GATE Overview and recent advances Nicolas Karakatsanis Irne

GATE Overview and recent advances Nicolas Karakatsanis Irène Buvat National Technical University of Athens, Greece Laboratory of Functional Imaging, U 678 INSERM, Paris, France

Outline • Evolution of the use of MC simulations in ET since 1996 • Open. GATE motivation and short history • New features in MC simulators in ET • New applications for MC simulations • Upcoming developments in MC simulations • Conclusion

Evolution of the codes used for MC simulations in ET since 1996 -2000 2001 -2005 • 14 different codes: - 10 « home-made » - 4 publicly released or available from authors • 15 different codes: - 8 « home-made » - 7 publicly released or available from authors No « standard » code for Monte Carlo simulations in SPECT and PET Most frequently used And recently Penelope Sim. SET SIMIND GATE

Motivation for developing GATE in 2001 • Provide a public code - based on a standard code to ensure reliability - enabling SPECT and PET simulations (possibly even more) - accommodating almost any detector design (including prototypes) - modeling time-dependent processes - user-friendly • Developed as a collaborative effort

The Open. GATE collaboration From 4 to 23 labs worldwide • Delft University of Technology, Delft, The Netherlands • Ecole Polytechnique Fédérale de Lausanne, Switzerland • Forschungszentrum Juelich, Germany • Ghent University, Belgium • National Technical University of Athens, Greece • Vrije Universiteit Brussel, Belgium • U 601 Inserm, Nantes • U 650 Inserm, Brest • U 678 Inserm, Paris • LPC CNRS, Clermont Ferrand • IRe. S CNRS, Strasbourg • UMR 5515 CNRS, CREATIS, Lyon, • CPPM CNRS, Marseilles • Subatech, CNRS, Nantes • SHFJ CEA, Orsay • DAPNIA CEA, Saclay • Joseph Fourier University, Grenoble • John Hopkins University, Baltimore, USA • Memorial Sloan-Kettering Cancer Center, New York, USA • University of California, Los Angeles, USA • University of Massachusetts Medical School, Worcester, USA • University of Santiago of Chile, Chile • Sungkyunkwan University School of Medicine, Seoul, Korea

Product of Open. GATE: GATE • Publicly released on May 2004 http: //www. opengatecollaboration. org • An official publication: Jan S, Santin G, Strul D, Staelens S, Assié K, Autret D, Avner D, Barbier R, Bardiès M, Bloomfield PM, Brasse D, Breton V, Bruyndonckx P, Buvat I, Chatziioannou AF, Choi Y, Chung YH, Comtat C, Donnarieix D, Ferrer L, Glick SJ, Groiselle CJ, Guez D, Honore PF, Kerhoas-Cavata S, Kirov AS, Kohli V, Koole M, Krieguer M, van der Laan DJ, Lamare F, Largeron G, Lartizien C, Lazaro D, Maas MC, Maigne L, Mayet F, Melot F, Merheb C, Pennacchio E, Perez J, Pietrzyk U, Rannou FR, Rey M, Schaart D, Schmidtlein CR, Simon L, Song TY, Vieira JM, Visvikis D, Van de Walle R, Wiers E, Morel C. GATE: a simulation toolkit for PET and SPECT. Phys Med Biol 49: 4543 -4561, 2004. • More than 800 subscribers to the Gate users mailing list

Tasks of the Open. GATE collaboration • Upgrade GATE for following GEANT 4 new releases (1 major release per year) • Incorporate new developments in GATE (1 minor release per year) e. g. : q variance reduction techniques (to be released soon) q speed-up options (e. g. , analytical modeling of the collimator response in SPECT) (to be released soon) q tools for running GATE on a cluster or on a grid environment (to be released soon) q extension of GATE for dosimetry applications q tools for interfacing GATE output with other software (STIR) • Organize training

GATE today: technical features • Based on GEANT 4 • Written in C++ • User-friendly: simulations can be designed and controlled using macros, without any knowledge in C++ • Appropriate for SPECT and PET simulations • Flexible enough to model almost any detector design, including prototypes • Explicit modeling of time (hence detector motion, patient motion, radioactive decay, dead time, time of flight, tracer kinetics) • Can handle voxelized analytical phantoms

GATE today: practical features • Can be freely downloaded, including the source codes • Can be run on many platforms (Linux, Unix, Mac. Os) • On-line documentation, including FAQ and archives of all questions (and often answers) about GATE that have been asked so far • Help about the use of GATE can be obtained through the gate-user mailing list • Many commercial tomographs and prototypes have already been modeled (including validation of the model)

www. opengatecollaboration. org

PET systems already modeled by the Open. GATE collaboration

Some examples GE Advance/Discovery LS PET scanner Schmidtlein et al, Med Phys 2006 GE Advance/Discovery ST PET, 3 D mode Schmidtlein et al. MSKCC

Some examples HRRT D. Guez et al. , HRRT, CEA/DAPNIA and SHJF Guez et al, DAPNIA and SHJF

SPECT systems already modeled by the Open. GATE collaboration

Example DST Xli camera Phantom holder Schielding Backcompartment Na. I(Tl) crystal MEHR collimator Scanning table Assié et al, Phys Med Biol 2005

Some examples Indium 111 source in water 14 7 12 10 6 Counts (AU) Indium 111 source in air 8 6 4 2 0 50 70 5 4 3 2 1 0 90 110 130 150 170 190 210 230 250 270 50 70 90 110 130 150 170 190 210 230 250 270 Energy (ke. V) Real data Simulated data FWHM (mm) 15 13 11 9 7 5 3 0 5 10 15 20 25 source - collimator distance (cm) Assié et al, Phys Med Biol 2005

Prototypes already modeled by the Open. GATE collaboration

Example crystal PSPMT collimator Energy spectrum Number of counts IASA Cs. I(Tl) gamma camera Experiment GATE … Experiment Energy (ke. V) Lazaro et al, Phys Med Biol 2004

Monte Carlo simulations today: what is new?

