) is fitted in a block of nylon, which is a highly scattering material

contrary to lead which is a strong absorber. A 1 mm wide and 6 cm long fan beam normal to the

X-axis illuminates the strip of lead and the nylon around it. A linear Gadolinium Orthosilicate

(GSO) array detector (4.0 cm x 1.5 cm x 0.1 cm) is placed 3.0 cm above the surface of the









object, and starts 1.5 cm away from the fan beam. A flux mesh tally records the flux averaged

over the pixelated detector. The dots seen in Figure 6-2 represent the collision sites of the X-ray

photons in the matter, their colors varying continuously from red corresponding to the highest

energy particles prior to collision (in this case, 60.6 keV) to blue corresponding to the lowest

energy (6.3 keV). In order to accelerate the MCNP simulation, a DXTRAN sphere (not shown in

Figure 6-2) was included in the MCNP input in such a manner that it encircled the entire

segmented detector.9

The source of X-ray photons for this simulation was located 3 cm above the surface of the

target object because a non-negligible flux was detected by the upper part of the detector when

the source was higher due to scattering in air toward the detector. In practice, the source would

need to be placed at a 20 cm distance to obtain a wide fan beam but the source scattering in air

problem could be solved by ensuring that the top of the detector is shielded. The voltage used for

this simulation was 65 kVp, meaning that the X-ray spectrum ranged from 0 to 65 keV. The

energy distribution of the X-ray source at 65 kVp, generated by the program "XRSPEC" from

the SXI group is shown in Figure 6-3. The hump around 10 keV is due to the L series

characteristic X-rays for tungsten.10

The flux of X-ray photons measured by the mesh tally in the MCNP simulation was then

compared with that obtained for the exact same geometry (4.0 cm x 1.5 cm x 0.1 cm GSO array

detector, 6.0 cm x 2.0 cm x 0.1 cm lead strip on nylon) but with a pencil beam source. This

MCNP simulation geometry can be seen in Figure 6-4.

As for the previous case, one measure of the flux for every 1 mm pixel (in the y-direction)

was taken over the volume of the detector. Instead of having to do multiple simulations for each

value (forty 1.0 mm pixels), the source was defined as 40 circular sources, 1.0 mm in diameter









placed on a line and 40 different flux tallies were obtained with the SCD option using FU cards.9

This association of data cards allowed the tally corresponding to each region of the detector to

count only the particles coming from the corresponding source. As a result, there is no

interaction between the different parts of the image. Figure 6-5 shows the averaged, normalized

flux in the uncollimated linear array detector with a pencil beam and a fan beam X-ray source

with the associated relative error.

It is clear in Figure 6-5 that the averaged flux in the detectors in the case of the pencil

beam shows a strong contrast between when the pencil beam is above the nylon or the lead,

because of the fact that lead is a highly absorbing material (which results in a low count rate of

backscattered X-rays) and that nylon is a highly scattering material (which results in a high count

rate). However, in the case of the fan beam source with an uncollimated pixelated detector, the

flux shows very little difference between pixels located over the nylon and those over the lead.

This phenomenon can be simply explained by the fact that each pixel can detect photons

backscattered toward any direction and not exclusively those coming directly from below.

Consequently, each pixel of the detector sees photons backscattered in both the lead and the

nylon as can be seen in Figure 6-6, and as a result, the contrast is averaged out and greatly

reduced. The exact same effect was observed in the case of the Snapshot Aperture Backscatter

Radiography (SABR), a backscatter X-ray imaging technique based on a snapshot illumination,

and which offers very poor contrast when there is a stand-off between the target object and the

detector (a CR plate in the case of SABR experiments).

Very little information can be obtained from the flux measured for the fan beam case in

Figure 6-5 when compared to the flux for the pencil beam, and it does not seem that a fan beam









associated with an uncollimated pixelated detector could really be of any use in backscatter

radiography.


























Figure 6-1. Backscatter radiography images with a 1 mm resolution of letters of lead on nylon:
A) with the RSD technique with 1 mm pixels and B) with the CIBR technique with
10 degrees rotational increments and 1 mm radial increments.


2cm


2 cm


Linear OSO array detector


Fan Beam


Nylon Block


Linear OSO array detector





Air



Lead Strip



Fan Beam
N% Ion Block


Figure 6-2. Geometry of the MCNP simulation of the backscatter radiography of a lead strip on a
nylon background using an uncollimated segmented detector: A) view of the XZ-
plane and B) view of the YZ-plane.


Lead Strip














4.50E-02

4.00E-02

3.50E-02

3.00E-02

2.50E-02

2.00E-02

1.50E-02

1.00E-02

5.00E-03

0.OOE+00


Figure 6-3. Energy spectrum of the X-ray source at 65 kVp.


L cm


Linear OSO aray detector .


Pencil Beams

,..nfl..


Lead Strip

_ .- : -- .....- rr---- T -


\ Lead Strip


N% lon Block


Figure 6-4. Geometry of the MCNP simulation of the pencil beam backscatter radiography of a
lead strip on a nylon background: A) view of the YX-plane and B) view of the YZ-
plane.


Relative Intensity of X-rays

























0 10 20 30 40 50 60 71

Phowon Energy in keV


2 cm


Nylon Block


I"I
i'











Normalized Flux


-4-Fan beam with
0.80 uncollimated
detector

0 .60 -- ------
-a-Pencil beam with
uncollimated
0.40 detector


0.20


0.00
-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5
Distance in cm (y-axis)


Figure 6-5. Comparison between the normalized fluxes observed in the uncollimated array
detector with a fan beam source and with a pencil beam source.


Figure 6-6. Possible paths of backscattered X-ray photons toward the uncollimated segmented
detector with a fan beam source.









CHAPTER 7
CSD-SXI: PRINCIPLE AND MONTE CARLO SIMULATIONS

Principle of Collimated Segmented Detector Scatter X-ray Imaging

A more suitable way to use a segmented detector with a fan beam source would be to

somehow force the photons backscattered in the target object to be detected only by the

corresponding pixels (i.e. the pixels located directly over the location at which they

backscattered) through collimation.

This can be done by fitting a grid made of a highly absorbing material onto the pixelated

detector, in such a way that pixels or clusters of pixels would be separated from each other at

regular intervals by the grid and could only see photons coming from a limited solid angle. Such

a grid could be made of tungsten or lead, which are both strongly absorbing materials. Tungsten

would be preferable to lead because its macroscopic absorption cross-section is about 26 %

higher at 30 keV7, due to a higher electron density, and also because it is a much stronger

material and would more easily resist deformation especially if the collimation grid is very fine.

Figure 7-1 shows the geometry of a MNCP simulation of Collimated Segmented Detector

backscatter radiography of a lead strip on a nylon background. The collimation grid, shown in

light blue, is composed of 0.4 mm thick tungsten plates (2 cm x 1 cm) separated by a regular 1

mm spacing. A 0.4 mm thick tungsten shielding was placed all around the array detector and the

collimation plates to ensure uniform collimation. The resolution of the image is limited by the

spacing between collimation plates, but also depends on other factors, such as the distance

between the target object and the detector, the thickness of the tungsten plates and their

collimation length (1 cm in this particular simulation). In practice, the spacing between the plates

should be adjusted in such a manner that it would correspond to a natural number of pixel widths

which would then be grouped to form a cluster. Matching the grid with the pixel clusters could









be one of the technical difficulties associated with this imaging method. In this chapter, this

problem will be ignored and clusters of pixels of the segmented detector will be referred as

pixels of the image.

With such a collimation grid, each pixel is shielded from the X-ray photons that do not

come from directly under it, as seen in Figure 7-2. Therefore, it can be expected that this

configuration reduces the averaging of the contrast between the pixels, which is the reason why

an uncollimated array detector with a fan beam source could not work.

Figure 7-3 shows the comparison between the fluxes recorded by each pixel of the

segmented detector with a fan beam source and with a pencil beam source from an MCNP

simulation. In both cases, the tungsten collimation grid is placed below the detector to make the

fluxes comparable. It is clear in Figure 7-3 that the introduction of a collimation grid has

improved the contrast between the lead and nylon regions in the case of the fan beam source and

that the contrast in now very similar with a fan beam source and with a pencil beam source.

In fact, the contrast is even slightly higher with the fan beam source; the relative contrast,

defined as the ratio of the average fluxes over nylon and over lead is 36 with the fan beam

against 25 with the pencil beam source. A possible explanation for this is shown in Figure 7-4.

The scattering-to-absorption cross section ratio for 30 keV photons is 5% in lead and 400% in

nylon and their mean free path in nylon is about 3 cm against 0.06 mm in lead.7 Therefore it can

be expected that with a fan beam source, a large number of particles can travel inside the nylon

and follow the same path as particle 1 in Figure 7-4. This increases the flux detected by pixels

located over nylon with a fan beam source by 70% relative to when the pencil beam source is

used. On the other hand, pixels located over lead will see a flux dominated by short path and

single scatter photons since the scattering-to-absorption ratio is so low in lead. Consequently,









with a fan beam source, the flux detected by pixels located over nylon is increased by 70% while

the one over lead is only increased by 15% and this results in a higher contrast between the nylon

and the lead.

For a pencil beam however, this is not true because only the area directly under the pixel is

illuminated so the surrounding regions cannot contribute to the flux. This theory is confirmed by

the fact that the flux normalized to the number of source particles is 1.7 times higher with a fan

beam source than with a pencil beam source over the nylon region and only 1.2 times higher over

the lead region.

The average fraction of source particles in the detected fluxes was, in the case of the

uncollimated detector (Figure 6-5), 8.49 x 10-5 and 2.78 x 10-6 with a fan beam and a pencil

beam source, respectively, and, in the case of the collimated detector (Figure 7-3), 1.64 x 10-6

and 9.87 x 10-7 with a fan beam and a pencil beam source respectively. The uncertainties on

these values are on the order of 1% for the uncollimated detector, and 4% for the collimated

detector.

These figures confirm the theory described in Figure 7-2. For the fan beam source, the flux

recorded in the detector is almost two orders of magnitude smaller when the tungsten collimation

is introduced because then, the segmented detector mainly detects photons that are backscattered

up and toward the corresponding pixel while other photons are absorbed in the tungsten. The

decrease in recorded flux with a pencil beam source when the tungsten collimation is introduced

is much smaller, less than a factor of three. Even though the count rate recorded with the fan

beam source is less than twice as large as with the pencil beam source in the case of the

collimated detector, it should be noted that experimentally, a fan beam source yields many more









photons per unit time than a pencil beam source, and this is why an acceleration relative to the

pencil beam techniques is expected.

Resolution

Figure 7-3 also shows that the resolution with the fan beam source is lower than with the

pencil beam source. Whereas going from within 10% of the largest flux value (corresponding to

the nylon) to within 10% of the smallest (corresponding to the lead) takes only 1 mm in the case

of the pencil beam, the same transition takes about 3 mm for the fan beam method.

Contrary to pencil beam imaging techniques for which the image resolution is roughly the

step between each measurement, the calculation of the resolution with a fan beam source is

slightly more complex. Figure 7-5 shows the span of the target object that can be seen by a

collimated detector with a fan beam source, where P is the pixel size, H is the distance between

the target object and the detector, L is the collimation length and R is the resolution. From the

intercept theorem, the resolution, R is defined as:


R = P L (7-1)
2

In the case of the lead strip laid on the nylon block for which P = 0.6 mm, H = 30 mm and

L = 10 mm, the resolution R was 3.0 mm which is about 3 times larger than the resolution

obtained by the pencil beam technique.

This value of R = 3 mm for the fan beam versus R = 1 mm is confirmed by the results from

Figure 7-3, in which the transition from the lead to nylon flux value is done in about 3 mm with

the fan beam source against only 1 mm with the pencil beam source. However, the resolution

cannot be so simply defined; indeed, the area below the central part of the target object can be

seen from every part of the pixel whereas some other areas can only be seen from a fraction of

the pixel's surface, and therefore have less impact on the measurement of the flux. In fact, if the









target object is homogeneous, it can be assumed that each point y = t of a pixel receives equal

flux contribution from every point of the target object that it sees, as seen in Figure 7-6. The

contribution of this point of the pixel to the image is therefore:


c(ty) =- 1 if (t H tan 01) < y < (t + H tan 2 )(7-2)
10 else

where:

t P-t
tan 1 = and tan 62 = (7-3)
L L

Consequently, a pixel with width P would see the following contribution from each point y

of the target obj ect:

C(y)= J c(t,y)dt (7-4)

Figure 7-7 shows the normalized contribution from a homogeneous target object to a pixel

of the image of width P = 0.06 cm, with collimation length L = 1 cm and with an H = 3 cm

stand-off. It can be noticed that the resolution is indeed 3 mm, as shown previously. The area

directly under the pixel has a contribution of 1 because it can be seen from any part of the image

pixel. On the other hand, areas that are not located directly under the pixel can only be seen by a

fraction of its surface, which results in a smaller contribution to that pixel.

Influence of Collimation

Figure 7-8 shows the geometry of another MCNP simulation. In this case, the target object

is the same lead strip located 2 cm deep inside a nylon block. For this simulation, for which we

obviously expect the contrast to be smaller between the two types of material, the stand-off

distance, H, is 3 cm, and the pixel width and the collimator length are 1 mm and 1 cm,

respectively. For all the simulations in this section, the maximum energy of X-ray photons was

set to 120 keV, in order to have a better depth penetration in the nylon.









The comparison between the flux observed by the collimated and shielded detector with

the pencil beam and the fan beam sources is shown in Figure 7-9. In this case, the contrast is also

higher with the fan beam source, with a relative contrast of 1.61 between the nylon and the lead

against 1.34 with the pencil beam source. However, whereas the resolution remains close to 1

mm with the pencil beam technique, it seems that the resolution with the fan beam was greatly

reduced, to about 5 mm (transition from the value of the flux over the nylon to over the lead),

and indeed the formula for the resolution gives R = 5.4 mm at the depth of the lead strip. The

average fraction of particles counted in the fluxes relative to the source was 5.76 x 10-6 with

the fan beam source, and 3.18 x 10-6 with the pencil beam source.

Other MCNP simulations were done with different collimation lengths in order to improve

contrast as well as resolution, and to understand the effect of collimation on resolution. In

practice, it would be too expensive and unpractical to purchase sets of tungsten grids with

different collimation lengths. Instead, simply lowering the position of the collimated detector

should be almost equivalent to extending the length of the collimator. The other collimation

lengths that were tried are L = 1.5 cm, L=2 cm and L = 2.5 cm for which the theoretically

calculated fan beam image resolutions are, respectively, R = 3.4 mm, R = 2.4 mm and R = 1.8

mm.

Figure 7-10 shows the difference between the detected fluxes for the four collimation

lengths with a fan beam source. The resolution improves as the collimation is increased: R is

about 4 mm for the 1.5 cm collimation, 3 mm for the 2 cm collimation and 2 mm for the 2.5 cm

collimation. The best contrasts were obtained with the 2.0 and 2.5 cm collimation, but, in the

case of the latter, the count rate was reduced by a factor of two due to over-collimation. The

average fraction of source particles in the detected fluxes was 5.76 x 10-6 with 1.0 cm









collimation, 2.40 x 10-6 with 1.5 cm collimation, 9.95 X 10-7 with 2.0 cm collimation and

5.06 x 10-7 with 2.5 cm collimation.

Depth information and Possibility of 3D Imaging

2D Collimation Grid for Better Depth Resolution

Replacing a linear grid by a 2D grid, as seen in Figure 7-11, could improve the depth

selection. Such a grid could prevent photons that do not backscatter in the region of interest from

being detected. In such a case, it might be more efficient to tilt the detector toward the fan beam;

otherwise, the count rate for the regions that are far away from the source would not detect

anything. For the particular grid used in the following MCNP simulations, there were 6 rows of

40 pixels for a total of 240 grid elements, each of which was 2.5 mm long and 1 mm wide. The

thickness of tungsten plates was 0.4 mm.

Three MCNP simulations, with geometries shown in Figure 7-12, were performed with a

linearly collimated segmented detector with no inclination, with a linearly collimated segmented

detector at a 450 inclination, and with a 2D grid collimated segmented detector at a 450

inclination. In all cases 1.5 cm collimation was used. The target object consists of a 1 cm thick

and 2 cm wide air gap located 1.0 cm deep inside an aluminum block. In the case of the linear

detector with no inclination, the stand-off was 3 cm from the bottom of the detector (1.5 cm from

the bottom of the collimator). The collimation length was chosen to obtain the best possible

contrast for the linearly collimated detector array without inclination. The maximum energy of

the photons was 120 keV.

Figure 7-13 shows the flux recorded in each pixel of the segmented detector for all 3 cases

from the MCNP simulations. For the 2D collimated detector, each of the 6 rows had different

weights to prevent the overall flux from being dominated by the first row only; the count rate

detected by the first row is about 3 times larger than for the second row, and even more than for









the others. This is why the flux for each row was normalized and then were all averaged to give

only one value.

The graph from Figure 7-13 shows that the contrast in the detected flux between the air and

the aluminum is in the 5 to 10% range for both the linearly collimated detector without any

inclination and the 2D collimated array detector at a 450 inclination. It is lower than 5% in the

case of the linearly collimated detector at a 450 inclination due to poor collimation.

The areas of the array detector corresponding to the air should detect a lower flux because

the scattering cross section of photons in air is lower than in aluminum. However, it can be

noticed that the contrast is inverted in the case of the linearly collimated detector without any

inclination. This is because with these settings, for which the contrast is optimal, the array

detector is over-collimated so the flux that is recorded represents mostly the area located under

the air. Because the absorption cross section of photons in air is smaller than in aluminum, more

photons reach that depth when they went through the air first and therefore the flux recorded by

the detectors is higher for this region. This is one type of shadowing effect.3

However, the flux recorded by each row of the 2D collimated array detector contains much

more information than the flux averaged over all rows.

Possibility of 3D Backscatter Imaging

Each row of the 2D collimated array detector corresponds to a certain depth. A closer view

of the 2D collimated array detector is shown in Figure 7-14. At the intersection with the fan

beam plane, row 1 corresponds to a depth of 0.80 cm below the surface of the aluminum, row 2

to a depth of 1.14 cm, row 3 to 1.50 cm, row 4 to 1.85 cm, row 5 to 2.21 cm and finally, row 6

corresponds to a depth of 2.59 cm.

However, each row represents in fact a certain depth distribution which can be seen as

resolution on the z-axis. Therefore, the contribution of each z-level of the target object to every









row of pixels of the array detector can be obtained similarly to the contribution of each point of

the y-axis. However, because of the inclination Od of the detector, the expression for the

contribution is slightly more complicated. Also, an absorption term must be introduced to

account for depth penetration. Each point t of a pixel row (0 < t < P) can directly see the photons

scattered in the plane containing the fan beam (red arrow in Figure 7-15). Therefore:


c(tz) f 1 if (D tan(Od 81) H) < z < (D tan(Od + 02) H)
10 else

where:

t P-t
tan 1 = and tan 2 =- (7-6)
L L

and, with He and D, the height of the center of the pixel and its distance to the beam,

respectively:

D D + (t 2) sin(8d) and H= H +(t 2 cos(Od) (7-7)

However, if P << D, then H z He and D z D,. The contribution of each point of the z-axis

to the pixel row is then:

C(z) e-2z/mfp c(t,z)dt (7-8)

where mfp is the mean free path of photons in the medium. The exponential term allows taking

into account the absorption in the medium as the photons travel downward and then upward,

toward the detector after backscattering.

Figure 7-16 shows the contribution of every z-level of the target object to each pixel row of

the array detector. The value of the mean free path used in this case was 1.5 cm which is the

average mean free path of 120 keV photons in the aluminum surrounding the void.

From the graph in Figure 7-16, it appears that row 1 is under-collimated and only a small

fraction of photons detected by this row would have been backscattered in the region of the void,









located between 1.0 and 2.0 cm deep in the aluminum; the image would be dominated by the

flux of photons scattering in the region above the void. On the contrary, rows 5 and 6 are over-

collimated. Rows 2, 3 and 4 are all reasonably collimated even though they also see significant

flux contribution from the regions located over and below the void. This is a simplified

contribution model (it was assumed that photons remain on the fan beam plane as they travel

through the target object). Figure 7-17 shows the fluxes detected in the MCNP simulation by

each of the six rows of pixels of the 2D array detector with a 450 inclination in the case of the 1

cm thick air gap.

