Improved Monte-Carlo Simulations for Dynamic PET



M. A. Shilov, Eric C. Frey, W. Paul Segars, Jingyan Xu and Benjamin M. W.Tsui



The objective of this research is to improve the speed and implementation convenience of computer simulations of positron emission tomography (PET) by combining the SimSET (Simulation System for Emission Tomography) and the GATE (Geant4 Application for Tomography Emission) Monte-Carlo codes. The improvement is particularly important for simulating dynamic FDG PET studies.



The new technique for computer simulations of PET involves combination of two powerful and well validated Monte-Carlo codes, SimSET and GATE. It takes advantage of the shorter simulation times for photon propagation inside a voxelized phantom using SimSET as compared to GATE. The histories of all photons, single and coincidence, escaping from the phantom are stored in a file which serves as the input to GATE. The convenience and precision of GATE interface to define and visualize complex scanner geometries is explored to simulate experimental data from specific PET systems. We used the design parameters and the geometry of the scintillators and the detector circuitry of a GE Advance PET scanner for generation of the preliminary data. The new technique was validated by comparing the simulated projection data of a simple voxelized source inside a uniform water phantom generated using the GATE code alone and using the new method with a combination of the SimSET and GATE codes. The two simulated projection data sets and the total computer times used to generate them were compared. To evaluate the new Monte-Carlo simulation method for realistic clinical applications, we simulated positron emission, annihilation and photon transport inside a voxelized 3D NCAT phantom with activity distribution modeling that of a typical F-18 FDG patient study.



We found the simulated coincidence projection data of the simple voxelized source inside a uniform water phantom obtained by using GATE alone and the new method with the combined Monte-Carlo codes are identical asides from statistical fluctuations. The total simulation times using the new technique on AMD Athlon® processors of our computer cluster are about 12 times shorter than that using GATE alone and are only about twice as long as the time required for the photon history generation with SimSET.  The time required to generate 1 minute of coincidence projection data from the 3D NCAT phantom with simulated 10 mCi of injected F-18 FDG using the new method is about 1.5 hours on a cluster of  80 AMD Athlon® MP 2200 CPUs.



Our preliminary results indicate that the new technique can be used for efficient Monte-Carlo simulations of a realistic voxelized phantom using an accurate model of the complex configuration and detector geometry of a modern PET scanner. The new simulation method is particularly useful in studying cardiac and respiratory motions and other dynamic studies using PET imaging techniques and instrumentation.



NIH grant # R01 EB168.