Implementation of Variance Reduction Techniques for Improvement in Computation Efficiency of GATE Monte Carlo Simulation Package on a Distributed Computing Platform- fastGATE

Partners: School of Electrical and Computer Engineering - National Technical University of Athens, Crump Institute for Molecular Imaging, David Geffen School of Medicine at UCLA., Hardware and Software Engineering Ltd

Duration: 2005 - 2008

GATE is a novel openSource toolkit for simulation of full-edged clinical PET and SPECT systems. It allows the user to model realistic nuclear medicine experiments with an easy to use, yet powerful, scripting language. GATE simulations are CPU-bound computations, in the sense that most of the work corresponds to numerical calculations related to particle transport. Thus, realistic imaging simulations, including transmission tomography, usually take weeks or months to complete in state-of-the-art single-CPU computers. However, it is also generally true that Monte Carlo simulations are excellent candidates for embarrassingly parallel solutions to the computational problem, because the amount of interprocess communication is small, and usually only at process start-up and termination. In addition, variance reduction techniques provide execution speed up since a number of physical processes such as photon splitting, electron history repetition etc can improve the efficiency of the simulated code. It is possible to further improve the efficiency by optimizing transport parameters such as electron energy cut-off, maximum electron energy step size, photon energy cut-off, a cut-off for kerma approximation and use of delta scattering technique, without loss of calculation accuracy. The aim of the proposed project is the implementation and application of variance reduction techniques (VRTs) in order to speedup the execution time of the GATE Monte Carlo simulation package on distributed computing platforms and the development of a cluster in BIOSIM.