All rows show contrast of at least 10% between the void and the aluminum, with rows 2

(30%), 3 (40%), 5 (30%) and 6 (50%) showing the best contrasts. It should be recalled that the

best contrast that could be achieved with a linearly collimated detector with no inclination was

below 10%. It can be noticed that the contrast in the flux detected at rows 1, 2, 3 and 4 is

inverted compared to the flux at rows 5 and 6. This is because rows 5 and 6 are over-collimated

and are tallying fluxes of particles backscattered below the region of the void. Indeed the

absorption cross section of photons in aluminum is higher than in air, therefore X-rays reaching

that depth are more numerous if they traveled through air. This is the same shadowing effect as

was observed for the linearly collimated detector in Figure 7-13. The flux measured at row 4

shows a smaller contrast than at rows 3 and 5 because in this collimation range, the contrast goes

from negative (row 3) to positive (row 5), as the shadowing effect cancels out the difference in

scattering cross section in air and aluminum.

It should be noted that the fluxes detected in this MCNP simulation were, on average,

1.36 x 10-6 for row 1, 7.14 x 10-7 for row 2, 4.27 x 10-7 for row 3, 3.19 x 10-7 for row 4,

2.56 x 10- for row 5 and 2.02 x 10-7 for row 6. By comparison, the linearly collimated









detector with no inclination saw an average flux of 9.04 x 10 -7. Because the flux detected for

rows corresponding to the highest depths is low, the statistical error for these rows can be high.

Also, actual detection of a feature is always better than detection of its shadow. This is why it is

expected that rows 2 and 3 give the best results.

The fluxes measured by the different rows of a 2D collimated segmented detector put

together could form a 3D image of a feature. It is roughly equivalent to changing the collimation

of a linearly collimated detector for each depth. Although the statistical error is larger, the

contrast is much higher and it only requires one scan so it is faster. However, true 3D imaging

with a backscatter X-ray method is difficult since features can show on an image even though

they are below the collimation plane, due to shadowing effects. Therefore, some features could

be hidden by others located at a shallower point.

CSD-SXI 3D Imaging: Example of a Ring of Air Inside Aluminum

X-ray backscatter images of a ring of air located 1.0 cm below the surface of an aluminum

block were obtained using MCNP simulations. This ring, shown in Figure 7-18, has an inner

diameter of 0.5 cm, an outer diameter of 1.5 cm and is 1.0 cm high.

Images of this ring were simulated for three different setups: a pencil beam system with a

Nal detector 5.9 cm in diameter, a fan beam system with a linearly collimated GSO array

detector with no inclination and a fan beam system with a 2D collimated GSO array detector at a

450 inclination. The pencil beam system is similar to the RSD system developed by the SXI

group; the detector is located 9.5 cm away from the pencil beam and the collimation used varied

between 0.1 and 1.5 cm. The distance between the bottom of the detector and the surface of the

target is 3.0 cm. The geometries of all three systems are shown in Figure 7-19.

The average normalized flux detected per 1 mm pixel at the bottom surface of the detector

after MCNP simulations for the pencil beam systems with a 1 cm collimation was equal to









4.08 x 10-6 averaged over a total detector surface of 27.3 cm2. In the case of the linearly

collimated segmented detector, the average normalized flux is 4.02 x 10-6 averaged over a

pixel surface of 1.5 x 10-1 cm2. For the 2D collimated segmented detector system, the average

normalized flux was 4.0 x 10-6 for row 1, 2.1 x 10-6 for row 2, 1.2 x 10-6 for row 3, 8.9 x

10-7 for row 4, 7.2 x 10-7 for row 5 and 5.6 x 10-7 for row 6 average over a pixel surface of

2.5 x 10-2 cm2. This means that the detection rate of the Nal detector is about two hundred

times larger than for each linearly collimated pixel, and more than one thousand times larger than

for each 2D collimated pixel.

However, it should be remembered that for the pencil beam system, only one pixel value

can be measured at a time. So if a 40x40 1 mm pixel image is considered, the pencil beam

system will have to spend some time on each of the 40 pixels of the line, while the fan beam

systems will record one entire 40 pixel line of the image at a time. In this case, the pencil beam

system would be 4.5 times faster than the linearly collimated array detector system with the same

statistical error on average for the pixels of the image. The 2D collimated array detector system

would be 27 times slower, with the same statistical uncertainty for the Nal detector and row 1, or

almost 200 times slower, with the same statistical uncertainty for the Nal detector and row 6,

although six images would be acquired simultaneously.

However, it should be remembered that these values are given for the image acquisition of

a 4x4 cm target object. The use of a longer segmented detector would reduce these ratios and it

can be estimated that with a 1 meter long pixelated detector, CSD-SXI would become faster than

pencil beam systems. CSD-SXI would then be suitable for large area inspection, such as the

inspection of the hull of commercial airplanes. Some manufacturers, such as Envision Product

Design can build linearly segmented detector up to 2 meters long.11









Figure 7-20 shows the MCNP generated X-ray backscatter images (40x40 pixels, 1 mm

per pixel) of the air ring located 1 cm deep inside aluminum with a RSD-like pencil beam

system. The four images were obtained with different collimation lengths: 0. 1 cm, 0.5 cm, 1.0

cm and 1.5 cm. The average flux values for each of these images were 1.24 x 10-5, 8.62 x

10-6, 4.08 x 10-6 and 8.58 x 10-7, respectively, averaged over a surface of 27.3 cm2. The

corresponding statistical error was below 8% for all collimation levels (less than 1% at 0.1 cm).

Each of these images required 1600 MCNP runs, each of them taking a few seconds, one for

every pixel. It seems that the best images of the ring with this pencil-beam system are the first

two, which were generated with 0.1 and 0.5 cm collimation. However all four images are rather

blurry and have fairly poor contrast. Moreover, the exact dimensions of the ring cannot be

measured in any of these images.

Figure 7-21 shows the MCNP generated images of the air ring in aluminum with a fan

beam source and using a linearly collimated array detector without any inclination. These three

images were generated using different collimation lengths (0.5 cm, 1.0 cm and 1.5 cm) and the

average flux recorded by the MCNP simulations were 3.98 x 10-5, 1.35 x 10-5 and 4.02 x

10-6 for a pixel surface of 0.15 cm2. The statistical error for each pixel of these images was

below 7%. 40 MCNP runs (30 minutes each) were necessary to generate these images. The last

two images show the best contrast. It can be noticed that the ring appears lighter than the

aluminum as was previously observed in Figure 7-13. Also the ring is not perfectly centered in

the middle of these images, and this indicates that they show some kind of shadow of the ring

rather than the ring itself. Even though the contours of the ring of air seem clearer than in the

case of the pencil beam system, the inverted contrast and the shadowing effect make these

images more difficult to read.









Figure 7-22 shows the MCNP generated images of the air ring in aluminum with a fan

beam source using a 2D collimated array detector at a 450 inclination. Each of these six images

represents one of the rows of the collimated pixelated detector and corresponds to the following

depths in the aluminum: 0.80 cm, 1.14 cm, 1.50 cm, 1.85 cm, 2.21 cm, and 2.59 cm. It should be

recalled that the air ring is located between 1 and 2 cm below the surface of aluminum. This is

better shown in Figure 7-23. All six images were generated using 40 MCNP runs (30 minutes

each), one for each row and, in practice, they would be generated by only one scan across. The

average flux recorded by the MCNP simulations was 4.00 x 10-6 for row 1, 2.09 x 10-6 for

row 2, 1.24 x 10-6 for row 3, 8.94 x 10-7 for row 4, 7.19 x 10-7 for row 5 and 5.61 x 10-7

for row 6. The statistical error for each pixel was of the order of 8% for the first row and about

15% for the last row. The image corresponding to the first row is under-collimated and this is

why its contrast is so poor. The images generated by rows 4, 5 and 6 are all over-collimated (the

collimation plane is either at the bottom of or below the ring). As a result they show either poor

contrast or blurry ring contours. Also, for the last two rows, the ring can only be detected

because of the shadowing effect and the contrast is inverted. However for rows 2 and 3, the

contrast is very high (of the order of 50%) and the shape of the ring is perfectly clear. The

dimensions of the ring can also be measured with a fairly good precision. These two rows

represent depths of around 1.15 cm and 1.50 cm, which correspond to the location of the air ring.

The contrasts observed in these six images are consistent with what was observed in Figure 7-17.













2ancm


Figure 7-1. Geometry of the MCNP simulation of a Collimated Segmented Detector Scatter X-
ray Image of a lead strip on a nylon background: A) view of the YZ -plane and B)
view of the XY-plane.


A


Figure 7-2. Possible paths of backscattered X-ray photons from a fan beam source toward the
segmented detector: A) without collimation and B) with a collimation grid.





I


i
I
I .
i








Normalized Flux


1.20 -r


Lead Strip
(6 x 2 x 0.1 cm)


Nylon
K .AmJa


-2.5 -2 -1.5 -1 -0.5 0 0.5
Distance in cm (y-axis)


--- Fan beam with
collimated and
shielded detector

-u-Pencil beam with
collimated and
shielded detector


1 1.5


Figure 7-3. Comparison between the normalized fluxes observed in the collimated array detector
with a fan beam source and with a pencil beam source for a lead strip on the surface
of a nylon block.


Image pixel


Tungsten
collimation


Particle 1


Fan beam
illumination field
only


Fan beam and
Pencil beam
illumination field


Figure 7-4. Possible paths of X-ray photons eventually reaching an image pixel. The path of
particle 1 is only valid with a fan beam source because it starts outside the pencil
beam illumination field.


Nylon


;r




























Figure 7-5. Measure of the resolution for Collimated Segmented Detector Scatter X-ray Imaging.


Figure 7-6. Backscatter X-ray secondary source as seen from point y













1.20

1.00

0.80

0.60

0.40

0.20

0.00


Normalized
contribution


-Normalized
contribution to
the pixel


-0.25 -0.2 -0.15 -0.1 -0.05 6E-16 0.05 0.1 0.15 0.2 0.25
Distance on the y-axis in cm


Figure 7-7. Normalized contribution to a 0.6 mm wide pixel, with a 1 cm collimation.


4 2cm


S 2anc


A B


Figure 7-8. Geometry of the MCNP simulation of a Collimated Segmented Detector Scatter X-
ray Image of a lead strip 2 cm deep inside a nylon block: A) view of the XZ-plane
and B) view of the YZ-plane.


I













Normalized Flux


-0Fan beam with
collimated and
shielded detector


-- Pencil beam with
collimated and
shielded detector


-2.5 -2 -1.5 -1 -0.5 0 0.5 1


1.5 2 2.5


Distance in cm (y-axis)



Figure 7-9. Comparison between the normalized fluxes observed in the collimated array detector

with a fan beam source and with a pencil beam source for a lead strip 2 cm deep in a
nylon block.






Normalized Flux


T


-*-Fan beam with 1.0
cm collimation
---Fan beam with 1.5
cmcollimation

-*-Fan beam with 2.0
cmcollimation

Fan beam with 2.5
cmcollimation


-2.5 -2 -1.5 -1


-0.5 0 0.5


1.5 2 2.5


Distance in cm (y-axis)


Figure 7-10. Comparison between the normalized fluxes observed in the collimated array
detector with a fan beam source for four different collimation lengths for a lead strip
2 cm deep in a nylon block.


1.00 -



0.80 -



0.60



0.40



0.20


ao-


T
I
r
t
r












V111G B1IG B B V B 111 VI B LB H1


Figure 7-11. Geometry of the MCNP simulation of a Collimated Segmented Detector Scatter X-
ray Image of a 1 cm deep air gap inside an aluminum block with a 2D grid collimated
detector at a 450 inclination: A) view of the XZ-plane and B) view of a plane parallel
to the detector.


Figure 7-12. Geometries of the 3 MCNP simulations of an air gap inside an aluminum block,
views of the XZ-plane: A) linearly collimated segmented detector, B) linearly
collimated segmented detector at a 450 inclination, and C) 2D collimated segmented
detector at a 450 inclination.


7?


~\


~












Normalized Flux

1 cm thick air gap


-- Fan beam with
linearly
collimated
detector


- Fan beam with
linearly
collimated
detector at a 45'
inclination

- Fan beam with
collimated 2D
array detector
at a 45'
inclination


-2.5 -2 -1.5 -1 -0.5 0 0.5


1.5 2 2.5


Distance in cm (y-axKs


Figure 7-13. Comparison between the normalized fluxes from MCNP simulations of a 1 cm deep
air gap inside an aluminum block with a fan beam source.


Figure 7-14. Geometry of the 2D collimated array detector at a 450 inclination, view of the XZ-
plane.
















Fanbeam source


-tan(B ) I ...--- -------- -- -------- m ------

ehlE- -' V L H

6 H



Tarpt Obj-ect \^' .t

















Z









Figure 7-15. Backscatter X-ray secondary source as seen from point t.
If C
/l
C
/"
/I













Contribution


1 cm thick air gap


3.50 4.00 4.50 5.0


- Collim ted 2Darray
with 45' inclination,
sum of 311 rows
- Collimated 2Darray
with s45' inclination,
row 2
- Colli mated 20 D rray
with -45' inclination,
raw 2
- Collimated 2Darr=y
with a45' inclination,
row 3
- Collimated 2Darray
with a45' inclination,
row 4
- Co 11limted 2D array
witha45' inclination,
row 5
-- Co lli mated 2 Darray
with s45' inclination,


Depth in the targetbjectin cm


Figure 7-16. Normalized contribution of each z-level in the target object to all 6 rows of pixels

and to the sum of the rows.






Normalized Flux


-u-Collimsted 2D
array detector,
r3W 1
-a-Collimated 2D
array detector,
row 2
rD3 2
-Collimated 2D
array detector,
row 3
Collimated 2D
array detactar,
ro1w4
-+-Collimated 2D
array detector,
row 5
--Coallimated 2D
array detector,
row 6


-2.5 -2 -1.5 -1 -0.5 0 0.5
Distance in cm (y-axis)


1 1.5 2 2.5


Figure 7-17. Comparison between the fluxes recorded by rows 1, 2, 3, 4, 5 and 6 of the 2D
collimated array detector, with a 450 inclination from an MCNP simulation of CSD-

SXI of an air gap inside an aluminum block.


-IJ


0.00 0.50 1.00 1.50 2.00 2.50 3.00

























Figure 7-18. Geometry of the ring of air located 1 cm deep inside aluminum: A) view of the XZ-
plane and B) view of the XY-plane.


I1 [F


II


Figure 7-19. Geometries of the three different imaging systems: A) Nal detector with pencil
beam source, B) linearly collimated array detector with fan beam source and C) 2D
collimated array detector at a 450 inclination with fan beam source.


































Figure 7-20. MCNP generated X-ray backscatter images of a ring of air inside aluminum using a
Nal detector with a pencil beam source with the following collimation lengths: A) 0.1
cm, B) 0.5 cm, C) 1.0 cm and D) 1.5 cm.


Figure 7-21. MCNP generated X-ray backscatter images of a ring of air inside aluminum using a
linearly collimated segmented detector with a fan beam source with the following
collimation lengths: A) 0.5 cm, B) 1.0 cm and C) 1.5 cm.















A BO I









D E F
Figure 7-22. MCNP generated X-ray backscatter images of a ring of air inside aluminum using a
2D collimated segmented detector at a 450 inclination with a fan beam source. Each
image correspond to one of the six rows of the pixelated detector A) row 1, B) row 2 ,
C) row 3, D) row 4, E) row 5 and F) row 6.


a K RU2W^















U nde rcol limi united





Good collimation






Overcollimated :
shadowing effects


-Ocm


- 1 cm


-2cm


-3 cm


A-


Figure 7-23. Z-levels corresponding to each of the six MCNP generated images of the ring of air.









CHAPTER 8
CSD-SXI: CONSTRUCTION AND TEST OF A FIRST PROTOTYPE

Experimental Setup

Segmented Detector

The detector used for the prototype is a 12 inch long linearly segmented CMOS detector

built by Envision Product Design. It is shown in Figure 8-1. It is composed of three thousand,

eight hundred and ten 80 micron wide pixels." X-rays are allowed into the detector through a 2

mm wide and 1 cm deep collimated slot in the tungsten housing and strike the scintillator. The

visible light produced is then guided via fiber optics to a CMOS active-pixel sensor array (Figure

8-2). The slot in the tungsten provides depth collimation. The detector is mounted on the X-ray

tube stand, 4 cm away and 9 cm below the X-ray tube window. It is given a 40 degree angle

toward the X-ray source because the collimation slot is so tight that no photon would be detected

otherwise (Figure 8-3).

The segmented detector is connected to a workstation through a PCI card. Software

displays in real time the intensity recorded by each pixel. It is possible to group up to 16 pixels

together, which is very useful because the pixel bins must be of the same size as the collimation

grid pitch. The integration time can vary from 6 to 4000 ms, and while times as short as 20 ms

are commonly used for direct transmission imaging, integration times between 500 and 2000 ms

are needed for X-ray backscatter imaging purposes. A function allows recording a grey level

image for a given scanning speed. The RSD motion control software was used to move the X-ray

source and the detector for the scan. Before the acquisition of an image, calibration of the

detector must be performed each time the target background and the X-ray tube voltage are

significantly modified. This is done automatically by the software and takes between two and









five minutes. A successful calibration should result in a straight line across the whole detector as

shown in Figure 8-4.

Collimation Grid

The collimation grid for this CSD-SXI prototype is made of about eighty 0.4 mm (or

1/64") thick lead plates and twice as many spacers glued together, in an alternating pattern

resulting in a 7 cm long, 4 cm wide and 2 cm thick grid with a 0.88 mm pitch (Figure 8-5).

Although the grid built with this method is not exactly evenly spaced, it was much less expensive

than a specially machined tungsten grid would have been. Moreover, thanks to the calibration of

the detector, irregularities in the grid, such as slightly bent lead plates, have no effect on CSD-

SXI images. This collimation grid is mounted on the bottom part of the detector (Figure 8-6).

Obviously, the lead collimation grid is critical to the quality of images. Figure 8-7 shows the

difference between the image obtained with the segmented detector when it is collimated by the

grid and when it is not. Shapes and details are only visible when the detector is collimated, while

the uncollimated detector yields images with very limited contrast information and in only one

dimension.

It was previously stated that pixel bins must be of the same size as the grid pitch. In fact, a

first lead collimation grid was built with 1.06 mm (or 1/24") thick lead. The pitch of this grid,

which is shown in Figure 8-8, was 2.18 mm. However, because the software only allows

grouping up to 16 pixels, the maximum pixel bin size was 1.41 mm. This resulted in artifacts on

CSD-SXI images; because more than one group of pixels or "super-pixel" are collimated by the

same lead plates, they do not have the same perspective of the target object (Figure 8-9).

Consequently, the different groups of pixels do not detect the same flux of X-rays during the

transition between the background and a feature with a different scattering-to-absorption ratio,









and this leads to oscillation in image contrast as shown in Figure 8-10. MCNP simulations have

confirmed this behavior.

X-ray fan Beam Source

Another important point is to have a sufficiently narrow fan beam X-ray source. In all

previous MCNP simulations, it was assumed that the width of the beam was only 1 mm. In

practice, a machined lead part with a 1 mm slit aperture that was designed to provide a fan beam

source for CIBR (Figure 8-11A) was positioned at the X-ray tube exit window. However, the

actual width of the fan beam with this lead shield is about 1 cm as shown in Figure 8-11 B.

MCNP simulations have shown that if the fan beam width is larger than a few millimeters, the

image quality is greatly reduced. This is why another lead part was design to allow changing the

aperture (Figure 8-12A). This shielding provides a narrower fan beam illumination, but on the

other hand, the flux of X-ray photons received by the detector can be too low which can be a

problem for both calibration and acquisition time (Figure 8-12B). Both Figure 8-11B and Figure

8-12B were obtained using the segmented detector pointed toward the moving X-ray source.

