GPU-accelerated Monte Carlo simulation for photodynamic therapy treatment planning

Abstract

Recent improvements in the computing power and programmability of graphics processing units (GPUs) have enabled the possibility of using GPUs for the acceleration of scientific applications, including time-consuming simulations in biomedical optics. This paper describes the acceleration of a standard code for the Monte Carlo (MC) simulation of photons on GPUs. A faster means for performing MC simulations would enable the use of MC-based models for light dose computation in iterative optimization problems such as PDT treatment planning. We describe the computation and how it is mapped onto the many parallel computational units now available on the NVIDIA GTX 200 series GPUs. For a 5 layer skin model simulation, a speedup of 277x was achieved on a single GTX280 GPU over the code executed on an Intel Xeon 5160 processor using 1 CPU core. This approach can be scaled by employing multiple GPUs in a single computer -a 1052x speedup was obtained using 4 GPUs for the same simulation.

Reducing branch divergence in GPU programs

“…We use two synthetic benchmarks (one for each optimization) and one highly-optimized realworld application called Monte Carlo simulation for Multi-Layered media (MCML) [9]. We parameterize the synthetic benchmarks to explore the impact of various kernel characteristics on the benefit of these optimizations.…”

Next-generation acceleration and code optimization for light transport in turbid media using GPUs

“…Unless each thread uses a unique sequence of random numbers, there is a risk that multiple threads will simply re-calculate one another’s results, which would affect the signal-to-noise ratio in the resulting simulation output. Simply seeding the PRNG state differently for each thread, an approach taken in [11, 16, 18], is not sufficient to ensure against inter-thread correlation of random numbers. The GPU implementation of the Mersenne Twister (MT) PRNG used by Fang and Boas [16] provides unique random numbers for threads within a block but still potentially suffers from correlation between different thread blocks.…”

A Coarse Grained Particle Transport Solver Designed Specifically for Graphics Processing Units

“…There has been a lot of recent activity in adapting Monte Carlo transport algorithms to streaming Downloaded by [Swinburne University of Technology] at 03:52 03 January 2015 processors (in the radiation treatment planning community see for instance, Badal and Badano, 2009b;Lo et al, 2009;Jia et al, 2010;Tickner, 2010;and in neutronics, Nelson and Ivanov, 2010;Aiping Ding et al, 2011). Although a Monte Carlo simulation is embarrassingly parallel, it does not map well to this type of architecture, which groups processing threads into a large number of Single Instruction Multiple Data (SIMD) like instruction units.…”