Performance of CUDA GPU in Monte Carlo simulation of light-skin diffuse reflectance spectra

We evaluated the simulation of light-skin diffuse reflectance spectra using graphic processing unit (GPU)-based Monte Carlo for Multi-Layered Media (MCML) which is known as CUDAMCML. The platform was a laptop with 2.3 GHz processor (Intel Core i5-2410M) supported by NVIDIA´s Compute Unified Device Architecture (CUDA) graphic card (GeForce GT 520M). A simplified three-layered skin tissue model at 380 - 780 nm wavelengths with interval of 10 nm was used as simulation problem. The comparison was on diffuse reflectance values computed by CUDAMCML and traditional MCML as well as their computation performance. This study observed small differences of diffuse reflectance values (0.03 - 0.68 %) between CUDAMCML and MCML which increased with increasing wavelengths. CUDAMCML offers an extremely practical computation time of 6.4 minutes compared to MCML which took 7.6 hours in the simulation of spectral diffuse reflectance. A total of 71.19-fold speed-up was achieved which varied across the wavelengths. The acceleration by CUDA GPU offers capability of running bulk data in computer-aided photodiagnosis of the skin tissue model.