Test of the First CSD-SXI Prototype

Contrast and Details

The various objects shown in Figure 8-13A, were imaged using CSD-SXI and RSD. Figure

8-13B shows the CSD-SXI image of these objects on an aluminum background, taken at 120

kVp and 8.75 mA with an acquisition time of 3 minutes and 42 seconds (1000 ms integration

time per pixel line). Figure 8-13C shows the RSD image of the same objects on the aluminum

background, obtained with the same X-ray tube setup in 16 minutes 21 second. The first

observation that can be made is that both images offer a similar level of detail, even though, most

features are clearer on the RSD image. The plastic screwdriver handle for instance, is easily

visible on the RSD image while it can barely be detected on the CSD-SXI image.









The ratio of contrast between the steel combination wrench and the aluminum background

is about 0.4 on both the RSD and CSD-SXI images. However, the contrast ratio between the

highly scattering plastic screwdriver handle and the background is 1.11 on the RSD image while

it is indeterminate on the CSD-SXI image. On other CSD-SXI images of the same objects this

plastic handle can be easily detected; it is a question of detector position and angle. The

statistical error for the background is close to 4% for the CSD-SXI image, and it is as low as

0.7% for the RSD image.

Figure 8-14 shows letters of lead on a nylon background which were also imaged using

CSD-SXI and RSD. Both images were obtained at 140 kVp and 4 mA, with acquisition time of 2

minutes and 46 seconds for CSD-SXI (1000 ms integration time per pixel line) and 8 minutes for

RSD. It is clearer in this example that the quality of RSD images is better. Not only is the

contrast better (lead-to-nylon ratio of 0.04 for RSD versus 0.12 for CSD-SXI), but the shape of

the target object is much more accurate in the RSD image, and the gaps between the lead tiles

composing the letters can even be detected although they are of the order of a few tenth of a

millimeter.

Resolution and Modulation Transfer Function (MTF) for CSD-SXI

An attempt was made to quantify the resolution associated with CSD-SXI images. A test

pattern composed of twelve 1 cm wide and 0.4 mm thick aligned lead tiles was scanned by CSD-

SXI and RSD (Figure 8-15A). Because the resolution for CSD-SXI is not necessarily the same in

both directions, parallel (Figure 8-15B) and perpendicular (Figure 8-15C) to the segmented

detector), two images were taken for this system against only one for RSD (Figure 8-15D).

Figure 8-16 shows the MTF of the two CSD-SXI images and of the RSD image. While

features as small as one millimeter can still be detected by RSD systems, this CSD-SXI

prototype yields images with a better resolution in the direction parallel to the linearly segmented









detector than in the perpendicular direction. This is due to the fact that resolution in the parallel

direction is controlled by the lead collimation grid with a 0.88 mm pitch whereas the resolution

in the perpendicular direction is a function of the 2 mm wide tungsten collimation slot. However,

increasing the collimation in the direction perpendicular to the segmented detector by decreasing

the width of the collimation slot, for instance would result in a lower count rate.

Depth Penetration

An important feature of X-ray backscatter imaging systems is the ability to detect features

located behind another object or deep below the surface of an object. Figure 8-17A shows

various objects such as a screwdriver, a wrench, steel and nylon washers, and screws on an

aluminum background. Figures 8-17B and 8-17C show the CSD-SXI and RSD images of these

objects without any cover at 140 kVp and 21.4 mA and at 120 kVp and 8.0 mA, respectively,

with a 1000 ms integration time per pixel line for CSD-SXI and 50 ms per pixel for RSD. The

contrast of both scattering (screwdriver's plastic handle, nylon washer) and absorbing (steel

wrench, screws) materials relative to aluminum is equivalent on both images. However, even if

all objects can be detected in both cases, the RSD image is clearer than the CSD-SXI; shapes are

more precisely represented and even the objects' shadows appear very clearly on the RSD image.

Figure 8-18 shows the RSD and CSD-SXI images of the same objects through 3.2 and 6.4 mm

thick aluminum plates. The objects appear slightly more distinctly and with more contrast on the

RSD images, but CSD-SXI images are also of fairly good quality. The plastic ruler on the other

hand can hardly be detected in both CSD-SXI images. Figure 8-19 also shows the RSD and

CSD-SXI images of the same objects on an aluminum background but this time through 2 mm

and 6.4 mm thick carbon-carbon composite. The RSD and CSD-SXI image quality are roughly

equivalent even though RSD offers images with much clearer contours than CSD-SXI. All the

RSD images shown in Figures 8-17, 8-18 and 8-19 were all acquired in 10 minutes and 47









seconds against only 3 minutes and 42 seconds for the CSD-SXI images. It should also be noted

that the lead collimation grid used for CSD-SXI was only 7 cm long. Increasing this length

would result in a longer active area for the segmented detector and therefore, CSD-SXI image

acquisition time would be reduced.































Figure 8-1. Photograph of the 12" Envision Product Design segmented detector.





Collimated Slot

Radiation Shield Lng
Tungsten Housing





CMOS Acrive Pixel
SerFnor array





Scintillator





Figure 8-2. Simplified diagram of the segmented detector used for CSD-SXI.
















-- X-ray tube


- Segmented detector


Collimation slot
-4


Figure 8-3. Segmented detector mounted on the X-ray tube at a 400 angle.


1>


Collimationleadgrid /


I -
\

i ^


With X-ray son

- With X-ravs off


Colliniaticnleadg -id


--------- ithX-rayson


With X-ra-.-r off


Figure 8-4. Flux detected by the linearly segmented detector with and without the X-rays on: A)
before calibration and B) after calibration.


:~; ~ A


/


>7
/"


w----.^-- .A* -' .



























Figure 8-5. Collimation grid made of 0.4 mm thick lead.


Figure 8-6. Lead collimation grid mounted on the bottom surface of the segmented detector.






















Figure 8-7. CSD-SXI image of letters of lead (1 mm thick) on nylon: A) with the lead
collimation grid and B) without the lead collimation grid.


I


Figure 8-8. Lead collimation grids made of 0.4 mm and 1.08 mm thick lead plates and spacers.


Iq
















Lead Ead
:oluniat 1211011' mo
plah te^ / '/ \late



I
I I








Figure 8-9. Explanation of the apparition of artifacts in CSD-SXI images when the collimation
grid pitch does not match the size of pixel bins.





A .. .. 1- B







A B





Figure 8-10. CSD-SXI images of a disk of lead on nylon: A) with a 2.18 mm collimation grid
pitch and 0.4 mm pixel bins and B) with a 0.88 mm collimation grid pitch and 0.88
mm pixels bins.













102
*", :- I^^^^ IN ^^^^^ l

































Figure 8-11. Lead shield with a 1 mm slit: A) photograph and B) Resulting X-ray fan beam at a
20 cm distance imaged with the segmented detector.



4a


Figure 8-12. Lead shield with variable aperture, in this case 0.2 mm: A) photograph and B)
Resulting X-ray fan beam at a 20 cm distance imaged with the segmented detector.







I
A(


ra
i


Figure 8-13. Various objects on an aluminum background: A) photograph, B) CSD-SXI image at
120 kVp and 8.75 mA and C) RSD image at 120 kVp and 8.75 mA.


Figure 8-14. Letters of lead on a nylon background: A) photograph, B) CSD-SXI image at 140
kVp and 4 mA and C) RSD image at 140 kVp and 4 mA.



















A






B





C





D

Figure 8-15. Lead test pattern on an aluminum background: A) photograph, B) CSD-SXI image
at 120 kVp and 8.80 mA with the pattern parallel to the segmented detector, C) CSD-
SXI image at 120 kVp and 8.80 mA with the pattern perpendicular to the segmented
detector, and D) RSD image at 120 kVp and 8.80 mA.











Relative contrast


--CSD-SXI parallel

--CSD-SXI perpendicular

RSD


Cycles percm


Figure 8-16. Modulation Transfer Functions (MTF) for the CSD-SXI and RSD images shown in
Figure 8-15.


Figure 8-17. Various objects on an aluminum background: A) photograph, B) CSD-SXI image at
140 kVp and 21.4 mA, and C) RSD image at 120 kVp and 8.0 mA.


























Figure 8-18. Various objects on an aluminum background: A) CSD-SXI image through a 3.2 mm
thick aluminum cover at 140 kVp and 25.0 mA B) RSD image through a 3.2 mm
thick aluminum cover at 120 kVp and 8.0 mA, C) CSD-SXI image through a 6.4 mm
thick aluminum cover at 160 kVp and 18.75 mA and D) RSD image through a 6.4
mm thick aluminum cover at 120 kVp and 8.0 mA.


A


Figure 8-19. Various objects on an aluminum background: A) CSD-SXI image through a 2 mm
thick carbon-carbon composite (C/C) cover at 140 kVp and 21.5 mA B) RSD image
through a 2 mm thick C/C cover at 120 kVp and 8.0 mA, C) CSD-SXI image through
a 6.4 mm thick C/C cover at 140 kVp and 21.5 mA and D) RSD image through a 6.4
mm thick C/C cover at 120 kVp and 8.0 mA.









CHAPTER 8
CSD-SXI: CONCLUSIONS AND FUTURE WORK

The association of collimated array detectors employing a lead or tungsten collimator grid

with a fan beam source offers several advantages for X-ray backscatter imaging. The image

acquisition time can be reduced relative to pencil beam systems such as RSD due to the use of

the fan beam source and a simpler scanning pattern. Experimental tests on the prototype system

have confirmed that CSD-SXI can yield images faster than RSD with an almost equivalent

quality. Increasing the length of the collimator grid, and therefore the segmented active length,

would also decrease image acquisition time.

However, the images obtained with RSD offer more distinct details of target objects than

those taken with CSD-SXI. Moreover, the resolution of images obtained with this first CSD-SXI

prototype in the direction perpendicular to the segmented detector is quite low compared to the

resolution obtained with RSD. This could be solved by decreasing the width of the segmented

detector collimation slot, but this would come at the cost of a smaller detection rate. More work

is needed to improve these factors, but given the relative simplicity of this first prototype,

significant progress can be expected.

MNCP simulations have shown that using a 2D collimated segmented detector could lead

to accurate 3D backscatter X-ray imaging. This could be experimentally done by fitting a laser-

cut tungsten collimation grid onto the bottom part of a 2D segmented detector. However, the fan

beam source needs to be improved by building a lead shield with a finer slit than what is

currently used because MCNP simulations have shown that too large a fan beam source

adversely impacts the quality of 3D CSD-SXI images.

Other, more immediate future work includes writing a LabView program to provide a more

adapted calibration for the segmented detector and to control the motion motors for easier image









acquisition. A small motor could also be placed on the segmented detector itself to control its

angle and therefore the corresponding image depth.









LIST OF REFERENCES


1. A. JACOBS, E. DUGAN, D. SHEDLOCK, University of Florida Research Foundation,
Inc, "Radiography by Selective Detection of Scatter Field Velocity Components", Patent
No.: US 7,224,772 B2 (May 29,2007).

2. D. SHEDLOCK, B. ADDICOTT, E. DUGAN, and A. JACOBS, "Optimization of a RSD
X-Ray Backscatter System for Detecting Defects in the Space Shuttle External Tank
Thermal Foam Insulation", University of Florida (2005).

3. D. SHEDLOCK, "X-Ray backscatter imaging for radiography by selective detection and
snapshot: evolution, development, and optimization", Ph.D. Dissertation, University of
Florida (2007).

4. R. EVANS, "The Atomic Nucleus", p 683, Krieger Publishing Company (June 1982)

5. "Technical data", Kodak INDUSTREX Digital Imaging Plates, KODAK Publication No.
TI-2632 (September 2006).
http://www.shawinspectionsystems.com/products/kodak/datasheets/EN ti2632.pdf

6. "Film Based Portable X-ray Systems", Source-Ray, Inc (2005).
http://sourceray.com/brochure/SR1 15%20-SR1 30%20Brochure%20web.pdf

7. M. BERGER, J. HUBBELL, S. SELTZER, J. CHANG, J. COURSE, R. SUKUMAR,
and D. ZUCKER, "XCOM: Photon Cross Sections Database", NIST Standard Reference
Database 8 (February 2009).
http://physics.nist.gov/PhysRefData/Xcom/Text/XCOM.html

8. C. MENG, "Computed Image Backscatter Radiography: proof of principle and initial
development", Master Thesis, University of Florida (2008)

9. X-5 Monte Carlo Team, "MCNP A General Monte Carlo N-Particle Transport Code,
Volume II: User's Guide", LA-CP-03 -0245, Los Alamos National Laboratory (April
2003)

10. G. KNOLL, "Radiation Detection and Measurement", p16, Third Edition, John Wiley &
Sons, Inc (2000)

11. "CMOS Segmented ArrayTM Specifications", Envision Product Design (April 2003)
http://cmosxray.com









BIOGRAPHICAL SKETCH

Olivier Bougeant was born in Brittany, in northwestern France. After obtaining his high

school diploma, he studied for two years in the "Classes Preparatoires aux Grandes Ecoles" in

Rennes. He was then admitted to the "Ecole Nationale Superieure de Physique de Grenoble"

(ENSPG), where he obtained a bachelor's degree in physics in 2007 and an Engineer's degree in

nuclear engineering in 2008. In 2007, he started a Master of Science in nuclear engineering at

the University of Florida where he joined the Scatter X-ray Imaging group.





PAGE 1

1 ALTERNATIVE TECHNIQUES OF BACKSCATTER RADIOGRAPHY: SNAPSHOT APERTURE BACKSCATTER RADIOGRAPHY AND COLLIMATED SEGMENTED DETECTOR SCATTER X RAY IMAGING By OLIVIER BOUGEANT A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2009

PAGE 2

2 2009 Olivier Bougeant

PAGE 3

3 To my family: Alain, Christiane, Carole and Nathalie, and to Verena

PAGE 4

4 ACKNOWLEDGMENTS I would like to thank Dr. Edward Dugan, my advisor, for his precious guidance and for sharing with me his knowledge and his wisdom during the past two years I thank Dr James Baciak for being on the committee. I would also li ke to thank Dan Shedlock who offered valuable help, Dan Ekdahl, Georgi Gorgiev, and my colleagues and friends: Nissia Sabri, Christopher Meng and Kara Beharry. I thank all the students, professors and staff members at the department of N uclear and Radiological E ngineering for so warmly welcoming me and for making me feel at home in Florida. Finally, I would like to thank my family and all my friends for their unconditional support. I especially want to thank Verena, my girlfriend, for her love and her patience. Financial acknowledgment: University of Florida, Department of Nuclear and Radiological Engineering Nucsafe

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS .................................................................................................................... 4 LIST OF FIGURES .............................................................................................................................. 7 ABSTRACT ........................................................................................................................................ 12 CHAPTER 1 INTRODUCTION ....................................................................................................................... 14 Radiograph y by Selective Detection .......................................................................................... 14 Limits to Radiography by Selective Detection.......................................................................... 15 2 SABR: INTRODUCTION .......................................................................................................... 19 Snapshot Bac kscatter Radiography ............................................................................................ 19 Principle of SABR ....................................................................................................................... 20 Experimental Setup for SABR ................................................................................................... 21 3 SABR: IMPORTANT FACTORS ON IMAGE QUALITY .................................................... 26 Effects of the X ray Generator Configuration ........................................................................... 26 Exposure ............................................................................................................................... 26 Voltage ................................................................................................................................. 27 Notes on MCNP Photon Source for SABR Simulations ................................................... 28 Effects of the Substrate ............................................................................................................... 30 Effects of the Size of the Lead Tiles .......................................................................................... 32 Effect of a Gap Between the CR Plate and the Object ............................................................. 33 Effect of a Gap Between the Object and the Substrate ............................................................. 34 4 SABR: VARIOUS OBJECTS THAT HAVE BEEN IMAGED ............................................. 45 SABR Images of a Spray -on -Foam Insulation Block ............................................................... 46 SABR Images of a Corroded Piece of Aluminum .................................................................... 47 Other Objects that have been Imaged with the SABR Technique ........................................... 49 5 SABR: CONCLUSIONS ............................................................................................................ 56 6 CSD SXI: INTRODUCTION .................................................................................................... 58 Computed Image Backscatter Radiography .............................................................................. 58 Backscatter Radiography using an Uncollimated Segmented Detector .................................. 58

PAGE 6

6 7 CSD SXI: PRINCIPLE AND MONTE CARLO SIMULATIONS ........................................ 65 Principle of Collimated Segmented Detector Scatter X ray Imaging ...................................... 65 Resolution .................................................................................................................................... 68 Influence of Collimation ............................................................................................................. 69 Depth information and Possibility of 3D Imaging .................................................................... 71 2D Collimation Grid for Better Depth Resolution ............................................................ 71 Possibility of 3D Backscatter Imaging ............................................................................... 72 CSD SXI 3D Imaging: Example of a Ring of Air Inside Aluminum .............................. 75 8 CSD SXI: CONSTRUCTION AND TEST OF A FIRST PROTOTYPE ............................... 92 Experimental Setup ..................................................................................................................... 92 Segmented Dete ctor ............................................................................................................. 92 Collimation Grid .................................................................................................................. 93 X ray fan Beam Source ....................................................................................................... 94 Test of the First CSD -SXI Prototype ......................................................................................... 94 Contrast and Details ............................................................................................................. 94 Resolution a nd Modulation Transfer Function (MTF) for CSD -SXI .............................. 95 Depth Penetration ................................................................................................................ 96 9 CSD SXI: CONCLUSIONS AND FUTURE WORK ........................................................... 108 LIST OF REFERENCES ................................................................................................................. 110 BIOGRAPHICAL SKETCH ........................................................................................................... 111

PAGE 7

7 LIST OF FIGURES Figure page 1 1 Principle of backscatter Radiography by Selective Detection ............................................ 17 1 2 NaI and YSO detectors mounted on a pencil beam RSD system ........................................ 17 1 3 Scanning pattern of pencil beam RSD systems .................................................................... 1 8 2 1 Letters of lead on a nylon substrate ....................................................................................... 23 2 2 Principle of Shadow Aperture Backscatter Radiography .................................................... 23 2 3 Differential scattering cross section per unit solid angle at 1 keV, 100 keV and 2 MeV ......................................................................................................................................... 24 2 4 SR 115 portable X ray generator .......................................................................................... 24 2 5 Typical energy spectrum of a medical X ray generator at 75 kVp with and without aluminum filter ....................................................................................................................... 25 2 6 Assemblies of 1", 1.5" and 2" lead tiles ................................................................................ 25 3 1 Foreign Object Debris on a nylon substrate ......................................................................... 36 3 2 Geometry of the MCNP simulation of the Shadow Aperture Backscatter Radiography of a steel washer on a nylon substrate. ........................................................... 36 3 3 Flux of photons received after a MCNP simulation by each point of the CR plate for three photon energy ................................................................................................................ 36 3 4 Flux of photons received after a MCNP simulation by each 0.5 pixel of the CR plate ..... 37 3 5 Steel Washer on nylon background ....................................................................................... 37 3 6 Foreign Object Debris on a nylon substrate ......................................................................... 38 3 7 MCNP simulated SABR images ........................................................................................... 38 3 8 Foreign Object Debris on an aluminum substrate ................................................................ 39 3 9 Profile plot of the previous SABR image on which can be seen two features (the hole in the large washer and a small washer). ...................................................................... 39 3 10 Foreign Object Debris on a lead substrate ............................................................................ 40 3 11 MCNP simulated SABR images of a steel washer .............................................................. 40

