Hybrid Monte Carlo CT Simulation on GPU

Developing image reconstruction algorithms for diagnostic medical devices requires physically accurate and effective simulation tools. In this paper we present a hybrid Monte Carlo (MC) particle simulation method for Computed Tomography (CT) scanners. To meet the performance requirements, we combine several variance reduction techniques and tailor the algorithms for effective GPU execution. Variance reduction methods include main part separation, sample weighting, reuse, forced collision, next event estimation and table driven importance sampling. We show that the resulting method can deliver accurate simulations orders of magnitude faster than direct physical simulation.