Validation of the Small Animal Biospace Gamma Imager Model Using - - PowerPoint PPT Presentation

▶
validation of the small animal biospace gamma imager
SMART_READER_LITE
LIVE PREVIEW

Validation of the Small Animal Biospace Gamma Imager Model Using - - PowerPoint PPT Presentation

May 11, 2023 347 likes •629 views

Validation of the Small Animal Biospace Gamma Imager Model Using GATE Monte Carlo Simulations on the Grid Joe Aoun 1,2 , Vincent Breton 2 , Laurent Desbat 1 , Bruno Bzeznik 3 , Mehdi Leabad 4 , Julien Dimastromatteo 5 1 Grenoble Joseph Fourier

slide-1
SLIDE 1

MICCAI Grid Workshop – New York 2008 1

Validation of the Small Animal Biospace Gamma Imager Model Using GATE Monte Carlo Simulations on the Grid

Joe Aoun1,2, Vincent Breton2, Laurent Desbat1, Bruno Bzeznik3, Mehdi Leabad4, Julien Dimastromatteo5

1 Grenoble Joseph Fourier University, TIMC-IMAG 2 Clermont-Ferrand Blaise Pascal University, LPC 3 Grenoble Joseph Fourier University, CIMENT 4 Biospace LAB, Paris 5 Grenoble Joseph Fourier University, INSERM U877

slide-2
SLIDE 2

Introduction

MICCAI Grid Workshop – New York 2008 2

SPECT images have a poor quality Correction of the attenuation and the scattering from images Modelling the detector and all the physical interactions Key: Solution: Monte Carlo Simulations => accurate model Disadvantage: Problem: Idea: Long computing time Solution: Grid computing [Buvat I., 2006] [Breton V., 2003]

slide-3
SLIDE 3

MICCAI Grid Workshop – New York 2008 3

Outline

1) Introduction 2) Tools and Experiments

  • Monte Carlo Simulations toolkit
  • CiGri Grid
  • The small animal gamma camera model
  • Experimental set-up
  • Parallelization of the simulations
  • Validation of the camera model

3) Results 4) Conclusions and Perspectives

slide-4
SLIDE 4

MICCAI Grid Workshop – New York 2008 4

Monte Carlo Simulation toolkit : GATE

G Geant4 A Application for T Tomographic Emission “GATE: a simulation toolkit for PET and SPECT”,

  • S. Jan et al, Phys. Med. Biol., 49 (2004) 4543-4561.

 based on GEANT4 : a standard simulation package for high energy physics  open source and modifiable  coded in C++ : more than 200 classes  easy to use : simulations are defined and controlled by macros and scripts

slide-5
SLIDE 5

MICCAI Grid Workshop – New York 2008 5

CIMENT Grid : CiGri

Exploits the idle resources of the CIMENT clusters of the University of Grenoble

1) Submission of a grid ‘bag of tasks’ application 2) Request of idle resources as a normal user 4) Submission of tasks on the best-effort queue of the batch scheduler. (ex. randomly select 5 tasks from the ‘bag’) 3) Returns the number of idle resources (ex. 5 nodes) 5) Execution of the tasks as low-priority best-effort jobs 6) Local job submission requesting resources (ex. 3 nodes) 7) Low-priority best-effort Jobs stopped (3 jobs killed) 8) Notification of the killing of the tasks 9) Killed tasks re-entered on the ‘bag’ to be re-scheduled for later execution

CiGri User Local User

Best-effort: type of jobs that have minimum priority and are submitted

  • nly if there is an idle resource.

Cluster Batch Scheduler CiGri Server

Use short time jobs

slide-6
SLIDE 6

MICCAI Grid Workshop – New York 2008 6

CiGri infrastructure

  • Resource management system : OAR (http://oar.imag.fr)
  • CiGri software: SQL database interacts with independent modules
  • scheduling jobs
  • submitting jobs
  • cluster synchronizing
  • monitoring jobs
  • collecting results
  • logging errors
  • killing jobs
  • Accessible through a User Interface and monitored through a web portal