PAGE 8

8 3 12 Foreign Object Debris on a nylon substrate ......................................................................... 41 3 13 MCNP simulated SABR images of a steel washer on a nylon background ....................... 41 3 14 SABR images of Foreign Object Debris on a nylon background at 75 kVp, 60mAs, at 25 inches ............................................................................................................................. 42 3 15 Possible paths of photons backscattered in the target object ............................................... 42 3 16 Geometry of the MCNP simulation of the Shadow Aperture Backscatter Radiography of a steel washer on a nylon substrate, with a 4 mm gap between the washer and the CR plate ........................................................................................................ 43 3 17 MCNP simulated SABR images of a steel washer on a nylon background ....................... 43 3 18 Nyl on washer suspended by a thread over a lead background ............................................ 43 3 19 Geometry of the MCNP simulation of the Shadow Aperture Backscatter Radiography of a steel washer located 3.5 inches over a nylon substrate. ......................... 44 3 20 MCNP simulated SABR images of a steel washer on a n ylon background ....................... 44 4 1 SABR of Foreign Object Debris on a nylon background taken at 75 kVp, 60 mAs, 25 inches ...................................................................................................................................... 50 4 2 MCNP simulated SABR images of a steel washer on a nylon background ....................... 50 4 3 SABR of Foreign Object Debris on a nylon background taken at 75 kVp, 60 mAs, 25 inches ...................................................................................................................................... 51 4 4 Block of spray-on-foam insulation ........................................................................................ 51 4 5 Five holes, 0.75 inch in diameter and of different depth in a spray on-foam block .......... 52 4 6 Five dimes placed in holes drilled in foam, 0.75 inch in diameter an d of different depth ........................................................................................................................................ 52 4 7 Zoom of the area of the previous SABR image that contained the third coin. .................. 53 4 8 Corroded aluminum plate on a nylon background ............................................................... 53 4 9 Corroded aluminum plate suspended 3.5 inches over a nylon background ....................... 54 4 10 Corroded aluminum plate covered by a 1mm thick aluminum plate and suspended 3.5 inches over a nylon background...................................................................................... 54 4 11 Image enhancement of the previous SABR image with features surrounded in red ......... 55 6 1 Backscatter ra diography images with a 1 mm resolution of letters of lead on nylon ........ 62

PAGE 9

9 6 2 Geometry of the MCNP simulation of the backscatter radio graphy of a lead strip on a nylon background using an uncollimated segmented detector ............................................ 62 6 3 Energy spectrum of the X ray source at 65 kVp. ................................................................. 63 6 4 Geometry of the MCNP simulation of the pencil beam backscatter radiography of a lead strip on a nylon background .......................................................................................... 63 6 5 Comparison between the normalized fluxes observed in the uncollimated array detector with a fan beam source and with a pencil be am source. ....................................... 64 6 6 Possible paths of backscattered X ray photons toward the uncollimated segmented detector with a fan beam source. ........................................................................................... 64 7 1 Geometry of the MCNP simulation of a Collimated Segmented Detector Scatter X ray Image of a lead strip on a nylon background ................................................................. 79 7 2 Possible paths of backscattered X ray photons from a fan beam source toward the segmented detector ................................................................................................................. 79 7 3 Comparison between the normalized fluxes observed in the collimated array detector with a fan beam source and with a pencil beam source for a lead strip on the surface of a nylon block ...................................................................................................................... 80 7 4 Possible paths of X ray photons eventually reaching an image pixel. .............................. 80 7 5 Measure of the resolution for Collimated Segmented Detector Scatter X ray Imaging .... 81 7 6 Backscatter X ray secondary source as seen from point y = t ............................................. 81 7 7 Normalized contribution to a 0.6 mm wide pixel, with a 1 cm collimation ....................... 82 7 8 Geometry of the MCNP simulation of a Collimated Segmented Detector Scatter X ray Image of a lead strip 2 cm deep inside a nylon block .................................................... 82 7 9 Comparison between the normalized fluxes observed in the collimated array detector with a fan beam source and with a pencil beam source for a lead strip 2 cm deep in a nylon block ............................................................................................................................. 83 7 10 Comparison between the normalized fluxes observed in the collimated array detector with a fan beam source for four differen t collimation lengths for a lead strip 2 cm deep in a nylon block ............................................................................................................. 83 7 11 Geometry of the MCNP simulation of a Collimated Segmented Detector Scatter X ray Image of a 1 cm deep air gap inside an aluminum block with a 2D grid collimated detector at a 45 inclination ................................................................................ 84 7 12 Geometries of the 3 MCNP simulations of an air gap inside an aluminum block, views of the XZ plane ............................................................................................................ 84

PAGE 10

10 7 13 Comparison between the normalized fluxes from MCNP simulations of a 1 cm deep air gap inside an aluminum block with a fan beam source .................................................. 85 7 14 Geometry of the 2D collimated array detector at a 45 inclination, view of the XZ plane ........................................................................................................................................ 85 7 15 Backscatter X ray secondary source as seen from point t ................................................... 86 7 16 Normalized contribution of each z -level in the target object to all 6 rows of pixels and to the sum of the rows ..................................................................................................... 87 7 17 Comparison between the fluxes recorded by rows 1, 2, 3, 4, 5 and 6 of the 2D collimated array det ector, with a 45 inclination from an MCNP simulation of CSD SXI of an air gap inside an aluminum block ........................................................................ 87 7 18 Geometry of the ring of air located 1 cm deep inside aluminum ........................................ 88 7 19 Geometries of the three different imaging systems .............................................................. 88 7 20 MCNP generated X ray backscatter images of a ring of air inside aluminum using a NaI detector with a pencil beam source with the following collimation lengths ............... 89 7 21 MCNP generated X ray backscatter images of a ring of air inside aluminum using a linearly collimated segmented detector with a fan beam source with the following collimation lengths ................................................................................................................. 89 7 22 MCNP generated X ray backscatter images of a ring of air inside aluminum using a 2D collimated segmented detector at a 45 inclination with a fan beam source ................ 90 7 23 Z -levels correspond ing to each of the six MCNP generated images of the ring of air ...... 91 8 1 Photograph of the 12 Envision Product Design segmented detector ................................ 98 8 2 Simplified diagram of the segmented detector used for CSD SXI ..................................... 98 8 3 Segmented detector mounted on the X ray tube at a 40 angle .......................................... 99 8 4 Flux detected by the linearly segmented detector with and without the X -rays on ........... 99 8 5 Collimation grid made of 0.4 mm thick lead ...................................................................... 100 8 6 Lead collimation grid mounted on the bottom surface of the segmented detector .......... 100 8 7 CSD SXI image of letters of lead (1 mm thick) on nylon ................................................. 101 8 8 Lead collimation grids made of 0.4 mm and 1.08 mm thick lead plates and spacers ...... 101 8 9 Explanation of the apparition of artifacts in CSD SXI images when the collimation grid pitch does not match the size of pixel bins ................................................................. 102

PAGE 11

11 8 10 CSD SXI images of a disk of lead on nylon ...................................................................... 102 8 11 Lead shield with a 1 mm slit ................................................................................................ 103 8 12 Lead shield with variable aperture, in this case 0.2 mm .................................................... 103 8 13 Various objects on an aluminum background .................................................................... 104 8 14 Letters of lead on a nylon background ................................................................................ 104 8 15 Lead test pattern on an aluminum background .................................................................. 105 8 16 Modulation Transfer Functions (MTF) for the CSD -SXI and RSD images shown in Figure 8 15............................................................................................................................ 106 8 17 Various objects on an aluminum background .................................................................... 106 8 18 Various objects on an aluminum background .................................................................... 107 8 19 Various objects on an aluminum background .................................................................... 107

PAGE 12

12 Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science ALTERNATIVE TECHNIQUES OF BACKSCATTER RADIOGRAPHY: SNAPSHOT APERTURE BACKSCATT ER RADIOGRAPHY AND COLLIMATED SEGMENTED DETECTOR SCATTER X RAY IMAGING By Olivier Bougeant August 2009 Chair: Edward T. Dugan Major: Nuclear Engineering Sciences Unlike standard transmission radiography, Compton Backscatter Imaging (CBI) techniques are non -destructive examination methods that rely on the de tection of X ray photons back scattered in the target object. They have the advantage of being single -sided imag ing techniques and can yield better images than transmission radiography f or certain applications The X ray backscatter imaging system currently used at the University of Florida employs a method called Radiography by Selective Detection (RSD) It uses an X ray pencil beam to illuminate the target object while scintillation detectors positioned around the X ray source count backscattered photons. As the X -ray beam scans the target object, one real time 2D image is created per detector based on the recorded counts. Lead coll imation sleeves placed around the detector prevent particles scattered above a given depth from being detected, and help provide good depth information in RSD images This system has been commercialized and can be used for detection of land mines, security inspections and detection of defects or foreign object debris. One of the drawbacks of this technique, however, is t he image acquisition time, especially when detectors are highly collimated. The two X ray backscatter imaging techniques presented in this thesis were originally designed to yield images with a shorter acquisition time.

PAGE 13

13 Snapshot Aperture Backscatter Radiography (SABR) is a single -sided Compton Backscatter Imaging technique that is based on a snapshot acquisition method, contrary to most other X ray Backscatter Imaging systems which generate images by scanning. This c haracteristic of the SABR technique greatly reduc es image acquisition time. The detector used with this technique is a CR plate shielded from direct X ray radiations by a lattice o f lead tiles: X rays illuminate the target object through apertures between the lead tiles and are backscattered toward the CR plate. However, both Monte Carlo simulations and actual experiments have shown that this technique with the employed aperture ar rangements, yields images with poor depth information Collimated Segmented Detector based Scatter X ray Imaging (CSD -SXI) is a new backscatter radiography technique. Its principle relies on the use of a pixelated detector, collimated by a fine grid of a st rongly absorbing material T he X ray source comes in the form of a fan beam, parallel to the segmented detector. Monte Carlo simulations and the first practical experimental tests have shown very promising results in both image quality and depth information

PAGE 14

14 CHAPTER 1 INTRODUCTION The purpose of this work is to present two new X ray backscatter imaging techniques: Snapshot Aperture Backscatter Radiography (SABR) and Collimated Segmented Detector Scatter X ray Imaging (CSD -SXI), both originally designed to yield faster images than existing Compton Backscatter Imaging (CBI) systems currently used at the U niversity of Florida, while bringing additional information. Radiography by Selective Detection Backscatter Radiography by Selective Detection (RSD) is a single-sided imaging technique developed by the Scatter X Ray Imaging (SXI) group at the University o f Florida Its principle is shown in Figure 1 1. X -rays are emitted in the form of a pencil beam from an X ray tube toward the object to be imaged. Sodium Iodide (NaI) and Yttrium O rthosilicate (YSO) s cintillation detectors (Figure 1 2) placed around the X ray tube then record the number of photons backscattered toward them for a particular pencil beam position. As the system scans across the target object, a real time image is formed based on the counts recorded by each detector. L ead collimation sleeves, positioned around the scintillation detector s prevent the detection of photons backscattered above a certain depth called the collimation plane Thanks to this feature, RSD can yield high quality images of features located at sele cted depths inside the target object.1 Applications of backscatter Radiography by Selective Detection include, among other things, detection of land mines and Homeland Security inspections It is however currently used for detection of flaws and defects, such as cracks, voids and corrosion in a wide variety of materials including aluminum, steel, concrete, carbon -carbon composite s and Spray On Foam -

PAGE 15

15 Insulation (SOFI). In particular, since 2004, six RSD systems have been used by NASA and the Lockheed Martin Space Systems Co. to detect flaws in the foam insulation on the external tank of the s pace s huttle prior to each launch .2 Such flaws caused parts of this foam to strike the wing of the Columbia S pace S huttle in 2003 shortly after lift-off damaging the s h uttle's heat shield. RSD is also used for the detection of Foreign Object Debris (FOD). Limits to Radiography by Selective Detection Because RSD is b ased on a pencil beam X -ray source, it allows obtaining very accurate backscatter images, with sub -millimeter resolution In order to detect small differences in contrast between features of a target object, the RSD technique needs to achieve a high count rate to reduce the statistical uncertainty associated with the measure ment However, because this t echnique requires a narrow pencil beam source, it is necessary to shield the majority of the photons as they exit the X ray tube to reduce the dispersion in the pencil beam. As only a fraction of the photons produced by the X ray tube is used, RSD systems must spend sufficient time on every pixel of the image to ensure that the statistical uncertainties are limited Total scanning time is obviously an important factor for an imaging technique, and much of the research work accomplished by the SXI group is a imed at reducing the image acquisition time. Nucsafe a company based in Oak Ridge, Tennessee, which is working with the University of Florida, has mobile pencil beam systems capable of imaging over a square meter per minute and portable systems capable of imaging several square meters per minute. However, these fast systems do not use collimated detectors and as a result, there is only limited depth information. The scanning pattern for RSD is shown in Figure 1 3. The X ray source sweeps across the target object to record the counts for one line of pixels, before going to the next line and so on until the whole area has been imaged. To scan a square foot area with a 1 mm resolution, assuming that the required illumination time per pixel is 0.1 second, the total image acquisition

PAGE 16

16 time with the RSD system currently in place at the University of Florida, and constructed in 2004, would be two hours and thirty five minutes. The two backscatter radiography systems presented in this thesis, Snapshot Aperture Backs catter Radiography (SABR) and Collimated Segmented Detector Scatter X ray Imaging (CSD -SXI) were designed in an effort to reduce the acquisition time

PAGE 17

17 Figure 1 1. Principle of backscatter Radiography by Selective Detection Figure 1 2 NaI and YSO detectors mounted on a pencil beam RSD system

PAGE 18

18 Figure 1 3. Scanning pattern of pencil beam RSD systems.

PAGE 19

19 CHAPTER 2 SABR: INTRODUCTION Snapshot Backscatter Radiography The SABR technique is a method of X ray backscatter imaging originally based on the Snapshot Backscatter Radiography (SBR), a technique developed by the Scatter X Ray Imaging group at the University of Florida.3 The SBR technique obtain s backscatter, single -sided images of an object without scanning, thus reducing the image acquisition time. SBR images are obtained by placing a Computed Radiography (CR) plate directly on the top the target object, which could, for example, be composed of Foreign Object Debris (FOD) place d on a background, and then exposing this target to an X -ray snapshot through the CR plate. As a result the CR plate is exposed a fi rst time by the X ray photons which are then scattered in the background and can be absorbed in the FOD placed under the CR plate. Dependi ng on the scatteringto absorption ratio of the background and the FOD object s a given fraction of these photons reach the Computed Radiography plate. The background or substrate behaves like a secondary source of X -rays and the Foreign Object Debris shadow s the CR plate from th is source. This is why the best images were obtained with a highly scattering substrate and strongly absorbing objects. A CR plate is a fi lm like plate that is made of photo -stimulable storage phosphors. As the X rays emitted by the X -ray generator strike the phosphor's atoms ; the electrons are excited to a higher energy level. Then a CR plate reader (in th is case, a K odak INDUSTREX ACR 2000 Digital System ) scans the plate with a laser that causes the electrons to go back to their ground state. In the process, they release visible light photons that the reader collects and counts in order to compute a digital image.

PAGE 20

20 However, the SBR technique often results in an overexposure of the CR plate and a very low signal to -noise rati o because the backscatter signal is superimposed onto the transmission signal on the image. Figure 2 1 shows the photograph of a target object and the corresponding SBR image. In this case letters of lead were placed on a nylon background to obtain the hi ghest possible contrast as the nylon is a highly scattering material and the lead is a strong absorber. This image was taken at 50 kVp (peak kilovoltage) with a 2.85 mAs exposure with the X ray source placed 23 inches away from the target object Principl e of SABR Because of the low signal -to -noise ratio of SBR based images, the SXI group developed a new technique, called Shadow Aperture Backscatter Radiography, which is based on the SBR method but in which the backscatter signal is not superimposed onto t he transmission signal.3 The principle of the SABR technique is shown in Figure 2 2. As for the SBR method, a CR plate is placed on the object that is to be imaged. However, in the case of the SABR technique, the CR plate is covered by tiles of lead to prevent it from being completely saturated by illumination photons. Instead, the illumination X rays can only reach the CR plate and the target object through apertures between the lead tiles. These X ray photons are then absorbed in the object or backscat tered toward the parts of the CR plate that are shadowed by the lead tiles. As a result the contrast in the shadowed parts of the CR plate is greatly enhanced when compared to the images obtained with the SBR technique, but while the exposure of most part s of the image is good, there are some saturation ( white ) lines on the image corresponding to the illumination apertures between the lead tiles. Consequently, these parts of the images cannot be used to detect features directly under the CR plate. This pro blem can be overcome by simply acquiring one image of an object, and then by shifting the lead tiles, and by reacquiring another image to make sure that all the missing parts from the fi rst image can be seen on the second one.

PAGE 21

21 However, because the intensi ty received by each point of the CR plate decreases exponentially with its distance to the apertures, image reconstruction is complex and two images cannot be simply overlapped to get rid of the white lines. The energy range used in this experiment was ro ughly from 0 to 100 keV, which corresponds to energies at which the Compton scattering is still relatively isotropic, as shown in Figure 2 3, whereas higher energy photons experience strongly forward peaked scattering.4 This range of energy allows a large proportion of X ray photons to be backscattered toward the CR plate, while still allowing them to travel through a moderately absorbing medium, such as aluminum for instance. Experimental Setup for SABR The X ray source used for SABR was a Source Ray SR 115 portable x-ray generator (Figure 2 4) which allowed maximum photon spectrum energies from 40 to 100 keV. The designation used for such spectra is e.g., 40 kVp or 100 kVp (meaning 40 or 100 keV peak). The energy spectrum of such a n X ray source usually resembles a Maxwell Boltzmann distribution with average photon energy of about 40% the maximum energy. Figure 2 5 displays a typical medical X ray source energy spectrum at 75 kVp with a 2.7 mm aluminum filter, with average photon energy of 39 keV This gr aph does not represent real measurements done on the Source Ray X ray generator used for the SABR experiments but is an accurate energy spectrum for a standard X -ray source The single shot exposure on this X ray generator can vary between 0.15 mAs and 60 mAs, but it was possible to take several shots to obtain even higher exposures (the cooling time between two 60 mAs shots is about two minutes). The CR plate used was a Kodak GP Digital Imaging Plate SO 170, which is about 0.6 mm thick and which is mainly composed of a layer of barium fl uorobromoiodide (BaFBr) protected

PAGE 22

22 by a thin polyester coat.5 In the MCNP calculations detailed later in this report, the CR plate was assumed to be only made of BaFBr. The density of th is type of CR plate is roughly 5.0 g / cm3. Although a CR plate can theoretically be scanned thousands of times if handled with extra caution, the plates used were not in perfect condition and this resulted in artifacts in some of the images. To obtain a SABR image, the target objects, which are generally composed of various objects on a scattering substrate, is placed on a lead sheet that is laid on a steel table. Then, the CR plate is put directly on the objects that are to be imaged. Finally, an assembly of tiles of lead is placed over the CR plate. Three di fferent assemblies of tiles were used to shade the CR plate and create the apertures to allow illumination of the objects. Each assembly consisted of 1 mm thick square tiles of lead glued next to each other on a sheet of paper with an average spacing of 1 mm between the tiles for the apertures. The square tiles of the three assemblies were about 1, 1.5 and 2 inches long (Figure 2 6).

