GPU-accelerated Monte-Carlo modeling for fluorescence propagation in turbid medium

In biomedical optics, the Monte Carlo (MC) simulation is widely recognized as a gold standard for its high accuracy and versatility. However, in fluorescence regime, due to the requirement for tracing a huge number of the consecutive events of an excitation photon migration, the excitation-to-emission convention and the resultant fluorescent photon migration in tissue, the MC method is prohibitively time-consuming, especially when the tissue has an optically heterogeneous structure. To overcome the difficulty, we present a parallel implementation of MC modeling for fluorescence propagation in tissue, on the basis of the Graphics Processing Units (GPU) and the Compute Unified Device Architecture (CUDA) platform. By rationalizing the distribution of blocks and threads a certain number of photon migration procedures can be processed synchronously and efficiently, with the single-instruction-multiple-thread execution mode of GPU. We have evaluated the implementation for both homogeneous and heterogeneous scenarios by comparing with the conventional CPU implementations, and shown that the GPU method can obtain significant acceleration of about 20-30 times for fluorescence modeling in tissue, indicating that the GPU-based fluorescence MC simulation can be a practically effective tool for methodological investigations of tissue fluorescence spectroscopy and imaging.