(https://ciment.imag.fr/cigri)

slide-7
SLIDE 7

MICCAI Grid Workshop – New York 2008 7

Submission of a job on CiGri

Output_i

  • 1. Macro_i
  • 2. Status_i

User Interface SSH SSH USER Cluster 3 Cluster 2 Cluster 1 JDL Automatically collected

slide-8
SLIDE 8

MICCAI Grid Workshop – New York 2008 8

GATE on CiGri

215 7 555 7 866 11 Nights and Weekends 125 7 430 7 886 11 Day CPUs Clusters CPUs Clusters CPUs Clusters GATE availability (average) GATE availability (max) Total

slide-9
SLIDE 9

MICCAI Grid Workshop – New York 2008 9

Outline

1) Introduction 2) Tools and Experiments

  • Monte Carlo Simulations toolkit
  • CiGri Grid
  • The small animal gamma camera model
  • Experimental set-up
  • Parallelization of the simulations
  • Validation of the camera model

3) Results 4) Conclusions and Perspectives

slide-10
SLIDE 10

MICCAI Grid Workshop – New York 2008 10

The Biospace small animal Ɣ Imager model

lead shielding crystal light guide PSPMT ruler capillary Al protection

Circular field of view D = 10 cm Continuous NaI(Tl) crystal : D = 12 cm & thickness = 4 mm PSPMT = Photomultiplier modelled as a 2 mm glass entrance window and a 11 cm nickel backpart LEHR parallel hole collimator with 35 mm thickness

collimator holes 1.3 mm

Septum thickness = 0.2 mm

slide-11
SLIDE 11

MICCAI Grid Workshop – New York 2008 11

Experimental set-up: Source in the center of the Field Of View

10.5 cm Source 99mTc 11.5 cm 4.6 cm

PSPMT

Crystal NaI(Tl)

Collimator 16.5 cm 10 cm 7 cm 2 cm Source 99mTc

PSPMT

Crystal NaI(Tl)

The radioactive background measured first and subtracted from all the other measurements Source placed at different distances from the camera in the air and above a beaker filled with water

slide-12
SLIDE 12

MICCAI Grid Workshop – New York 2008 12

Experimental set-up: Source 2 cm off-centered and image of a 4 capillaries phantom

Capillary

C1 C2 C3 C4

2,1 mm 1,9 mm

Y X

3,15 cm

Concentrations :

  • C1 = 611 µCi
  • C2 = 220 µCi
  • C3 = 129 µCi
  • C4 = 81 µCi

Y X

2 cm Source centered Source off- centered Field Of View

  • f the camera

Previous measurements were repeated with the source 2 cm off-centered

slide-13
SLIDE 13

MICCAI Grid Workshop – New York 2008 13

Parallelization of the simulations

Optimization of the camera model: ~ 200 different models were tested

1 configuration  1 big simulation  1 billion emitted events  30 billions random numbers

1 million 1 million 1 000 000 000 1 million 1 million 1 million 1 million 1 million 1 million 1 2 3 … … … … 1 000

1 small simulation  1 million emitted events  30 millions random numbers  10 minutes

Local CPU: Pentium IV, 3.2 GHz, 1 Go RAM

The Random number streams should be independent

[Reuillon R., 2008]

slide-14
SLIDE 14

MICCAI Grid Workshop – New York 2008 14

Output files of the simulations

  • 1. Retrieved from CiGri
  • 2. Merged into one file on a local CPU
  • 3. Analyzed with the ROOT object oriented data analysis framework

(http://root.cern.ch/)

slide-15
SLIDE 15

MICCAI Grid Workshop – New York 2008 15

Validation of the Ɣ Imager model

Comparison of 4 parameters measured experimentally with the corresponding simulated data Energy spectra: events recorded in the whole FOV (40 – 186 keV) Spatial Resolution: events recorded in the photopic window 126 – 154 keV Sensitivity: Nb of detected events / Nb of emitted events Image of a capillary phantom: a visual comparison of an inhomogeneous phantom

FWHM FWHM FWHM FWHM FWHM

Features

  • f a

gamma camera Features

  • f a

gamma camera

6 pixels 6 pixels 6 pixels Y X

slide-16
SLIDE 16

MICCAI Grid Workshop – New York 2008 16

Outline

1) Introduction 2) Tools and Experiments 3) Results

  • Energy Spectra
  • Sensitivity
  • Spatial Resolution
  • Image of a capillary phantom
  • CiGri performance