PAGE 23

23 A B Figure 2 1. Letters of lead on a nylon substrate : A) Photograph B) SBR image at 50 kVp, 2.85 mAs and 23 inches between the X -ray source and target Figure 2 2 Principle of S hadow Aperture Backscatter Radiography

PAGE 24

24 Figure 2 3. Differential scattering cross section per unit solid angle at 1 keV, 100 keV and 2 MeV .4 Figure 2 4 SR 115 portable X ray generator .6

PAGE 25

25 Figure 2 5 Typical energy spectrum of a medical X -ray generator at 75 kVp with and without aluminum filter. Figure 2 6 Assemblies of 1", 1.5" and 2" lead tiles

PAGE 26

26 CHAPTER 3 SABR: IMPORTANT FACTORS ON IMAGE QUALITY Effects of the X -ray Generator Con fi guration There are several factors that can a ff ect the quality of images obtained with the SABR technique. Perhaps the most important ones are the maximum energy of the X ray photons and the exposure. These factors can be modi fi ed directly on the SR 115 X ray generator. Exposure The exposure, measured in mAs (on the SR 115, the current is fi xed at 15 mA, and the exposure time varies between 0.01 and 4.0 seconds6), is very important fo r obtaining a reasonable amount of photons to avoid under or overexposure of the CR plate. This factor is also strongly linked to the distance between the X -ray source and the CR plate. Indeed, because X ray photons are roughly emitted isotropically at the anode in the X ray tube for energies below 100 kVp, the number of photons reaching the CR plate is proportional to 1/R2, where R is the distance between the source and the CR plate. For example, by increasing the distance between the X ray source and the CR plate by a factor of two, the exposure needs to be increased by a factor of four to obtain roughly the same image. A ffi rmation of this behavior is demonstrated in Figure 3 1 A number of metallic objects ( a lead wedge, some steel washers and a penny sho wn in Figure 3 1A ) were placed on a nylon substrate which is a very good scattering material. A fi rst SABR image (Figure 3 1B) was obtained at 75 kVp with an exposure of 60 mAs and with the CR plate located 25 inches away from the X ray source. The exposur e of this image is, given the very high sensitivity of CR plates, very close to the exposure of the second image (Figure 3 1C) obtained with the SABR method at 240 mAs and with a 47 inch distance between the source and the CR plate (47 inches is the maximu m possible distance between the source and the top of the table that was used), still

PAGE 27

27 at 75 kVp. The exposure and voltage were chosen to obtain the best possible images. The vertical and horizontal white lines that can be seen on the two SABR images corres pond to the illumination apertures. Voltage The effects of voltage variations are even more important. For instance, at 50 kVp, with a source to target distance of 25 inches, the objects on a nylon surface are only visible for exposures above 240 mAs, and with an aluminum substrate, only the white lines are visible at exposures as high as 360 mAs; the rest of the CR plate remains completely dark. This can be explained by the fact that at lower energies (below 20 keV), the dominant collision type of photons in the nylon and the aluminum is photoelectric absorption, and as a result a lower fraction of X rays can be backscattered in the substrate through the various objects and toward the CR plate. On the other hand, at energies greater than 80 kVp, for tiles o f lead 1 mm thick, there is too much transmission through the 1 mm thick lead shadow shields and the signal to -noise ratio is decreased because the backscatter signal is superimposed onto the transmission signal. This is because, at energies higher than 80 keV, a significant fraction of photons have a mean free path in lead on the order of 1 mm. For instance, for a photon of 30 keV, which is roughly the average energy of X ray photons for a source voltage of 70 kVp, the mean free path in lead is about 0.03 mm7; so, at this energy, very few photons actually go through the lead. However, at 90 kVp, a non -negligible fraction of photons have energies higher than 80 keV for which the mean free path in lead is about 0.3 mm, which is of the same order as the thickn ess of the lead tiles. MCNP calculations were performed to confi rm the impact of photon energy on the SABR image contrast. MCNP is a Monte Carlo particle simulation code developed by the Los Alamos National Lab. Figure 3 2 shows the MCNP geometry used to s imulate the SABR image of a steel

PAGE 28

28 washer on a nylon substrate. The air is represented in purple, the lead in dark blue, the CR plate in light blue, the nylon in yellow and the steel washer in green. The source for this geometry was a circular surface sourc e placed above the lead tiles and which emitted photons at a given energy directed downward (Because the point source for the SABR experiment was 25 inches away, and the total size of the nylon background was 6 inches, it can be safely assumed that all pho tons had the same direction ). Figure 3 3 shows the flux received by each point of the CR plate when the source emitted 500 million photons at 10 keV, 50 keV and 100 keV. The average relative error per pixel of the background for the simulation at 50 keV is close to 50%, whereas, on a SABR image at 75 kVp, 60 mAs and 25 inches, the average statistical uncertainty per pixel is 3%. Therefore, it can be estimated that 500 millions photons in the MCNP simulation correspond to a SABR image with an exposure of abo ut 0.2 mAs only. However, due to the lack of sensitivity of the CR plate, SABR images taken at 0.2 mAs appear dark. The parts of the image in white have not received any photon during the MCNP calculations, each of which lasted roughly 90 minutes, and woul d be represented by very dark pixels on the corresponding SABR image. As explained before, the image with the best contrast is by far the one at 50 keV, which is close to the mean energy used with the SABR technique. The pixels size for each image was 0.5 mm, and each lead tile measures 2 inches. Because the circular source of the MCNP simulation did not cover the whole area, the corners are underexposed compared to the rest of the image. Notes on MCNP Photon Source for SABR Simulations Although, the image quality at 50 keV is better than at 10 or 100 keV, some parts of the CR plate, corresponding to white pixels were not crossed by any photon during the MCNP calculation. Moreover, the relative error of the fl ux for pixels that received very few photons is c lose to 100 %, even after hours of MCNP calculation. In fact, this is due to the very low

PAGE 29

29 probability of photons go ing through the lead tile assembly. Indeed, the aperture holes only take less than 0.5 % of the total area of the assembly, and the mean free path of 50 keV photons in lead is about 0.1 mm, which is less than the tenth of the thickness of the lead tiles, so the probability of a photon to pass through the lead is roughly 4 10 5. As a result, more than 99.5% of photons are wasted if a continuous surface source is used. Consequently, a better source was introduced in all the other MCNP input fi les. The particles were then emitted over the aperture holes between the lead tiles only. Figure 3 4 shows the fl ux received at each point of the CR plate for the two different source types. The image shown in Figure 3 4A was created with the continuous source, with 500 million photons emitted and the calculation lasted for more than 90 m inutes. The other image shown in Figure 3 4B was made using the other source, for which only 5 million particles were emitted in less than 50 minutes (the ine ff iciency of the source sampling is responsible for such a small number of particles emitted per unit time). The di fference in image quality is explained by the fact that on average, the CR plate pixels received ten times more particles with the modified source than with the regular one (this can be seen on the scales). Also the largest uncertainty of fl ux value for a 0.5 mm wide pixel is only about 20% with the modified source Figure 3 5 shows the proximity between the image observed experimentally with the SABR of a steel washer (Figure 3 5A) and the image created after a MCNP simulation of a SABR of the same steel washer (Figure 3 5B) This similitude validates the MCNP geometry and source models used. Finally, it was observed that typically, variations of voltage between 60 and 80 kVp do not seem to dramatically change the contrast between the obj ects and the substrate they are placed on. In general, for most objects and backgrounds 75 kVp, 60 mAs and 25 inches are settings that

PAGE 30

30 o ffer fairly good SABR images and the contrast is rarely improved signi fi cantly by changing the energy or the voltage, as long as the image exposure is acceptable. Effects of the Substrate The substrate, or background, is a plate on which the target objects are placed. Its composition can greatly modify the SABR image. Three types of substrates were used and compared: a ny lon substrate, an aluminum substrate and a lead substrate. The fi rst two o ffered fairly good images, because their scattering -to absorption ratios are high enough. At 30 keV, the approximate average energy of photons for voltages of about 70 kVp, the scatt ering to absorption ratio is 0.3 for the aluminum, and 3.0 for the nylon.7 However, the SABR images of objects placed directly on a lead substrate were very dark, except for the white lines corresponding to the apertures. This is due to the very low scatte ring -to absorption ratio of lead (0.05 for 30 keV photons). Indeed, the vast majority of photons that are allowed through the apertures are absorbed in the lead substrate instead of being scattered back toward the CR plate. In Figure 3 6 can be seen a va riety of objects on a 0.5 inch thick nylon substrate and the resulting SABR image taken at 75kVp with an exposure of 60 mAs, the CR plate being 25 inches away from the X ray source. The tiles used to obtain this image were medium sized tiles, of about 1.5 inches. All the di fferent objects are visible on this SABR image, except the nylon washer, as could be expected, because the average mean free path of photons in that energy range in the nylon is about 4 cm, so a very small fraction of X -rays collided in t he 2 mm thick nylon washer. As a result the CR plate was not shaded by this object enough to detect the nylon washer. It can be noticed that the holes in some of the washers cannot be detected, probably because of their relative position to the aperture gr id (objects that are too close or too far from the aperture lines tend to appear with less contrast).

PAGE 31

31 MCNP calculations con fi rm that on nylon or aluminum background only a relatively strong absorbing material can be detected. Figure 3 7 shows MCNP simulations of SABR images of a steel washer ( Figure 3 7A) and a nylon washer ( Figure 3 7B) laid on a nylon substrate. The steel washer is clearly visible on the fi rst image, whereas the nylon washer is invisible on the other image. These MCNP calculations also used five millions 50 keV photons Figure 3 8 shows a similar SABR image but this time, the various objects were placed on an aluminum substrate. The X ray generator setup was the same: 75 kVp, 60 mAs and at 25 inches. Also, the same medi um sized lead tiles were used. It can be clearly noticed that the quality of this image is not as good as the one with the nylon substrate. This is due to the low scattering to absorption ratio of X rays in aluminum at this energy range compared to nylon. As a result, fewer photons are backscattered toward the CR plate and can therefore be detected. However, because most of the object s laid on the aluminum substrate are strong absorbers they still appear on the SABR image but with a much lower contrast. It becomes extremely di ffi cult to see the hole in the large steel washer or even to see the smallest washers. However, with a pro fi le plot, those features can be easily detected (Figure 3 9 ). Contrary to more conventional X -ray backscatter imaging methods, wi th which the contrast between objects in aluminum and nylon can be easily detected (the nylon being a strong scatter ing material and the aluminum a relatively good absorber), SABR cannot detect a thin nylon washer on an aluminum plate. What appears very su rprising at fi rst can be easily explained by the fact that objects placed on the substrate are merely shading the CR plate from the photons backscattered in the substrate. Because the average mean free path for photons between 20 and 50 keV in the nylon is on the order of 4.0 cm very few photons are absorbed in the nylon washer which does not provide su ffi cient shade for it to be detected on the CR plate. In fact, the SABR

PAGE 32

32 technique was never able to detect scattering material with a very small absorption cross section, which reduces the fi eld of possible applications for the SABR technique to the detection of strong absorbers, like metals. Figure 3 10 shows another SABR image of various objects on a 1.5 mm thick lead background. The setup for this experiment was exactly the same as previously: 75 kVp, 60 mAs of exposure and a distance between the X -ray source and the CR plate of 25 inches. As can be seen in the SABR image, not a single object can be detected at all. The fact that these objects do no t appear on the SABR image, is not due to an underexposure of the CR plate for this particular X ray generator setting, because gradually increasing the exposure did not help to detect any object, and the only e ffect was the broadening of the white lines o n the image. In fact, the impossibility to see these objects is due to the very low scattering to absorption ratio of photons in lead in the 10 keV to 75 keV range (about 0.05).7 Indeed, a large fraction of the photons which pass through the aperture grid are absorbed in the lead and very few are scattered back toward the objects, which prevent these objects from being visible. This SABR experiment was also simulated by MCNP calculations, which gave results very close to what was observed experimentally. Fi gure 3 11 shows the fl ux received by each pixel of the CR plate after MCNP simulation of Shadow Aperture Backscatter Radiography of a steel washer on nylon, aluminum or lead substrate. The white pixels in the third image represent areas that were not cross ed by any photon. Effects of the Size of the Lead Tiles The size of the lead tiles is also an important factor for SABR image quality. First of all, the amount of exposure should be increased with the size of the tiles because for large lead tiles a very small fraction of the X -ray photons that are going through the aperture holes backscatter to ward the center part of the tiles. For a smaller tile size, on the other hand, a large part of the

PAGE 33

33 resulting SABR image is overexposed because the surface taken by the aperture grid (white lines) is increased, and the surface taken by the shadowed areas is decreased (Figure 3 12). The overall trend is that image quality tends to be better for medium -sized or large tiles. The e ffect s of the size of the lead tiles obs erved on the SABR images are in accordance with MCNP simulations too. O n Figure 3 13 can be seen the two simulated images of a steel washer on a nylon substrate, with 1 and 2 inches lead tiles. For both images, the contrast defined as the ratio of the average flux detected over the steel washer to the average flux detected over the nylon background is equivalent (steel-to -nylon ratio of about 0.5%). However, a much larger area of the CR plate is overexposed (red lines) wit h the small lead and this can lead to more hidden features in SABR images 50 keV photons were emitted by the enhanced source for both MCNP runs. Effect of a Gap Between the CR Plate and the Object Another interesting and surprising factor a ff ecting the im age quality is the distance between the CR plate and the objects to be imaged. Indeed, even a small gap of a few millimeters can hurt the image quality a lot. This is shown in Figure 3 14, where the fi rst SABR image represents metallic objects placed on a nylon substrate without any gap between the CR plate and the objects (the CR plate was placed directly on the objects), at 75 kVp, 60 mAs, and 25 inches. The second SABR image shows the same objects but with a 4 mm gap between the CR plate and the objects, the X ray generator setup being the same (75 kVp, 60mAs and 25 inches). It is obvious in these images that introducing even a small gap between the CR plate and the target objects dramatically reduces the image quality. Even by modifying the X ray generat or setup it was not possible to increase the contrast for the SABR image with the 4 mm gap. This implies that the SABR technique using the configurational setups examined in this work, can only be used to detect near -surface defects or Foreign Object Debr is that are very close to a

PAGE 34

34 reachable thin surface. (It was possible for instance to detect large metallic objects just behind aluminum or carbon -carbon composite plates, as seen later in this report.) The explanation for this phenomenon is shown in Figure 3 15. It is at fi rst surprising that for the SABR technique, a small gap between the CR plate and the target object can hurt the image quality so dramatically whereas for a pencil -beam scanning backscatter imaging system, such as the RSD (Radiography by S elective Detection) system, the image quality remains good even with a large gap between the detectors and the target object. In fact, with the pencil beam scanning technique, if multiple -scattered photons are ignored, then, when the beam is over the target object, all the X ray photons reaching the detectors were previously backscattered in the object (Figure 3 15A ). On the other hand, with the SABR technique the entire background is illuminated directly or indirectly through the aperture holes. Conse quently, as the gap between the CR plate and the object is increased, the solid angle with which the area of the CR plate directly over the object sees distant parts of the background is increased, and X -ray photons backscattered in another part of the background reach that area of the CR plate ( Figure 3 15B ). As a result, the contrast is rapidly nulli fi ed. Once again this experimental observation was validated by a MCNP calculation for which the geometry can be seen in Figure 3 16. Figure 3 17 shows the di fference between the images obtain with MCNP simulation of a SABR image of a steel washer on a nylon background with the CR plate directly laid on the object, or with a 4 mm gap. The steel washer appears much sharper, and with more contrast between the obj ects and the background when there is no gap between the object and the CR plate. Effect of a Gap Between the Object and the Substrate The last noticeable e ffect is caused by the introduction of a gap between the object and the substrate, with the CR plate directly on the object. Shadow Aperture Backscatter Radiography of

PAGE 35

35 objects suspended with the help of threads 3.5 inches above the substrate gave images with an extremely good contrast. This phenomenon is explained by the fact that the suspended object is crossed by photons which have been backscattered from every single area of the substrate. Therefore from the point of view of both the object and the CR plate just above it, the substrate acts as a much more uniform secondary source. Moreover, the areas o f the CR plate that are close to the illumination apertures are not crossed by a larger number of particles than the areas under the center of t he tiles. This causes less area of the SABR image to be overexposed. In fact, suspending objects a few centimete rs over the substrate allows detecting objects that could not be detected when the object was in contact with the substrate. The nylon washer could even be detected when suspended over aluminum or lead background, despite the fact that those materials are strong absorbers. Indeed, Figure 3 18 shows the SABR image of a nylon washer suspended 3.5 inches over a lead background, at 70 kVp, 60 mAs and at 25 inches, and proves that this object can be detected when suspended, wher eas it could not be seen in Figure 3 6 for instance. This feature of the SABR method could have some application, for instance to help detect objects located inside of a plane, in contact with the external hull and with an aluminum background a few centime ters behind them A MCNP simulation of a steel washer suspended 3.5 inches over a nylon substrate also proves that the contrast of SABR images is improved when the object is suspended. The geometry of this MCNP calculation can be seen in Figure 3 19 and th e comparison between the MCNP simulated SABR images of a steel washer laid or suspended 3.5 inches over the nylon substrate is shown in Figure 3 20. The background for the suspended washer (Figure 3 20B) is more uniform than in Figure 3 20A, and therefore, the steel washer is more easily detectable.

PAGE 36

36 A B C Figure 3 1 Foreign Object Debris on a nylon substrate : A) p hotograph, B) SABR images at 75 kVp, for 60 mAs at 25 inches C) SABR images at 75 kVp, for 240 mAs at 47 inches Figure 3 2 Geometry of the MCNP simulation of the Shadow Aperture Backscatter Radiography of a steel washer on a nylon substrate A B C Figure 3 3. Flux of photons received after a MCNP simulation by each point of the CR plate for three photon energy: A) at 10 keV, B ) at 50 keV and C) at 100 keV

PAGE 37

37 A B Figure 3 4 Flux of photons received after a MCNP simulation by each 0.5 pixel of the CR plate : A) with the regular source and B) with the modi fi ed, more e fficient source A B Figure 3 5. Steel Washer on nylon background: A) SABR image obtained at 75 kVp, 60mAs, at 25 inches and B) MCNP simulated SABR image based on the flux of photons detected by the CR plate

PAGE 38

38 A B Figure 3 6 Foreign Object Debris on a nylon substrate : A) Photograph and B) SABR image obtained at 75 kVp, 60mAs, at 25 inches A B Figure 3 7. MCNP simulated SABR image s of: A) a steel washer and B) a nylon washer on a nylon substrate

PAGE 39

39 A B Figure 3 8. Foreign Object Debris on an aluminum substrate: A) Photograph and B) SABR image obtained at 75 kVp, 60mAs, at 25 inches Figure 3 9 Profile plot of the previous SABR image on which can be seen two features (the hole in the large washer and a small washer)

PAGE 40

40 A B Figure 3 10. Foreign Obj ect Debris on a lead substrate : A) Photograph and B) SABR image obtained at 75 kVp, 60mAs, at 25 inches A B C Figure 3 11. MCNP simulated SABR image s of a steel washer : A) on a nylon background, B) on an aluminum background and C) on a lead background.

PAGE 41

41 A B C Figure 3 12. Foreign Object Debris on a nylon substrate: A) Photograph, B) SABR image with small lead tiles and C) with large lead tiles, obtained at 75 kVp, 60mAs, at 25 inches A B Figure 3 13. MCNP simulated SABR images of a steel washer on a nylon background : A) with 1 inch lead tiles and B) with 2 inches lead tiles

PAGE 42

42 A B Figure 3 14. SABR images of Foreign Object Debris on a nylon background at 75 kVp, 60mAs, at 25 inches: A) with no gap between the CR plate and the objects and B) with a 4 mm gap A B Figure 3 1 5 Possible paths of photons backscattered in the target object : A) for the RSD technique and B) for the SABR technique with a gap between the CR plate and the target object

PAGE 43

43 Figure 3 1 6 Geometry of the MCNP simulation of the Shadow Aperture Backscatter Radiography of a steel washer on a nylon substrate with a 4 mm gap between the washer and the CR plate A B Figure 3 1 7 MCNP simulated SABR images of a steel washer on a nylon background : A) without a gap and B) with a 4 mm gap between the washer and the CR plate Figure 3 1 8 N ylon washer suspended by a thread over a lead background : A) Photograph and B) SABR image taken at 75 kVp, 60mAs, at 25 inches

PAGE 44

44 Figure 3 1 9 Geometry of the MCNP simulation of the Shadow Aperture Backscatter Radiography of a steel washer located 3.5 inches over a nylon substrate A B Figure 3 20. MCNP simulated SABR images of a steel washer on a nylon background : A) without a gap and B) with a 3.5 inches gap between the background and the washer

PAGE 45

45 CHAPTER 4 SABR: VARIOUS OBJECTS THAT HAVE BEEN IMAGED Foreign Object Debris behind a thin aluminum or carbon -carbon composite plate One of the conditions for the SABR techniq ue to be used to detect Foreign Object Debris is the ability to take images though thin layers of common material s such as aluminum or carbon-carbon composite. The three SABR images of lead and steel pieces on a nylon substrate, which can be seen in Figur e 4 1 were all taken at 75 kVp, 60 mAs, and 25 inches. The SABR image shown in Figure 4 1A was obtained with the CR plate directly on the objects, while the SABR image in Figure 4 1B shows the same objects but with a 1 mm thick aluminum plate placed betwe en the objects and the CR plate. Even though most of the metallic objects can still be detected, the image quality is greatly reduced when the image is taken through 1 mm of aluminum. However, this loss in image quality is not necessarily due only to xray attenuation in the aluminum, but also to the 1 mm gap between the CR plate and the metallic objects induced by introducing the aluminum plate. It was stated earlier in Chapter 3 that a gap between the CR plate and the objects dramatically reduces the cont rast in the SABR image. The SABR image of the same objects in Figure 4 1C was also taken through a 1 mm thick aluminum plate, but with a 5 mm stand -o ff distance between the objects and the aluminum plate. The CR plate was placed on the aluminum. In this i mage, it becomes very di ffi cult to detect even the largest metallic objects. This is again due to the e ffect of a gap between the CR plate and the objects on the SABR image contrast. In order to understand the e ffect of introducing an aluminum plate and a stand o ff distance all three images were taken with the same experimental setup. As a result, the last two images of Figure 4 1 are both slightly underexposed. However, even by increasing the exposure for these two images, the contrast remained very low, and no other object could be detected.

