Emission tomography, where the detection of radiation emitted by a radioactive tracer administered to a patient allows the estimation of the tracer's distribution inside the body, is becoming increasingly important in medicine, in both diagnosis and treatment. This increased interest has led to a demand for higher imaging quality, accuracy and speed. The development of new medical-imaging devices, image-reconstruction algorithms and correction techniques, and the optimization of acquisition protocols all depend on appropriate simulations, in particular using the Monte Carlo techniques familiar in particle physics.
The main techniques for emission tomography are PET (positron emission tomography) and SPECT (single photon emission computed tomography). Both involve surrounding the subject to some extent with an array of suitable detectors (e.g. scintillator crystals) and detecting the radiation emitted from the tracer. At least a dozen Monte Carlo simulation packages can be used for either technique, with different advantages and disadvantages. Accurate and versatile general-purpose simulation packages such as Geant3 from CERN, EGS4 from SLAC, MCNP from the Los Alamos National Laboratory and most recently Geant4 include well validated physics models, geometry-modelling tools and efficient visualization utilities, but they require a major effort to be tailored to PET and SPECT.
On the other hand, dedicated Monte Carlo codes developed for PET and SPECT suffer from a variety of drawbacks. For example, SimSET, developed at the University of Washington, is one of the most powerful dedicated codes for PET and SPECT simulations, modelling physics phenomena and basic detector designs precisely and efficiently. However, it has limitations with respect to the range of detector geometries that can be simulated: for instance, a detector ring cannot be subdivided into individual crystals. In addition, neither SimSET nor other dedicated codes account explicitly for time, which limits their use for modelling time-dependent processes such as the movement of tracers.
Clearly, there is a need for a Monte Carlo tool that readily accommodates complex scanner geometries, while retaining the comprehensive physics-modelling abilities of the general-purpose codes. To meet this demand an international collaboration of physicists at centres in different countries has developed the simulation toolkit GATE - the Geant4 Application for Tomographic Emission.
Opening the GATE
The origin of GATE can be traced back to a workshop held in July 2001 in Paris, which focused on the future of Monte Carlo simulations in nuclear medicine. The various drawbacks of the existing dedicated and general-purpose codes were discussed, and it became clear that it would be to everyone's advantage to develop a simulation toolkit that would combine the best of both worlds: namely, to have a dedicated Monte Carlo platform for emission tomography that could on one hand model decay kinetics, dead time and movement, and on the other hand benefit from the versatility and support of general-purpose simulation tools. Moreover, object-oriented technology appeared to be the best choice to ensure high modularity and reusability for the simulation tools developed specifically for PET and SPECT.
The consensus was therefore to select the simulation toolkit developed in C++ by the Geant4 collaboration and to foster long-term support and maintenance by sharing this development among many research groups. This effort was launched by the PET instrumentation group in the Laboratory for High Energy Physics at the Ecole Polytechnique Fédérale de Lausanne (EPFL), in the first instance as an aid for the design of the ClearPET prototype developed by the Crystal Clear Collaboration at CERN (figure 1).
Specifications of the Geant4-based simulation framework were circulated in December 2001. The C++ coding began at the Lausanne PET instrumentation group with the help of the Geant4 low-energy electromagnetic physics working group, the corpuscular physics for life science group at Laboratoire de Physique Corpusculaire in Clermont-Ferrand and the medical image and signal-processing group in the Electronics and Information Systems Department at the University of Ghent. The development strategy was defined at a second workshop organized in January 2002 in Lausanne, and on 23 May 2002, at a meeting again held in Lausanne, a live demonstration of the first version of the GATE platform was given. At this meeting the research groups at Lausanne, Clermont-Ferrand and Ghent decided to start the OpenGATE collaboration in order to improve, validate, document and test GATE with a view to preparing a public release of the software.
Since then, the collaboration has grown, currently comprising 21 laboratories in nine countries in Europe, the US and Asia, and in May 2004 the first public release of GATE was made available. At present, GATE has been downloaded and is run by more than 200 individuals in academic institutes and commercial companies around the world.
How does it work?