Modeling time dependent processes SPECT and PET intrinsically involves time: • Change of tracer distribution over time (tracer biokinetic) • Detector motions during acquisition • Patient motion • Radioactive decay • Dead time of the detector • Time-of-flight PET GEANT 4 (hence GATE) is perfect in that regard

Modeling of radioactive decay 15 O (2 min) 11 C (20 min) Santin et al, IEEE Trans Nucl Sci 2003

Modeling of time of flight PET 3 D ML-EM Timing resolution NO TOF 3 ns TOF 700 ps TOF 500 ps TOF 300 ps Type of study Detector Energy Resolution (FWHM) Low Energy Threshold (ke. V) Coincidence Time Window (ns) Non-TOF GSO 15% 410 8 TOF La. Br 3 6. 7% 470 6 Groiselle et al, IEEE MIC Conf Rec 2004

Modeling realistic phantoms Realistic phantoms can be easily used as GATE input NCAT Segars et al, IEEE TNS 2001 MOBY Segars et al, Mol Imaging Biol 2004 Descourt et al, IEEE MIC Conf Records 2006

Modeling original detector designs GEANT 4 is a very flexible tool Non-conventional geometries TEP/CT BIOGRAPH Siemens detection block lead end-shielding Spherical geometry of the Hi-Rez PET scanner Michel et al, IEEE Conf Records 2006

Modeling original detector designs Small animal imaging TEP/CT BIOGRAPH Siemens Mouse-size gamma camera 2 Hamamatsu H 8500 PSPMT Na. I pixelized scintillator Tungsten collimator Mouse-size PET 1 Hamamatsu H 8500 PSPMT pixelated LYSO scintillator (30 x 35 crystals per block, 1 head = 1 block) Sakellios et al, IEEE MIC Conf Records 2006

Modeling original detector designs Small animal imaging Simulated mouse gamma camera image (MOBY phantom) Simulated mouse PET image (MOBY phantom) Sakellios et al, IEEE MIC Conf Records 2006

New applications for Monte Carlo simulations 1995 -1999 2000 -2004 Design and assessment of correction and reconstruction methods Study of an imaging system response Data production for evaluation purpose Use in the very imaging process Description and validation of a code

Optimizing detector design NEC curves as a function of the crystal in the PET Hi. Rez scanner Michel et al IEEE MIC Conf Records 2006

Using Monte Carlo simulations for calculating the system matrix “Object” f GATE is very appropriate but slow Projection p j i p=Rf R(i, j): probability that a photon (or positron) emitted in voxel j be detected in pixel i Calculating R using Monte Carlo simulations: • for non conventional imaging design (small animal) • to account for fully 3 D and patient-specific phenomena difficult to model analytically (mostly scatter) e. g. , Lazaro et al Phys Med Biol 2005, Rafecas et al IEEE TNS 2004, Rannou et al IEEE MIC Conf Rec 2004

Example: SPECT Tc 99 m phantom Simulated data Real data 21. 2 cm 22 cm El Bitar et al IEEE MIC Conf Records 2006

What next?

Bridging the gap between MC modeling in imaging and dosimetry GATE SIMIND DPM GATE The validity of the physics at low energy will have to be checked Problems in G 4 have been identified, e. g. , multiple scattering and corresponding energy deposit calculation Dewaraja et al, J Nucl Med 2005

Modeling hybrid machines (PET/CT, SPECT/CT, OPET) PET/CT SPECT/CT OPET GATE TOAST Brasse et al, IEEE MIC Conf Rec 2004 Alexandrakis et al, Phys Med Biol 2005 Integrating Monte Carlo modeling tools for: - common coordinate system - common object description - consistent sampling - convenient assessment of multimodality imaging On-going studies regarding the use of GATE for CT simulations

Conclusion • GATE is a very relevant tool for Monte Carlo simulations in ET • Simulations will be more and more present in (nuclear) medical imaging in the future: - as a invaluable guide for designing scanners, imaging protocols and interpreting SPECT and PET scans - in the very imaging process of a patient

Last but not least: next GATE training • 3 days, March 7 -9 th, 2007 in Clermont-Ferrand, France • GATE installation, GATE use through lectures and practical sessions given by GATE experts • Registration will open on December, attendance will be limited Registration from mid-december on http: //www. opengatecollaboration. org

Acknowledgments The Open. GATE collaboration

Thank you!!!

Throughput of the simulations • High throughput needed for efficient data production The major problem with GATE and GEANT 4! • Big “World”: - detectors have a “diameter” greater than 1 m - emitting object (e. g. , patient) is large (50 cm up to 1. 80 m) - emitting object is finely sampled (typically 1 mm x 1 mm cells) - voxelized objects are most often used • Large number of particles to be simulated - low detection efficiency - in SPECT, typically 1 / 10 000 is detected - in PET, 1 / 200 is detected At least 4 approaches can be used to increase throughput of the simulations

Using acceleration methods • Variance reduction techniques such as importance sampling (e. g. in Sim. SET) speedup factors between 2 and 15 • Fictitious cross-section (or delta scattering)

Combining MC with non MC modeling Full MC Collimator Angular Response Function increase in efficiency > 100 Song et al, Phys Med Biol 2005

Parallel execution of the code on a distributed architecture Speed-up factor ~ number of jobs new merger old merger PET De Beenhouwer et al, IEEE MIC Conf Records 2006

Smart sampling Taschereau et al, Med Phys 2006