PAGE 46

46 Each of the three previous SABR images was simulated by MCNP calculations, with results displayed in Figure 4 2 The various metallic objects were replaced by a single steel washer. This object can be detected on the fi rst two simulated SABR images (Figure 4 2A and Figure 4 2B) but not in the one with the 5mm gap between the object and the aluminum plate (Figure 4 2C) as was observed experimentally on the SABR images. The same three SABR images were taken, but this time, with a 1.5 mm thick carbon carbon composite plate, consisting of carbon fi bers reinforcing a carbon matrix, instead of the aluminum plate to test for depth penetration capabilities for other materials These SABR images can be seen in Figure 4 3 The fi rst two ( Figure 4 3 A and Figure 4 3B), which are images without and with the carbon -carbon plate, were taken at 75 kVp, 6 0 mAs and at 25 inches. The SABR image in Figure 4 3C was taken with the carbon-carbon composite plate and a 5 mm stand -o ff at 75 kVp and at 25 inches, with an exposure of 90 mAs, to avoid underexposure. Again, the image quality was greatly decreased when a carbon -carbon plate was introduced between the objects to be imaged and the CR plate. And with a 5 mm stand-o ff between the o bjects and the 1.5 mm carbon-carbon composite plate, even the largest objects were almost invisible. SABR Images of a Spray -on -Foam Insulation Block The next object to be imaged was a block of foam (12 inches long, 12 inches wide, 2 inches thick), with 5x5 regularly spaced holes in it, all di fferent in diameter (0.25 0.38 0.5 0.65 and 0.75 ) or in depth (0.125 0.25 0.375 0.5 and 0.65 ). This block of foam, which was placed on an aluminum plate, can be seen in Figure 4 4 Radiography by Selective Detection is able to detect fla ws in this kind spray on-foam insulation that covers the Space Shuttle external tank.2 As stated before, fla ws inside the foam were responsible for parts of the external

PAGE 47

47 tank's insulation striking the Colombia's hea t shield in the 2003 accident. Detecting fl a ws in the foam could help prevent such an accident from happening again. At fi rst, the 5 holes with the largest diameter (0.75 inches) were imaged with the SABR technique. However, contrary to what can be obtaine d with the Radiography by Selective Detection, the holes are invisible on the SABR image. This can be seen in Figure 4 5 This SABR image was taken at 70 kVp, 60 mAs, and 25 inches with the large tiles (2 inches). Di fferent energy levels and exposures were tried but the holes could never be detected in the SABR images. In order to detect those holes, a dime ( c upronickel, 18 mm in diameter) was placed in each of them. This is shown i n Figure 4 6 The SABR image was taken at 70 kVp, 60 mAs, and 25 inches. On ly the coins placed in the two shallowest holes were clearly visible. This can again be explained by the dramatic loss of contrast in the SABR images when a gap between the objects and the CR plate is introduced. Because a dime is about 1.35 mm thick, the true gap between the CR plate and the last clearly visible dime measures only roughly 0.5 mm. A surface plot shows that the third dime is also visible with some image processing as shown in Figure 4 7 However, it is very unlikely that this third coin would have been detected if the operator did not know where to look. Because of the very low density of this foam, it is likely that the image contrast would be the same if the coins were introduced inside of the foam at the equivalent depths instead of ju st being placed in holes. SABR Images of a Corroded Piece of Aluminum A small, extremely corroded plate of aluminum, shown in Figure 4 8 A was also imaged with the SABR technique. The ability to see corrosion with an X ray backscatter technique can be used for instance during the maintenance of airplanes to check the general state of the inside

PAGE 48

48 of the plane, without the need of tearing it apart. The SABR image on Figure 4 8B was taken at 70 kVp, 120 mAs, at 25 inches. The exposure was increased compared to most of the previous scans because almost the entire nylon substrate was covered by the 1 mm thick corroded aluminum plate which has a high absorption cross -section. Even by increasing the exposure, the image quality remained very poor, and some of the hol es could not even be detected. Also the corrosion is invisible on this SABR image. A similar SABR image was taken, with the aluminum plate suspended between two 5 cm thick blocks of lead, about 3.5 inches above the nylon substrate, as shown in Figure 4 9A Th e corresponding SABR image ( Figure 4 9B) was also taken at 70 kVp, 120 mAs and 25 inches. As shown earlier, it is not surprising to observe that the image quality was greatly improved compared to when the corroded part of aluminum was laid on the nylon substrate. For this setup, lead and aluminum can also be used for the background, with no noticeable loss in relative contrast. This was surprising because much brighter images had been expected for the nylon background than for the lead background. Also, despite the relatively good quality of this SABR image, the corrosion itself could not be detected, mainly because the contrast could not be improved enough for it to be visible. Another SABR image of this corroded aluminum part was taken at 75 kVp, 120 m As and at 25 inches, through a 1 mm thick aluminum plate. This image, along with the experimental setup can be seen in Figure 4 10. As mentioned earlier, the introduction of an aluminum plate decreased the image quality by a large factor. However, some of the holes in the corroded aluminum plate are still slightly visible with the help of some image enhancement, as can be seen in Figure 4 11. The corrosion can obviously not be detected through the aluminum plate.

PAGE 49

49 Other Objects that have been Imaged with the SABR Technique SABR images have been taken of many di fferent objects, including metallic objects inside a block of foam, bones covered by tissue simulant material, objects made of plastic (a s crewdriver for example) and other highly scattering materials. None of them showed a reasonable image of the objects, mainly because of the inability of the SABR technique to detect materials in which the mean free path of X ray photons is too long, and because this technique using the configurational setups examined in this work, can only image fl at, near surface objects, due to the dramatic loss in image quality when the CR plate is not directly in contact with the objects.

PAGE 50

50 A B C Figure 4 1 SABR of Foreign Object Debris on a nylon background taken at 75 kVp, 60 mAs, 25 inches : A) without an aluminum plate B) with a 1 mm thick aluminum plate between the CR plate and the objects and C) with the 1 mm thick aluminum plate with a 5 mm stand -o ff distance. A B C Figure 4 2 MCNP simulated SABR image s of a steel washer on a nylon background: A) without an aluminum plate, B) with a 1 mm thick aluminum plate between the CR plate and the washer and C) with the 1 mm thick aluminum plate with a 5 mm stand -off.

PAGE 51

51 A B C Figure 4 3 SABR of Foreign Object Debris on a nylon background taken at 75 kVp, 60 mAs, 25 inches : A) without a carbon -carbon plate B) with a 1 .5 mm thick carbon -carbon plate between the CR plate and the objects and C) with the 1 .5 mm thick carbon -carbon plate with a 5 mm stand -o ff (taken at 90 mAs) Figure 4 4. Block of spray on-foam insulation

PAGE 52

52 A B Figure 4 5 Five holes 0.75 inch in diameter and of di fferent depth in a spray on-foam block : A) p hotograph and B) SABR image at 70 kVp, 60 mAs, and 25 inches A B Figure 4 6 Five dimes placed in holes drilled in foam, 0.75 inch in diameter and of di fferent depth: A) p hotograph and B) SABR image at 70 kVp, 60 mAs, and 25 inches

PAGE 53

53 Figure 4 7 Zoom of the area of the previous SABR image that contained the thir d coin. A B Figure 4 8 Corroded aluminum plate on a nylon background: A) p hotograph and B) SABR image at 70 kVp, 120 mAs, and 25 inches

PAGE 54

54 A B Figure 4 9 Corroded aluminum plate suspended 3.5 inches over a nylon background: A) p hotograph and B) SABR image at 70 kVp, 120 mAs, and 25 inches A B Figure 4 10. Corroded aluminum plate covered by a 1mm thick aluminum plate and suspended 3.5 inches over a nylon background: A) photograph and B) SABR image at 70 kVp, 120 mAs, and 25 inches

PAGE 55

55 Figure 4 11. Image enhancement of the previous SABR image with features surrounded in red.

PAGE 56

56 CHAPTER 5 SABR: CONCLUSIONS The Shadow Aperture Backscatter Radiography technique is an X ray backscatter imaging technique tha t is based on a snapshot acquisition method instead of the scanning acquisition method which is used for X -ray backscatter imaging. It uses assemblies of lead tiles to block the transmission signal. This prevents overexposure of the CR plate as the X ray photons cross it to reach the target objects. By limiting the superimposition of irradiation photons on the backscatter signal, the signal to -noise ratio is dramatically improved compared to the SBR technique. The snapshot mode of acquisition allows the operator to obtain images much faster than with systems that use scanning to acquire images (two images can be obtained in a couple of minutes with the SABR technique while the acquisition time with the RSD system currently in use at UF would be of the ord er of hours ). But on the other hand, the image quality is not nearly as good, and fewer objects can be detected in a SABR image. Also, because the areas just under the apertures, between the tiles, are overexposed (lines about 1 mm wide), small objects can be completely hidden on the image and therefore, at least two images of a given target need to be taken in order to be sure to see all the areas in examined target The SABR technique o ffers fairly accurate X -ray backscatter images of near surface highly absorbing objects (metallic objects for instance) that are placed on a substrate made of a good scattering material such as nylon. Images acquired on more absorbing substrates, such as aluminum, can be good too despite a los s in contrast. However, it was n ot possible to detect objects made of highly scattering material. For example, a nylon washer on an aluminum substrate remained invisible in the SABR images when a steel washer would have appeared, and this was true no matter what X ray energy level was us ed. Also, introducing a small gap between the CR plate and the objects reduces the image contrast, and gaps as small as 5 mm can prevent

PAGE 57

57 any object from being detected. However, a gap between the object and the background can greatly increase the SABR imag e quality, to the point where objects that did not appear when laid directly on the substrate were very clear in the image when suspended a few inches over the background. Some large metallic objects could also be seen through a plate of carbon -carbon comp osite or aluminum of about 1 mm in thickness which was placed directly over them. However, in this case the image quality was greatly decreased, and most of the smallest objects (less than 1 cm in size) were invisible. When a stand -o ff was introduced betwe en the plate and the objects, nothing could be seen in the SABR image because of the gap between the CR plate and the objects This problem reduces the fi eld of possible applications of the SABR technique because in most cases, the objects that need to be detected are not directly in contact with an external surface. A lead or tungsten collimation grid placed between the target and the CR plate could prevent the averaging of contrast

PAGE 58

58 CHAPTER 6 CSD SXI: INTRODUCTION Computed Image Backscatter Radiography Computed Image Backscatter Radiography (CIBR) is a nother X ray backscatter imaging technique developed by the SXI group. It employs a fan beam source and rotational motion rather than a pencil beam source and rastering motion; it uses the same collimated scintillation detectors as RSD. It was originally designed to acquire X -ray backscatter images faster than RSD This is b ecause a larger fraction of the X ray photons produced in the tube is used to obtain the image, the count rate is higher th an with RSD, and therefore a scan of the same object and with the same statistical uncertainty tolerance is faster with the CIBR technique. C omputed Image Backscatter Radiography requires scanning across the target objects at different angles with relative ly small increments in order to obtain acceptable images, and it also needs image reconstruction, much like Computed Radiography.8 However, further research is required to improve resolution and reduce reconstruction artifacts through algorithm improvement Figure 6 1 shows a comparison between the RSD and CIBR images of letters of lead on nylon. Some artifacts are visible on Figure 6 1B. Backscatter Radiography using an Uncollimated Segmented Detector T he possibility of adding a linear pixe lated detector parallel to the fan beam source, to the CIBR technique in order to acquire more information on the target object was considered This case, with geometry shown in Figure 6 2 was simulated using MCNP In this simulation, a lead strip (6.0 cm x 2.0 cm x 0.1 cm) is fitted in a block of nylon, which is a highly scattering material contrary to lead which is a strong absorber. A 1 mm wide and 6 cm long fan beam normal to the X axis illuminates the strip of lead and the nylon around it. A linear Gadolinium Orthos ilicate (GSO ) array detector (4.0 cm x 1.5 cm x 0.1 cm) is placed 3.0 cm above the surface of the

PAGE 59

59 object, and starts 1.5 cm away from the fan beam. A flux mesh tally records the flux averaged over the pixe lated detector. The dots seen in Figure 6 2 represent the collision sites of the X ray photons in the matter, their colors varying continuously from red corresponding to the highest energy particles prior to collision (in this case, 60.6 keV) to blue corresponding to the lowest energy (6.3 keV). In order t o accelerate the MCNP simulation, a DXTRAN sphere (not shown in Figure 6 2 ) was included in the MCNP input in such a manner that it encircled the entire segmented detector.9 The source of X ray photons for this simulation was located 3 cm above the surface of the target object because a non -negligible flux was detected by the upper part of the detector when the source was higher due to scattering in air toward the detector. In practice, the source would need to be placed at a 20 cm distance to obtain a wide fan beam but the source scattering in air problem could be solved by ensuring that the top of the detector is shielded. The voltage used for this simulation was 65 kVp, meaning that the X ray spectrum ranged from 0 to 65 keV. The energy distribution of the X ray source at 65 k Vp generated by the program XRSPEC from the SXI group is shown in Figure 6 3 The hump around 10 keV is due to the L series characteristic X rays for tungsten.10 The flux of X -ray photons measured by the mesh tally in the MCNP simu lation was then compared with that obtained for the exact same geometry (4.0 cm x 1.5 cm x 0.1 cm GSO array detector, 6.0 cm x 2.0 cm x 0.1 cm lead strip on nylon) but with a pencil beam source. This MCNP simulation geometry can be seen in Figure 6 4 As f or the previous case, one measure of the flux for every 1 mm pixel (in the y-direction) was taken over the volume of the detector. Instead of having to do multiple simulations for each value (forty 1.0 mm pixels), the source was defined as 40 circular sour ces, 1.0 mm in diameter

PAGE 60

60 placed on a line and 40 different flux tallies were obtained with the SCD option using FU cards.9 This association of data cards allowed the tally corresponding to each region of the detector to count only the particles coming from the corresponding source. As a result, there is no interaction between the different parts of the image. Figure 6 5 shows the averaged normalized flux in the uncollimated linear array detector with a pencil beam and a fan beam X ray source with the associ ated relative error It is clear in Figure 6 5 that the averaged flux in the detectors in the case of the pencil beam shows a strong contrast between when the pencil beam is above the nylon or the lead, because of the fact that lead is a highly absorbing material (which results in a low count rate of backscattered X rays) and that nylon is a highly scattering material (which results in a high count rate). However, in the case of the fan beam source with an uncollimated pixe lated detector, the flux shows very little difference between pixels located ove r the nylon and those over the lead. This phenomenon can be simply explained by the fact that each pixel can detect photons backscattered toward any direction and not exclusively those coming directly from below. Consequently, each pixel of the detector se es photons backscattered in both the lead and the nylon as can be seen in Figure 6 6 and as a result, the contrast is averaged out and greatly reduced. The exact same effect was observed in the case of the Snapshot Aperture Backscatter Radiography (SABR), a backscatter X ray imaging technique based on a snapshot illumination, and which offers very poor contrast when there is a stand off between the target object and the detector (a CR plate in the case of SABR experiments). Very little information can be o btained from the flux measured for the fan beam case in Figure 6 5 when compared to the flux for the pencil beam, and it does not seem that a fan beam

PAGE 61

61 associated with an uncollimated pixe lated detector could really be of any use in backscatter radiography.

PAGE 62

62 A B Figure 6 1 Backscatter radiograph y images with a 1 mm resolution of letters of lead on nylon: A) with the RSD technique with 1 mm pixels and B) with the CIBR technique with 10 degrees rotational increments and 1 mm radial increments A B Figure 6 2 Geometry of the MCNP simulation of the backscatter radiography of a lead strip on a nylon background using an uncollimated segmented detector: A) view of the XZ plane and B) view of the YZ -plane

PAGE 63

63 Figure 6 3. Energy spectrum of the X ray source at 65 kVp A B Figure 6 4 Geometry of the MCNP simulation of the pencil beam backscatter radiography of a lead strip on a n ylon background: A) view of the YX -plane and B) view of the YZ plane

PAGE 64

64 Figure 6 5. Comparison between the normalized flu xes observed in the uncollimated array detector with a fan beam source and with a pencil beam source Figure 6 6. Possible paths of backscattered X -ray photons toward the uncollimated segmented detector with a fan beam source

PAGE 65

65 CHAPTER 7 CSD SXI: PRINCIPLE AND MONTE CARLO SIMULATIONS Principle of Collimated Segmented Detector Scatter X -ray Imaging A more suitable way to use a segmented detector with a fan beam source would be to somehow force the photons backscattered in the target object to be detected only by the corresponding pixels (i.e. the pixels located directly over the location at which they backscattered) through collimation. This can be done by fitting a grid made of a highly absorbing material onto the pixe lated detect or, in such a way that pixels or clusters of pixels would be separated from each other at regular intervals by the grid and could only see photons coming from a limited solid angle. Such a grid could be made of tungsten or lead, which are both strongly absorbing material s Tungsten would be preferable to lead because it s macroscopic absorption cross -section is about 26 % higher at 30 keV7, due to a higher electron density, and also because it is a much stronger material and would more easily re sist deformation especially if the collimation grid is very fine. Figure 7 1 shows the geometry of a MNCP simulation of Collimated Segmented Detector backscatter radiography of a lead strip on a nylon background. The collimation grid, shown in light blue, is composed of 0.4 mm thick tungsten plates (2 cm x 1 cm) separated by a regular 1 mm spacing. A 0. 4 mm thick tungsten shielding was placed all around the array detector and the collimation plates to ensure uniform collimation. The resolution of the image is limited by th e spacing between collimation plates but also depends on other factors, such as the distance between the target object and the detector, the thickness of the tungsten plates and their collimation length (1 cm in this particular simulation) In practice, the spacing between the plates should be adjusted in such a manner that it would correspond to a natural number of pixel widths which would then be grouped to form a cluster M atching the grid with the pixel clusters could

PAGE 66

66 be one of the technical difficulties associated with this imaging method. In this chapter this problem will be ignored and cluster s of pixels of the segmented detector will be referred as pixel s of the image With such a collimation grid, each pixel is shielded from the X -ray photons that do not come from directly under it, as seen in Figure 7 2 Therefore, it can be expect ed that this configuration reduces the averaging of the contrast between the pixels, which is the reason why an uncollimated array detector with a fan b eam source c ould not work. Figure 7 3 shows the comparison between the fluxes recorded by each pixel of the segmented detector with a fan beam source and with a pencil beam source from a n MCNP simulation. In both cases, the tungsten collimation grid is placed below the detector to make the fluxes comparable. It is clear in Figure 7 3 that t he introdu ction of a collimation grid has improved the contrast between the lead and nylon regions in the case of the fan beam source and that th e contrast in now very similar with a fan beam source and with a pencil beam source In fact, the contrast is even slightly higher with the fan beam source ; the relative contrast, defined as the ratio of the ave rage fluxes over nylon and over lead is 36 with the fan beam against 25 with the pencil beam source. A possible explanation for this is shown in Figure 7 4 The scattering to absorption cross section ratio for 30 keV photons is 5% in lead and 400% in nylon and their mean free path in nylon is about 3 cm against 0.06 mm in lead.7 Therefore it can be expected that with a fan beam source, a large number of particles can travel inside the nylon and follow the same path as particle 1 in Figure 7 4 This increase s the flux detected by pixels located over nylon with a fan beam source by 70% relative to when the pencil beam source is used On the other hand, pixels located over lead will see a flux dominated by short path and single scatter photons since the scatter ingto absorption ratio is so low in lead. Consequently,

