Parallel Design For Molecular Optical Simulation Environment Based On Monte Carlo Method

Molecular imaging is a new emerging interdiscipline, which involves the molecular biology, physics, mathematics, information science, radiation medicine and so on. As a new imaging modality, in vivo optical imaging has become one of the research focuses in the field of molecular imaging. Monte Carlo (MC) has been widely used to simulate light transport in tissues for various applications and gone through several improvements. Based on the Monte Carlo method, a platform of Molecular Optical Simulation Environment (MOSE) is developed in our lab, which supports the simulation of photon transport in 2D/3D geometries or arbitrary-shaped phantoms described by triangle meshes.There are mainly two contributions in this thesis:Firstly, Based on the original MOSE, a parallel mechanism is introduced and implemented on a small PC cluster by making use of the easy parallelism of Monte Carlo method.Secondly, in the view of the computational performance, two improved methods are proposed. On one hand, because each cell of the PC clusters has different computational performance, a load balance system which dynamically can assign jobs is designed to implement the load balance of different cells. On the other hand, a policy of splitting parallel random number sequence is employed, which can guarantee the randomness and independence of the random number sequences in different computing cells.Through parallel program verification experiments, we can see that the parallel programs effectively reduce the computation time, and that the dynamic task splitting algorithm to achieve the load balancing and the splitting of sequence of random number algorithm to achieve consistency of the calculation results.