Performance Evaluation And Optimization Of Numerical Solution For FEM Based On Light Transfer In Biological Tissues

Optical Molecular Imaging has the advantage of safety, noninvasive, low cost, high specificity and Preparation of molecular probes stored convenience, so that it has become the hot topic into today’s Optical Molecular Imaging researches. The primary problem in Optical Molecular Imaging is how to describe Optical transmission in biological tissue more accurately and then, modeling it precisely and solving it quickly. Among many optical transmission model, SPN model, which is the higher-order approximation of RTE, receiving more focus in the field of Optical Molecular Imaging。However, with the rise of order N, model has rapid increase in computational cost, that limits its application. Therefore, basing on GPU, this article pose a speeded-up strategy to accelerate the solving process of generate stiffness matrix and Linear Equations, which are the two largest consuming modules during the whole solving process. This speeded-up strategy in views of Compute Unified Device Architecture(CUDA), which is general purpose parallel computing framework and lunching by the NVIDIA Company.Considering the features of CPU and GPU, this article will optimize the process correspondingly, which includes distribute computing tasks, rational use of memory and data communication, as well as simple data processing. The optimization process raises the advantage of CPU/GPU hybrid platform and improves the acceleration effect. Finally, this paper will prove the accuracy of the speeded-up strategy by experiment. Through the experiment, Algorithm’s speedup ratio can be found out and that shows the speeded-up strategy has excellent acceleration effect.