PAGE 67

67 with a fan beam source, the flux detected by pixels located over nylon is increased by 70% while the one over lead is only increased by 15% and this results in a higher contrast between the nylon and the lead. For a pencil beam however, this is not true because only the area directly under the pixel is illuminated so the surrounding regions cannot contribute to the flux. This theory is confirmed by the fact that the flux normalized to the number of s ource particles is 1.7 times higher with a fan beam source than with a pencil beam source over the nylon region and only 1.2 ti mes higher over the lead region. The average fraction of source particles in the detected fluxes was, in the case of the uncollim ated detector ( Figure 6 5 ), 8 49 10 5 and 2 78 10 6 with a fan beam and a pencil beam source respectively, and, in the case of the collimated detector ( Figure 7 3 ), 1 64 10 6 and 9 87 10 7 with a fan beam and a pencil beam source respectively. The uncertainties on these values are on the order of 1% for the uncollimated detector, and 4% for the collimated detector. These figures confirm the theory described in Figure 7 2 For the fan beam source, t he flux record ed in the detector is almost two order s of magnitude smaller when the tungsten collimation is introduced because then, the segmented detector mainly detects photons that are backscattered up and toward the corresponding pixel while other photons are absorb ed in the tungsten. The decrease in recorded flux with a pencil beam source when the tungsten collimation is introduced is much smaller, less than a factor of three Even though the count rate recorded with the fan beam source is less than twice as large a s with the pencil beam source in the case of the collimated detector, it should be noted that experimentally, a fan beam source yields many more

PAGE 68

68 photons per unit time than a pencil beam source, and this is why an acceleration relative to the pencil beam te chnique s is expected. Resolution Figure 7 3 also shows that the resolution with the fan beam source is lower than with the pencil beam source W hereas going from within 10% of the largest flux value (corresponding to the nylon) to within 10% of the smallest (corresponding to the lead) takes only 1 mm in the case of the pencil beam, the same transition takes about 3 mm for the fan beam method Cont rary to pencil beam imaging technique s for which the image resolution is roughly the step between each measurement, the calculation of the resolution with a fan beam source is slightly more complex. Figure 7 5 shows the span of the target object that can b e seen by a collimated detector with a fan beam source, where P is the pixel size H is the distance between the target object and the detector, L is the collimation length and R is the resolution. From the intercept t heorem, the resolution, R is defined a s: = 2 2 (7 1) In the case of the lead strip laid on the nylon block for which P = 0.6 mm, H = 30 mm and L = 10 mm, the resolution R was 3.0 mm which is about 3 times larger than the resolution obtained by the pencil beam technique. This value of R = 3 mm for the fan beam versus R = 1 mm is confirmed by the results from Figure 7 3 in which the transition from the lead to nylon flux value is done in about 3 mm with the fan beam source against only 1 mm with the pencil beam source How ever, the resolution cannot be so simply defined; indeed, the area below the central part of the target object can be seen from every part of the pixel whereas some other areas can only be seen from a fraction of the pixels surface, and therefore have les s impact on the measurement of the flux. In fact, if the

PAGE 69

69 target object is homogeneous, it can be assumed that each point y = t of a pixel receives equal flux contribution from every point of the target object that it sees, as seen in Figure 7 6 The contri bution of this point of the pixel to the image is therefore: ( ) = 1 if ( tan 1) < < ( + tan 2) 0 else (7 2) where: tan 1=t L and tan 2=P t L (7 3) Consequently, a pixel with width P would see the following contribution from each point y of the target object: ( ) = ( ) 0 (7 4) Figure 7 7 shows the normalized contribution from a homogeneous target object to a pixel of the image of width P = 0.06 cm, with collimation length L = 1 cm and with an H = 3 cm stand -off. It can be noticed that the resolution is indeed 3 mm, as shown previously. The area directly under the pixel has a contribution of 1 because it can be seen from any part of the image pixel. O n the other hand, areas that are not located directly under the pixel can only be seen by a fraction of its surface, which results in a smaller contribution to that pixel. Influence of Collimation Figure 7 8 shows the geometry of another MCNP simulation. In this case, the target object is the same lead strip located 2 cm deep inside a nylon block. For this simulation, for which we obviously expect the contrast to be smaller between the two types of material, the stand-off distance, H, is 3 cm, and the pixe l width and the collimator length are 1 mm and 1 cm, respectively. For all the simulations in this section, the maximum energy of X ray photons was set to 120 keV, in order to have a better depth penetration in the nylon.

PAGE 70

70 T he comparison between the flux ob served by the collimated and shielded detector with the pencil beam and the fan beam sources is shown in Figure 7 9 In this case, the contrast is also higher with the fan beam source, with a relative contrast of 1.61 between the nylon and the lead against 1.34 with the pencil beam source. However, whereas the resolution remains close to 1 mm with the pencil beam technique, it seems that the resolution with the fan beam was greatly reduced to about 5 mm (transition from the value of the flux over the nylon to over the lead), and indeed the formula for the resolution gives R = 5.4 mm at the depth of the lead strip. The average fraction of particles counted in the fluxes relative to the source was 5 76 10 6 with the fan beam source and 3 18 10 6 with the pen cil beam source Other MCNP simulations were done with different collimation length s in order to improve contrast as well as resolution, and to understand the effect of collimation on resolution. In practice, it would be too expensive and unpractical to purchase set s of t ungsten grids with different collimation length s Instead, simply lowering the position of the collimated detector should be almost equivalent to extending the length of the collimator. The other collimation lengths that were tried are L = 1.5 cm, L=2 cm and L = 2.5 cm for which the theoretically calculated fan beam image resolutions are, respectively, R = 3.4 mm R = 2.4 mm and R = 1.8 mm. Figure 7 10 shows the difference between the detected fluxes for the four collimation lengths with a fan beam source. The resolution improves as the collimation is increased: R is about 4 mm for the 1.5 cm collimation, 3 mm for the 2 cm collimation and 2 mm for the 2.5 cm collimation. The best contrasts were obtained with the 2.0 and 2.5 cm collimation, but, in the case of the lat te r, the count rate was reduced by a factor of two due to over -collimation. The average fraction of source particles in the detected fluxes was 5 76 10 6 with 1 .0 cm

PAGE 71

71 collimation, 2 40 10 6 with 1.5 cm collimation, 9 95 10 7 with 2 .0 cm collimation and 5 06 10 7 with 2.5 cm collimation D epth information and Possibility of 3D Imaging 2D Collimation Grid for Better Depth Resolution Replacing a linear grid by a 2D grid, as seen in Figure 7 11, could improve the depth selection. S uch a grid could prevent photons that do not backscatter in the region of interest from being detected. In such a case, it might be more efficient to tilt the detector toward the fan beam ; otherwise, the count rate for the regions that are far away from the source would not detect anything. For th e particular grid used in the following M CNP simulations t here were 6 rows of 40 pixels for a total of 240 grid elements, each of which was 2.5 mm long and 1 mm wide. The thickness of tungsten plates was 0.4 mm. Three MCNP simulations, with geometries shown in Figure 7 12, were performed with a linearly collimated segmented detector with no inclination, with a linearly collimated segmented detector at a 45 inclination, and with a 2D grid collimated segmented detector at a 45 inclination. In all cases 1.5 cm collimation was used. The tar get object consists of a 1 cm thick and 2 cm wide air gap located 1.0 cm deep inside an aluminum block. In the case of the linear detector with no inclination, the stand off was 3 cm from the bottom of the detector (1.5 cm from the bottom of the collimator ). The collimation length was chosen to obtain the best possible contrast for the linearly collimated detector array without inclination. The maximum energy of the photons was 120 keV. Figure 7 13 shows the flux recorded in each pixel of the segmented dete ctor for all 3 cases from the MCNP simulations For the 2D collimated detector, each of the 6 rows had different weights to prevent the overall flux from being dominated by the first row only ; the count rate detected by the first row is about 3 times large r than for the second row, and even more than for

PAGE 72

72 the others. This is why the flux for each row was normalized and then were all averaged to give only one value. The graph from Figure 7 13 shows that the contrast in the detected flux between the air and t he aluminum is in the 5 to 10% range for both the linearly collimated detector without any inclination and the 2D collimated array detector at a 45 inclination. It is lower than 5% in the case of the linearly collimated detector at a 45 inclination due t o poor collimation. The areas of the array detector corresponding to the air should detect a lower flux because the scattering cross section of photons in air is lower than in aluminum. However, it can be noticed that the contrast is inverted in the case of the linearly collimated detector without any inclination. This is because with these settings, for which the contrast is optimal, the array detector is over -collimated so the flux that is recorded represents mostly the area located under the air. Becaus e the absorption cross section of photons in air is smaller than in aluminum, more photons reach that depth when they went through the air first and therefore the flux recorded by the detectors is higher for this region. This is one type of shadowing effec t .3 However, the flux recorded by each row of the 2D collimated array detector contains much more information than the flux averaged over all rows. Possibility of 3D Backscatter Imaging Each row of the 2D collimated array detector corresponds to a certain depth. A closer view of the 2D collimated array detector is shown in Figure 7 14. At the intersection with the fan beam plane, row 1 corresponds to a depth of 0.80 cm below the surface of the aluminum, row 2 to a depth of 1.14 cm, row 3 to 1.50 cm, row 4 to 1.85 cm, row 5 to 2.21 cm and finally, row 6 corresponds to a depth of 2.59 cm. However, each row represents in fact a certain depth distribution which can be seen as resolution on t he z axis. Therefore the contribution of each z -level of the target object to every

PAGE 73

73 row of pixels of the array detector can be obtained similarly to the contribution of each point of the y d of the detector, the expression for the contribution is slightly more complicated. Also, an absorption term must be introduced to account for depth penetration. Each point t of a pixel row ( 0 < t < P ) can directly see the photons scattered in the plane containing the fan beam (red arrow in Figure 7 15). Therefore : ( ) = 1 if ( tan ( d 1) ) < < ( tan ( d+ 2) ) 0 else (7 5) where: tan 1=t L and tan 2=P t L (7 6) and, with Hc and Dc the height of the center of the pixel and its distance to the beam, respectively: = + 2 sin ( ) = + 2 cos ( ) (7 7 ) However if P << D then H c and D c. The contribution of each point of the z axis to the pixel row is then: ( ) 2 / ( ) 0 (7 8 ) where mfp is the mean free path of photons in the medium. The exponential term allows taking into account the absorption in the medium a s the photons travel downward and then upward, toward the detector after backscattering. Figure 7 16 shows the contribution of every z level of the target object to each pixel row of the array detector. The value of the mean free path used in this case was 1.5 cm which is the average mean free path of 120 keV photons in the aluminum surrounding the void. From th e graph in Figure 7 16, it appears that row 1 is under -collimated and only a small fraction of photons detected by this row would have been backscat tered in the region of the void,

PAGE 74

74 located between 1.0 and 2.0 cm deep in the aluminum; the image would be dominated by the flux of photons scattering in the region above the void. On the contrary, rows 5 and 6 are over collimated. Rows 2, 3 and 4 are all re asonably collimated even though they also see significant flux contribution from the regions located over and below the void. This is a simplified contribution model (it was assumed that photons remain on the fan beam plane as they travel through the targe t object). Figure 7 17 sh ows the fluxes detected in the MCNP simulation by each of the six rows of pixels of the 2D array detector with a 45 inclination in the case of the 1 cm thick air gap. All rows show contrast of at least 10% between the void and the aluminum, with rows 2 (30%), 3 (40%), 5 (30%) and 6 (50%) showing the best contrasts. It should be recalled that the best contrast that could be achieved with a linearly collimated detector wi th no inclination was below 10%. It can be noticed that the contrast in the flux detected at rows 1, 2, 3 and 4 is inverted compared to the flux at rows 5 and 6. This is because rows 5 and 6 are over -collimated and are tallying fluxes of particles backscat tered below the region of the void. Indeed the absorption cross section of photons in aluminum is higher than in air, therefore X rays reaching that depth are more numerous if they traveled through air. This is the same shadowing effect as was observed for the linearly collimated detector in Figure 7 13. The flux measured at row 4 shows a smaller contrast than at rows 3 and 5 because in this collimation range, the contrast goes from negative (row 3) to positive (row 5), as the shadowing effect cancels out t he difference in scattering cross section in air and aluminum. It should be noted that the fluxes detected in this MC NP simulation were, on average, 1 36 10 6 for row 1, 7 14 10 7 for row 2, 4 27 10 7 for row 3, 3 19 10 7 for row 4, 2 56 10 7for row 5 and 2 02 10 7 for row 6. By comparison, the linearly collimated

PAGE 75

75 detector with no inclination saw an average flux of 9 04 10 7. Because the flux detected for rows corresponding to the highest depths is low, the statistical error for these rows can be high. Als o, actual detection of a feature is always better than detection of its shadow. This is why it is expected that rows 2 and 3 give the best results. The fluxes measured by the different rows of a 2D collimated segmented detector put together could form a 3D image of a feature. It is roughly equivalent to changing the collimation of a linearly collimated detector for each depth. Although the statistical error is larger, the contrast is much higher and it only requires one scan so it is faster. However, true 3D imaging with a backscatter X ray method is difficult since features can show on an image even though they are below the collimation plane, due to shadowing effects. Therefore, some features could be hidden by other s located at a shallower point CSD -SXI 3D Imaging : Example of a Ring of Air Inside Aluminum X ray backscatter images of a ring of air located 1.0 cm below the surface of an aluminum block were obtained using MCNP simulations. This ring, shown in Figure 7 -18, has an inner diameter of 0.5 cm, an outer diameter of 1.5 cm and is 1.0 cm high. Images of this ring were simulated for three different setups: a pencil beam system with a NaI detector 5.9 cm in diameter, a fan beam system with a linearly collimated GSO array detector with no inclination and a fan beam system with a 2D collimated GSO array detector at a 45 inclination. The pencil beam system is similar to the RSD system developed by the SXI group; the detector is located 9.5 cm away from the pencil beam and the collimation used varied betwe en 0.1 and 1 .5 cm. The distance between the bottom of the detector and the surface of the target is 3.0 cm. The geometries of all three systems are shown in Figure 7 19 The average normalized flux detected per 1 mm pixel at the bottom surface of the dete ctor after MCNP simulations for the pencil beam systems with a 1 cm collimation was equal to

PAGE 76

76 4 08 10 6 averaged over a total detector surface of 2 7.3 cm2. In the case of the linearly collimated segmented detector the average normalized flux is 4 02 10 6 averaged over a pixel surface of 1 5 10 1 cm2. For the 2D collimated segmented detector system, the average normalized flux was 4 0 10 6 for row 1, 2 1 10 6 for row 2, 1 2 10 6 for row 3, 8 9 10 7 for row 4, 7 2 10 7 for row 5 and 5 6 10 7 for row 6 averag e over a pixel surface of 2 5 10 2 cm2. This means that the detection rate of the NaI detector is about two hundred times larger than for each linearly collimated pixel, and more than one thousand times larger than for each 2D collimated pixel. However, it should be remembered that for the pencil beam system, only one pixel value can be measured at a time. So if a 40x40 1 mm pixel image is considered, the pencil beam system will have to spend some time on each of the 40 pixels of the line, while the fan bea m systems will record one entire 40 pixel line of the image at a time. In this case, the pencil beam system would be 4.5 times faster than the linearly collimated array detector system w ith the same statistical error on average for the pixels of the image. T he 2D collimated array detector system would be 27 times slower with the same statistical uncertainty for the NaI detector and row 1, or almost 200 times slower with the same statist ical uncertainty for the NaI detector and row 6, although six images would be acquired simultaneously. However, it should be remembered that these values are given for the image acquisition of a 4x4 cm target object. The use of a longer segmented de tector would reduce these ratios and it can be estimated that with a 1 meter long pixelated detector CSD SXI would become faster than pencil beam systems. CSD -SXI would then be suitable for large area inspection, such as the inspection of the hull of commercial airplanes. Some manufacturers, such as Envision Product Design can build linearly segmented detector up to 2 meters long.11

PAGE 77

77 Figure 7 20 shows the MCNP generated X ray backscatter images (40x40 pixels, 1 mm per pixel) of the air ring located 1 cm deep inside aluminum with a RSD -like pencil beam system. The four images were obtained with different collimation lengths: 0.1 cm, 0.5 cm, 1.0 cm and 1.5 cm. The average flux values for each of these images were 1 24 10 5, 8 62 10 6, 4 08 10 6 and 8 58 10 7, respectively averaged over a surface of 27.3 cm2. The corresponding statistical error was below 8% for all collimation levels (less than 1% a t 0.1 cm). Each of these images required 1600 MCNP runs, each of them taking a few seconds, one for every pixel. It seems that the best images of the ring with this pencil -beam system are the first two, which were generated with 0.1 and 0.5 cm collimation. However all four images are rather blurry and have fairly poor contrast. Moreover, the exact dimensions of the ring cannot be measured in any of these images. Figure 7 21 shows the MCNP generated images of the air ring in aluminum with a fan beam source and using a linearly collimated array detector without any inclination. These three images were generated using different collimation lengths (0.5 cm, 1.0 cm and 1.5 cm) and the average flux recorded by the MCNP simulations were 3 98 10 5, 1 35 10 5 and 4 02 10 6 for a pixel surface of 0.15 cm2. The statistical error for each pixel of these images was below 7%. 40 MCNP runs (30 minutes each) were necessary to generate these images. The last two images show the best contrast. It can be noticed that the ring appears lighter than the aluminum as was previously observed in Figure 7 13. Also the ring is not perfectly centered in the middle of these images, and this indicates that they show some kind of shadow of the ring rather than the ring itself. Even though t he contours of the ring of air seem clearer than in the case of the pencil beam system, the inverted contrast and the shadowing effect make these images more difficult to read.

PAGE 78

78 Figure 7 22 shows the MCNP generated images of the air ring in aluminum with a fan beam source using a 2D collimated array detector at a 45 inclination Each of these six images represents one of the rows of the collimated pixe lated detector and corresponds to the following depths in the aluminum : 0.80 cm, 1. 14 cm, 1.50 cm, 1.85 cm, 2.21 cm, and 2.59 cm. It should be recalled that the air ring is located between 1 and 2 cm below the surface of aluminum. This is better shown in Figure 7 23. All six images were generated using 40 MCNP runs (30 minutes each), one for each row and, in pr actice, they would be generated by only one scan across. The average flux recorded by the MCNP simulations was 4 00 10 6 for row 1, 2 09 10 6 for row 2, 1 24 10 6 for row 3, 8 94 10 7 for row 4, 7 19 10 7 for row 5 and 5 61 10 7 for row 6. The statistical error for each pixel was of the order of 8% for the first row and about 15% for the last row. The image corresponding to the first row is under collimated and this is why its contrast is so poor. The images generated by rows 4, 5 and 6 are all over -collima ted (the collimation plane is either at the bottom of or below the ring). As a result they show either poor contrast or blurry ring contours. Also, for the last two rows, the ring can only be detected because of the shadowing effect and the contrast is inv erted. However for rows 2 and 3, the contrast is very high (of the order of 50%) and the shape of the ring is perfectly clear. The dimensions of the ring can also be measured with a fairly good precision. These two rows represent depths of around 1.15 cm a nd 1.50 cm, which correspond to the location of the air ring. The contrasts observed in these six images are consistent with what was observed in Figure 7 17.

PAGE 79

79 A B Figure 7 1. Geometry of the MCNP simulation of a Collimated Segmented Detector Scatter X ray Image of a lead strip on a n ylon background: A) view of the YZ plane and B) view of the XY -plane A B Figure 7 2. Possible paths of backscattered X -ray photons from a fan beam source toward the segmented d etector : A) without collimation and B) with a collimation grid

PAGE 80

80 Figure 7 3. Comparison between the normalized fluxes observed in the collimated array detector with a fan beam source and with a pencil beam source for a lead strip on the surface of a nylon block Figure 7 4 Possible paths of X ray photons eventually reach ing an image pixel. The path of particle 1 is only valid with a fan beam source because it starts outside the pencil beam illumination field.