4) Conclusions and Perspectives

slide-17
SLIDE 17

MICCAI Grid Workshop – New York 2008 17

Energy Spectra

slide-18
SLIDE 18

MICCAI Grid Workshop – New York 2008 18

Sensitivity in Air

Source in Air

0,0E+00 1,0E-05 2,0E-05 3,0E-05 4,0E-05 5,0E-05 6,0E-05 7,0E-05 2 7 10 16,5 Distance from collimator (cm) Sensitivity (cps/Bq) Experiment 40-186 keV GATE 40-186 keV Experiment 126-154 keV GATE 126-154 keV Experiment 92-125 keV GATE 92-125 keV

Relative difference < 5%

slide-19
SLIDE 19

MICCAI Grid Workshop – New York 2008 19

Sensitivity with the Beaker

Source in Water

0,0E+00 1,0E-05 2,0E-05 3,0E-05 4,0E-05 5,0E-05 6,0E-05 4,62 11,55 Water thickness in the beaker (cm) Sensitivity (cps/Bq) Experiment 40-186 keV GATE 40-186 keV Experiment 126-154 keV GATE 126-154 keV Experiment 92-125 keV GATE 92-125 keV

Relative difference ~ 5%

slide-20
SLIDE 20

MICCAI Grid Workshop – New York 2008 20

Spatial Resolution

2.85 7.61 7.83 0.91 7.65 7.58 16.5 cm (11.55 cm water) 4.69 7.26 7.62 0.91 7.32 7.26 16.5 cm (4.62 cm water) 3.32 7.38 7.63 1.63 7.39 7.27 16.5 cm (0 cm water) 3.49 7.37 7.64 1.30 7.34 7.25 16.5 cm 3.95 5.39 5.61 1.47 5.40 5.32 10 cm 4.49 4.53 4.75 1.89 4.54 4.53 7 cm 10.5 3.20 3.58 3.64 3.22 3.34 2 cm Difference (%) Simulated FWHM (mm) Experimental FWHM (mm) Difference (%) Simulated FWHM (mm) Experimental FWHM (mm) Off-centered Source Centered Source Distance source- collimator (water thickness)

slide-21
SLIDE 21

MICCAI Grid Workshop – New York 2008 21

Image of the capillary Phantom

Experiment GATE 6 pixels profile 6 pixels

Y X

slide-22
SLIDE 22

MICCAI Grid Workshop – New York 2008 22

CiGri performance

13.4

56

25 days 3 h

CiGri – estimated average

10.2

67

21 days 2.5 h

CiGri – nights and weekends

16.9

42

37 days 4 h

CiGri – day

1

1392 days (~ 4 years) 167 h

Local CPU

Pentium IV, 3.2 GHz, 1 Go RAM

Resubmission percentage (%) Gain 200 simulations 1 simulation (1000 jobs)

slide-23
SLIDE 23

MICCAI Grid Workshop – New York 2008 23

Outline

1) Introduction 2) Tools and Experiments 3) Results 4) Conclusions and Perspectives

slide-24
SLIDE 24

MICCAI Grid Workshop – New York 2008 24

Conclusions

GATE was able to accurately model the Biospace Ɣ Imager. Computing grid = indispensable tool in the field of nuclear medical imaging => Deployment of large scale computations and reducing considerably the elapsed time. => Getting more accurate statistical results by increasing the number of tests.

slide-25
SLIDE 25

MICCAI Grid Workshop – New York 2008 25

Perspectives

F Fully 3D 3D M Monte C Carlo reconstruction method [El Bitar Z., 2006].

  • Using the camera model => investigating new algorithms such

as an iterative reconstruction algorithm

  • Fully 3D Monte Carlo => Huge matrices => compression

Increasing the number of CPUs

  • by adding gradually new clusters of the University of Lyon
  • By using the European grid EGEE (http://public.eu-egee.org/)

Improving CiGri performance Current efforts are focused on "check-pointing" to allow execution of longer jobs that can be restored in case of a best-effort kill.

slide-26
SLIDE 26

MICCAI Grid Workshop – New York 2008 26

Thank You for your attention !!