论文部分内容阅读
Backgrounds: Owing to unique radiological physics properties associated with the Bragg peak,proton and heavy ion therapies are potentially advantageous than the x-ray photon based external beam therapy.An accurate dose calculation method is of great importance in treatment planning and dose verification.Although Monte Carlo(MC)method provides the most accurate dose calculations,the time is often prohibitively long unless parallel computing techniques are used.Benefiting from the rapid development of computer science,the graphics processing unit(GPU)has recently shown a great potential for high performance MC computing.To date,although a dozen studies have explored the GPU-based MC acceleration techniques,few focused on proton and heavy ion therapies.The goal of this study is to extend a parallel computing MC code,called ARCHER(Accelerated Radiation-transport Computations in Heterogeneous EnviRonments),to proton and heavy ion therapies.ARCHER is designed to run Monte Carlo calculations using a number of emerging heterogeneous hardware architectures including devices such as NVIDIA GPUs and the Intel Xeon Phi coprocessors.This paper describes the results from ARCHER in comparison with those from production codes,GEANT4 and FLUKA.Methods: The ARCHER was written in C++for the CPU version and for a GPU version for direct performance comparison.The code was designed to model the proton and heavy ion transport in media using condensed history method and the MC simulation process was executed in the double precision mode.A desktop PC with a dual NVIDIA TeslaTM M2090 GPUs and an Intel Xeon X5660 CPU were employed as the computation platform.To focus on software testing,a simple water phantom was considered in the comparison.Results: In this project,the depth doses in the water phantom were calculated,as shown in Figure 1.A total of 10 7 particles were processed by the given MC codes with the relative error less than 1%.For a proton pencil beam of 100 MeV,150 MeV and 200 MeV,the dose differences between ARCHER and GEANT4 was found to be within 2%of the dose maximum for 97%of all tallies and,in the same way,within 2%for 95%of FLUKA tallies.Table 1 illustrates the comparison of execution times.The speedup factor of our GPU code over the CPU code was found to be 57 for single GPU card.The speedup for the dual GPU system over the singe-CPU GEANT4 was founded to be 177,and there is also 208 times acceleration for the dual GPU system over the single-CPU FLUKA.Conclusion: The preliminary results on proton and heavy ion simulations in the ARCHER code agree well in our study that involved simply physics and phantom.On-going work will enhance the physics and more realistic phantom modeling in the comparison.