PAGE 81

81 Figure 7 5. Measure of the resolution for Collimated Segmented Detector Scatter X ray Imaging Figure 7 6 Backscatter X ray secondary source as se en from point y = t

PAGE 82

82 Figure 7 7 Normalized contribution to a 0.6 mm wide pixel, with a 1 cm collimation. A B Figure 7 8. Geometry of the MCNP simulation of a Collimated Segmented Detector Scatter X ray Image of a lead strip 2 cm deep inside a n ylon block: A) view of the XZ -plane and B) view of the YZ -plane 0.00 0.20 0.40 0.60 0.80 1.00 1.20 0.25 0.2 0.15 0.1 0.05 6E 16 0.05 0.1 0.15 0.2 0.25 Normalized contribution Distance on the y axis in cm Normalized contribution to the pixel

PAGE 83

83 Figure 7 9. Comparison between the normalized fluxes observed in the collimated array detector with a fan beam source and with a pencil beam source for a lead strip 2 cm deep in a nylon block Figure 7 10. Comparison between the normalized fluxes observed in the collimated array detector with a fan beam source for four different collimation lengths for a lead strip 2 cm deep in a nylon block

PAGE 84

84 A B Figure 7 11. Geometry of the MCNP simulation of a Collimated Segmented Detector Scatter X ray Image of a 1 cm deep air gap inside an aluminum block with a 2D grid collimated detector at a 45 inclination : A) view of the XZ -plane and B) view of a plane parallel to the detector A B C Figure 7 12. Geometries of the 3 MCNP simulations of an air gap inside an aluminum block, views of the XZ plane : A) linearly collimated segmented detector, B) linearly collimated segmented detector at a 45 inclination and C) 2D collimated segmented detector at a 45 inclination

PAGE 85

85 Figure 7 13. Comparison between the normalized fluxes from MCNP simulation s of a 1 cm deep air g ap inside an aluminum block with a fan beam source Figure 7 14. Geometry of the 2D collimated array detector at a 45 inclination, view of the XZ plane

PAGE 86

86 Figure 7 15. Backscatter X ray secondary source as se en from point t

PAGE 87

87 Figure 7 16. Normalized contribution of each z -level in the target object to all 6 rows of pixels and to the sum of the rows Figure 7 1 7 Comparison between the fluxes recorded by rows 1, 2, 3, 4, 5 and 6 of the 2D collimated array detector, with a 45 inclination from a n MCNP simulation of CSD SXI of an air gap inside an aluminum block

PAGE 88

88 A B Figure 7 18. Geometry of the ring of air located 1 cm deep inside aluminum: A) view of the XZ plane and B) view of the XY -plane A B C Figure 7 19. Geometries of the three different imaging systems: A) NaI detector with pencil beam source, B) linearly collimated array detector with fan beam source and C) 2D collimated array detector at a 45 inclination with fan beam source

PAGE 89

89 A B C D Figure 7 20. MCNP generated X ray backscatter images of a ring of air inside aluminum using a NaI detector with a pencil beam source with the following collimation lengths: A) 0.1 cm, B) 0.5 cm, C) 1.0 cm and D) 1.5 cm A B C Figure 7 21. MCNP generated X ray backscatter images of a ring of air inside aluminum using a linearly collimated segmented detector with a fan beam source with the following collimation lengths: A) 0. 5 cm, B) 1.0 cm and C) 1. 5 cm

PAGE 90

90 A B C D E F Figure 7 22. MCNP generated X ray backscatter images of a ring of air inside aluminum using a 2D collimated segmented detector at a 45 inclination with a fan beam source. Each image correspond to one of the six rows of the pixe lated detector A) row 1, B) row 2 C) row 3, D) row 4, E) row 5 and F) row 6.

PAGE 91

91 Figure 7 2 3 Z -levels corresponding to each of the six MCNP generated images of the ring of air

PAGE 92

92 CHAPTER 8 CSD SXI: CONSTRUCTION AND TEST OF A FIRST PROTOTYPE Experimental Setup Segmented Detector The detector used for the prototype is a 12 inch long linearly segmented CMOS detector built by E nvision Product Design It is shown in Figure 8 1. It is composed of three thousand, eight hundred and ten 80 micron wide pixels .11 X rays are allowed into the detector through a 2 mm wide and 1 cm deep collimated slot in the tungsten h ousing and strike the scintillator. The visible light produced is then guided via fiber optics to a CMOS active -pixel sensor array (Figure 8 2 ). The slot in the t ungsten provides depth collimation. The detector is mounted on the X ray tube stand, 4 cm away and 9 cm below the X ray tube window. It is given a 40 degree angle toward the X ray source because the collimation slot is so tight that no photon would be detected otherwise (F igure 8 3). The segmented detector is conne cted to a workstation t hrough a PCI card. S oftware displays in real time the intensity recorded by each pixel. It is possible to group up to 16 pixels together, which is very useful because the pixel bins must be of the same size as the collimation grid pitch. The integration time can vary from 6 to 4000 ms, and while times as short as 20 ms are commonly used for direct transmission imaging, integration time s bet ween 500 and 2000 ms are needed for X ray backscatter imaging purposes. A function allows recording a grey level image for a given scanning speed. The RSD motion control software was used to move the X -ray source and the detector for the scan. Before the acquisition of an image, calibration of the detector must be performed each time the target background and the X ray tube voltage are significantly modified. This is done automatically by the software and takes between two and

PAGE 93

93 five minutes. A successful cal ibration should result in a straight line across the whole detector as shown in Figure 8 4 Collimation Grid The collimation grid for this CSD SXI prototype is made of about eighty 0.4 mm (or 1/64) thi ck lead plates and twice as many spacers glued togethe r in an alternating pattern resulting in a 7 cm long, 4 cm wide and 2 cm thick grid with a 0.88 mm pitch (Figure 8 5 ). Although the grid built with this method is not exactly evenly spaced, it was much less expensive than a specially machined tungsten gr id would have been. Moreover, thanks to the calibration of the detector, irregularities in the grid, such as slightly bent lead plates, have no effect on CSD SXI images. This collimation grid is mounted on the bottom part of the detector (Figure 8 6) Obvi ously, the lead collimation grid is critical to the quality of images. Figure 8 7 shows the difference between the image obtained with the segmented detector when it is collimated by the grid and when it is not. Shapes and details are only visible when the detector is collimated, while the uncollimated detector yields images with very limited contrast information and in only one dimension. It was previously stated that pixel bins must be of the same size as the grid pitch. In fact, a first lead collimation grid was built with 1.06 mm (or 1/24) thick lead. The pitch of this grid, which is shown in Figure 8 8 was 2.18 mm. However, because the software only allows grouping up to 16 pixels, the maximum pixel bin size was 1.41 mm. This resulted in artifacts on CSD SXI images ; because more than one group of pixels or super -pixel are collimated by the same lead plates, they do not have the same perspective of the target object (Figure 8 9 ). Consequently, the dif ferent groups of pixels do not d etect the same flux of X ray s during the transition between the background and a feature with a different scattering to absorption ratio,

PAGE 94

94 and this leads to oscillation in image contrast as shown in Figure 8 10. MCNP simulati ons have confirmed this behavior. X -ray fan Beam Source Another important point is to have a sufficiently narrow fan beam X ray source. In all previous MCNP simulations, it was assumed that the width of the beam was only 1 mm. In practice, a machined lead part with a 1 mm slit aperture that was designed to provide a fan beam source for CIBR (Figure 8 11A ) was positioned at the X -ray tube exit window However, the actual width of the fan beam with this lead shield is about 1 cm a s shown in Figure 8 11B. MCNP simulations have shown that if the fan beam width is larger than a few millimeters, the image quality is greatly reduced. This is why another lead part was design to allow changing the aperture (Figure 8 12A). This shielding provides a narrower fan beam illumination, but on the other hand, the flux of X -ray photons received by the detector can be too low which can be a problem for both calibration and acquisition time (Figure 8 12B). Both Figure 8 11B and Figure 8 12B were obta ined using the segmented detector pointed toward the moving X ray source. Test of the First CSD -SXI Prototype Contrast and Details The various objects shown in Figure 8 13A, were imaged using CSD -SXI and RSD. Figure 8 13B shows the CSD -SXI image of these objects on an aluminum background, taken at 120 kVp and 8.75 mA with an acquisition time of 3 minutes and 42 seconds (1000 ms integration time per pixel line) Figure 8 13C shows the RSD image of the same objects on the aluminum background, obtained with the same X ray tube setup in 16 minutes 21 second. The first observation that can be made is that both images offer a similar level of detail, even though, mos t features are clearer on the RSD image. The plastic screwdriver handle for instance is easily visible on the RSD image while it can barely be detected on the CSD -SXI image.

PAGE 95

95 The ratio of contrast between the steel combination wrench and the aluminum backg round is about 0.4 on both the RSD and CSD SXI images. However, the contrast ratio between the highly scattering plastic screwdriver handle and the background is 1.11 on the RSD image while it is indeterminate on the CSD SXI image. On other CSD -SXI image s of the same objects this plastic handle can be easily detected; it is a question of detector position and angle The statistical error for the background is close to 4% for the CSD SXI image, and it is as low as 0.7% for the RSD image. Figure 8 14 shows l etters of lead on a nylon background which were also imaged using CSD SXI and RSD. Both images were obtained at 140 kVp and 4 mA, with acquisition time of 2 minutes and 46 seconds for CSD SXI (1000 ms integration time per pixel line) and 8 minutes for RSD. It is clearer in this example that the quality of RSD images is better. Not only is the contrast better (lead to nylon ratio of 0.04 for RSD versus 0.12 for CSD SXI), but the shape of the target object is much more accurate i n the RSD image, and the gaps between the lead tiles composing the letters can even be detected although they are of the order of a few tenth of a millimeter. Resolution and Modulation Transfer Function (MTF) for CSD -SXI An attempt was made to quantify the resolution associated with CS D -SXI images. A test pattern composed of twelve 1 cm wide and 0.4 mm thick aligned lead tiles was scanned by CSD SXI and RSD (Figure 8 15A). Because the resolution for CSD -SXI is not necessarily the same in both directions, parallel (Figure 8 15B) and perpendicular (Figure 8 15C) to the segmented detector), two images were taken for this system against only one for RSD (Figure 8 15D). Figure 8 16 shows the MTF of the two CSD SXI images and of the RSD image. While features as small as one millime ter can still be detected by RSD systems, this CSD -SXI prototype yields images with a better resolution in the direction parallel to the linearly segmented

PAGE 96

96 detector than in the perpendicular direction. This is due to the fact that resolution in the paralle l direction is controlled by the lead collimation grid with a 0.88 mm pitch whereas the resolution in the perpendicular direction is a function of the 2 mm wide tungsten collimation slot. However, increasing the collimation in the direction perpendicular t o the segmented detector by decreas ing the width of the collimation slot, for instance would result in a lower count rate. Depth Penetration An important feature of X ray backscatter imaging systems is the ability to detect features located behind another object or deep below the surface of an object. Figure 8 17A shows various objects such as a screwdriver, a wrench, steel and nylon washers, and screws on an aluminum background. Figures 8 17B and 8 17C show the CSD -SXI and RSD images of these objects witho ut any cover at 140 kVp and 21.4 mA and at 120 kVp and 8.0 mA, respectively with a 1000 ms integration time per pixel line for CSD -SXI and 50 ms per pixel for RSD The contrast of both scattering (screwdrivers plastic handle, nylon washer) and absorbing (steel wrench, screws) materials relative to aluminum is equivalent on both images. However, even if all objects can be detected in both cases, the RSD image is clearer than the CSD SXI; shapes are more precisely represented and even the objects shadows a ppear very clearly on the RSD image. Figure 8 18 show s the RSD and CSD -SXI images of the same objects through 3.2 and 6.4 mm thick aluminum plates. The objects appear slightly more distinctly and with more contrast on the RSD images, but CSD -SXI images are also of fairly good quality. The plastic ruler on the other hand can hardly be detected in both CSD -SXI images. Figure 8 19 also shows the RSD and CSD SXI images of the same objects on an aluminum background but this time through 2 mm and 6.4 mm thick car bon -carbon composite. The RSD and CSD SXI image quality are roughly equivalent even though RSD offers images with much clearer contours than CSD SXI. All the RSD images shown in Figures 8 17, 818 and 8 19 were all acquired in 10 minutes and 47

PAGE 97

97 seconds aga inst only 3 minutes and 42 seconds for the CSD -SXI images. It should also be noted that the lead collimation grid used for CSD SXI was only 7 cm long. Increasing this length would result in a longer active area for the segmented detector and therefore, CSD -SXI image acquisition time would be reduced.

PAGE 98

98 Figure 8 1 Photograph of the 12 Envision Product Design segmented detector Figure 8 2 Simplified diagram of the segmented detector used for CSD -SXI

PAGE 99

99 Figure 8 3 Segmented detector mounted on the X ray tube at a 40 angle A B Figure 8 4 Flux detected by the linearly segmented detector with and without the X -rays on: A) before calibration and B) after calibration

PAGE 100

100 Figure 8 5 Collimation grid made of 0.4 mm thick lead Figure 8 6 Lead collimation grid mounted on the bottom surface of the segmented detector

PAGE 101

101 A B Figure 8 7 CSD -SXI image of letters of lead (1 mm thick) on nylon: A) with the lead collimation grid and B) without the lead collimation gr id Figure 8 8 Lead collimation grids made of 0.4 mm and 1.08 mm thick lead plates and spacers.

PAGE 102

102 Figure 8 9 Explanation of the apparition of artifacts in CSD -SXI images when the collimation grid pitch does not match the size of pixel bins. A B Figure 8 10. CSD -SXI images of a disk of lead on nylon: A) wit h a 2.18 m m collimation grid pitch and 0.4 m m pixel bins and B) with a 0.88 m m collimation grid pitch and 0.88 m m pixels bins

PAGE 103

103 A B Figure 8 1 1 Lead shield with a 1 mm slit: A) photograph and B) Resulting X ray fan beam at a 20 cm distance imaged with the segmented detector A B Figure 8 12. Lead shield with variable aperture, in this case 0.2 mm: A) photograph and B) Resulting X ray fan be am at a 20 cm distance imaged with the segmented detector

PAGE 104

104 A B C Figure 8 13. Various objects on an aluminum background: A) photograph, B) CSD -SXI image at 120 kVp and 8.75 mA and C) RSD image at 120 kVp and 8.75 mA A B C Figure 8 14. Letters of lead on a nylon background: A) photograph, B) CSD -SXI image at 140 kVp and 4 mA and C) RSD image at 140 kVp and 4 mA

PAGE 105

105 A B C D Figure 8 1 5 Lead test pattern on an aluminum background: A) photograph, B) CSD -SXI image at 1 20 kVp and 8.80 mA with the pattern parallel to the segmented detector, C) CSD SXI image at 120 kVp and 8.80 mA with the pattern perpendicular to the segmented detector, and D) RSD image at 120 kVp and 8.80 mA

PAGE 106

106 Figure 8 16. Modulation Transfer Functions (MTF) for the CSD -SXI and RSD images shown in Figure 8 15. A B C Figure 8 1 7 Various objects on an aluminum background: A) photograph, B) CSD -SXI image at 140 kVp and 21.4 mA, and C) RSD image at 120 kVp and 8.0 mA

PAGE 107

107 A B C D Figure 8 1 8 Various objects on an aluminum background: A) CSD -SXI image through a 3.2 mm thick aluminum cover at 140 kVp and 25.0 mA B) RSD image through a 3.2 mm thick aluminum cover at 1 20 kVp and 8.0 mA C) CSD SXI image through a 6.4 mm thick aluminum cover at 1 60 kVp and 18.75 mA and D) RSD image through a 6.4 mm thick aluminum cover at 120 kVp and 8.0 mA A B C D Figure 8 1 9 Various objects on an aluminum background: A) CSD -SXI image through a 2 mm thick carbon-carbon composite (C/C) cover at 140 kVp a nd 21.5 mA B) RSD image through a 2 mm thick C/C cover at 120 kVp and 8.0 mA, C) CSD SXI image through a 6.4 mm thick C/C cover at 1 40 kVp and 21.5 mA and D) RSD image through a 6.4 mm thick C/C cover at 120 kVp and 8.0 mA

PAGE 108

108 CHAPTER 8 CSD SXI: CONCLUSIONS AND FUTURE WORK The association of collimated array detectors employing a lead or tungsten collimator grid with a fan beam source offer s several advantages for X ray backscatter imaging. The image acquisition time can be reduced relative to pencil beam systems such as RSD due to the use of the fan beam source and a simpler scanning pattern. Experimental test s on the pr ototype system have confirmed that CSD SXI can yield images faster than RSD with an almost equivalent quality. Increasing the length of the collimator grid, and therefore the segmented active length, would also decrease image acquisition time. However, the images obtained with RSD offer more distinct details of target objects than those taken with CSD SXI Moreover, the resolution of images obtained with this first CSD -SXI prototype in the direction perpendicular to the segmented detector is quite low compared to the resolution obtained with RSD. This could be solved by decreasing the width of the segmente d detector collimation slot but this would come at the cost of a smaller detection rate. More work is needed to improve these factors, but given the relative simplicity of this first prototype significant progress can be expected. MNCP simulations have s hown that using a 2D collimated segmented detector could lead to accurate 3D backscatter X ray imaging. This could be experimentally done by fitting a laser cut tungsten collimation grid onto the bottom part of a 2D segmented detector. However, the fan bea m source needs to be improved by building a lead shield with a finer slit than what is currently used because MCNP simulations have shown that too large a fan beam source adversely impact s the quality of 3D CSD -SXI images. Ot her, more immediate future work include s writing a LabView program to provide a more adapted calibration for the segmented detector and to control the motion motors for easier image

PAGE 109

109 acquisition. A small motor could also be place d on the segmented detector itself to control its angle and therefore the corresponding image depth.

PAGE 110

110 LIST OF REFERENCES 1 A. JACOBS, E. D UGAN, D S HEDLOCK, University of Florida Research Foundation, Inc Radiography by Selective Detection of Scatter Field Velocity Components, Patent No.: US 7,224,772 B2 (May 29,2007). 2 D SHEDLOCK, B A DDICOTT E DUGAN, and A JACOBS Optimization of a RSD X Ray Backscatter System for Detecting Defects in the Space Shuttle External Tank Thermal Foam Insulation University of Florida (2005) 3 D. SHEDLOCK X Ray backscatter imaging for radiography by selective detection and snapshot: evolution, development, and optimization, Ph.D. Dissertation University of Florida (2007). 4 R. E VANS, The Atomic Nuc leus" p 683, Krieger Publishing Company (June 1982) 5 Techni cal data, Kodak INDUSTREX Digital Imaging Plates, KODAK Publication No. TI 2632 (September 2006). http://www.shawinspectionsystems.com/products/kodak/datasheets/ EN_ti2632.pdf 6 Film Based Portable X -ray Systems, Source -Ray, Inc (2005). http://sourceray.com/brochure/SR115%20 SR130%20Brochure%20web.pdf 7 M. BERGER, J. HUBBELL, S. SELTZER, J. CHANG, J. COURSEY, R. SUKUMAR, and D. ZUCKER, XCOM: Photon Cross Sections Database, NIST Standard Reference Database 8 (February 2009). http://physics.nist.gov/PhysRefData/Xcom/Text/XCOM.html 8 C. MENG, Computed Image Backscatter Radiography : p roof of principle and initial development, Master Thesis, University of Florida (2008) 9 X 5 Monte Carlo Team MCNP A General Monte Carlo N -Particle Transport Code, Volume II: Users Guide LA CP 03 0245, Los Alamos National Laboratory (April 2003) 10. G. KNOLL, Radiation Detection and Measurement, p16, Third Edition, John Wiley & Sons, Inc (2000) 11. CMOS Segmented Array Specifications, Envision Product Design (April 2003) http://cmosxray.com

PAGE 111

111 BIOGRAPHICAL SKETCH Olivier Bougeant was born in Brittany, in northwestern France. After obtaining his high school diploma, he studied for two years in the Classes Preparatoires aux Grandes Ec oles in Rennes. He was then admitted to the Ecole Nationale Sup e rieure de Physique de Grenoble (ENSPG), where he obtained a b achelors degree in p hysics in 2007 and an Engineers degree in n uclear e ngineering in 2008. In 2007, he started a Master of Sc ience in n uclear e ngineering at the University of Florida where he joined the Scatter X ray Imaging group.