GATE incorporates the Geant4 libraries in a modular, versatile and scripted simulation toolkit adapted to nuclear medicine. In addition, it allows the accurate description of time-dependent phenomena such as source and detector movement and source decay kinetics. The elements of the geometry of the system can be set into movement via scripting, and all the movements of these elements are kept synchronized with the evolution of the activity of the radioactive source. For this purpose, the data acquisition is subdivided into a number of time steps during which the elements of the geometry are considered to be at rest.
Radioactive decay times are generated within these time steps so that the number of events decreases exponentially from one time step to the next according to the decay profile of the particular radioisotope. The ability to synchronize all time-dependent components in this way is one of the most innovative features of GATE, making it possible to perform realistic simulations of data acquisition in time.
GATE uses approximately 200 C++ classes from the Geant4 simulation toolkit and an application layer allows the implementation of user classes derived from the core layer classes. Provided the application layer contains appropriate features, the use of GATE does not require any C++ programming, thanks to a dedicated scripting mechanism that extends the native command interpreter of Geant4.
The GATE source code resides in a CVS (Concurrent Versions System) repository maintained by the OpenGATE collaboration. There is documentation to support the simulation toolkit, including installation and user guides, online source-code documentation (via the doxygen documentation tool) and a list of frequently asked questions (FAQs). In addition, two benchmarks have been developed for PET and SPECT to check the installation of GATE and as a tutorial for users. The SPECT benchmark, for example, simulates a gamma camera based on four detector heads, with a cylinder of water as specimen, containing the gamma-emitter technetium-99m (figure 2).
GATE has been validated by comparing its simulated data to real data obtained from commercial systems currently in use or under consideration. Factors such as spatial resolution and energy resolution generally agree to within a few per cent or better. Overall these studies illustrate the flexibility and reliability of GATE for the accurate modelling of different detector designs.
The price of GATE's versatility, compared with simpler codes such as SimSET, comes in terms of longer computation times, but already there are efforts to improve this. One approach is the "gridification" of GATE - subdividing the simulations to run on processors geographically distributed. Following successful tests on the European DataGrid testbed, GATE is now deployed on EGEE grid production infrastructure, where computing time can be reduced by a factor of 16 when a simulation is run on 20 processors as opposed to one. However, the study also showed that computing time is not directly inversely proportional to the number of jobs running in parallel.
In general, the future of GATE is closely linked to the developing role of Monte Carlo simulations in nuclear medicine, where they are becoming increasingly important. GATE also has the potential to be useful beyond PET and SPECT, for example in in-line tomography in hadron therapy and dose calculations in radiotherapy. By providing free access to the GATE source code, the OpenGATE collaboration hopes that GATE will continue to evolve to become a comprehensive simulation tool at the service of the nuclear medicine community.
Further reading
S Jan et al. 2004 Phys. Med. Biol. 49 4543.
See the GATE website located at www-lphe.epfl.ch/GATE/.
• Member institutes of the OpenGATE collaboration are: Ecole Polytechnique Fédérale de Lausanne (LPHE); University of Clermont-Ferrand (LPC); University of Ghent (ELIS); CHU Pitié-Salpêtrière (U494 INSERM), Paris; Vrije Universiteit Brussel (IIHE); Centre d'Exploration et de Recherche Médicales par Emission de Positons (CERMEP), Lyon; Service Hospitalier Frédéric Joliot (SHFJ), CEA-Orsay; CHU Nantes (U601 INSERM); Sungkyunkwan University School of Medicine (Division of Nuclear Medicine), Seoul; University Claude Bernard (IPNL), Lyon; University Louis Pasteur (IRES), Strasbourg; University Joseph Fourier (LPSC), Grenoble; Forschungszentrum-Juelich (IME); University of Massachusetts Medical School (Division of Nuclear Medicine); CHU Morvan (LATIM, U650 INSERM), Brest; University of California (Crump Institute for Molecular Imaging); University of Toronto (CAMH); DAPNIA, CEA-Saclay; Memorial Sloan-Kettering Cancer Center (Department of Medical Physics), New York; University of Athens (IASA); Delft University of Technology (IRI).
Author:
Christian Morel, spokesman of the OpenGATE collaboration, Laboratory for High Energy Physics, Ecole Polytechnique Fédérale de Lausanne.