GPU Acceleration On Tetrahedron-based Monte Carlo Method

Optical molecular imaging possesses the ability to characterize and measurebiological processes at the cellular and molecular levels in vivo. It uses molecularmarker techniques such as fluorescence labeling to implement in vivo bioluminescentimaging which is used to non-invasively visualize the physiological and pathologicalprocess of biological tissues in real time. Because the behavior of light propagation inthe medium is various, the study of light propagation in turbid media is significant.Monte Carlo (MC) method is statistic method, which has been widely used to solve theprocess of light propagation in the field of optical molecular imaging. However,traditional MC method is rather time-consuming in turbid media for its statisticalproperties.The main work of this dissertation is focused on how to improve and speed up theMC method. On the one hand, the structural information of heterogeneous medium isdescribed by tetrahedral mesh, with the help of tetrahedron-based model’s specialtopological structure, calculation of intersection point between photon and trajectoryand geometry boundary becomes more faster. On the other hand, based on computeunified device architecture (CUDA) platform, we present a GPU based parallelimplementation for MC simulation of light propagation in complex heterogeneousmedium which is constructed by tetrahedral mesh. The influencing factors and theeffectiveness of the parallel MC simulation are analyzed and validated by the simulationexperiments.