中文摘要：

目的：蒙特卡洛模拟被认为是目前剂量计算方面最为精确的算法,但是因为其模拟时间过长,在临床应用上受到限制。EGSnrc作为目前在医学物理领域应用最为广泛的蒙特卡洛模拟软件,因为其过长的执行时间,其在临床方面的应用受到很大限制。为了克服这一障碍,我们开发了一个基于GPU的蒙特卡洛模拟程序,以期为放疗计量提供一个高效和低成本的蒙卡程序。方法：本文给出了一种基于GPU（Graphic Processing Unit）的蒙特卡洛模拟的新方法,开发语言是CUDA 5.0,将目前最为通用的蒙特卡洛程序EGSnrc移植到GPU平台,保留EGSnrc的核心物理过程以及输运过程的算法,这可以在最大限度保持原来EGSnrc模拟精度的前提下,极大地提高蒙特卡洛模拟的效率。GPU版本的蒙特卡洛模拟程序运行在一块英伟达Tesla C2050显卡上。GPU版本的EGSnrc精度的验证采用了纯水模体,同时,入射的射线我们选择为6 MV的光子。为了进一步检验GPU版本的EGSnrc的精度,我们进行了一个逐体素的检验,检验结果显示,GPU版本的EGSnrc和EGSnrc符合的很好。结果：最终实验结果表明,在模拟20亿个相空间事例的情况下,使用NVIDIA Tesla C2050显卡,新的基于GPU的蒙特卡洛程序的速度比在单核的Intel Xeon 2.0 GHz CPU上的模拟速度提高了43倍,且其精度与EGSnrc的精度相当。计算结果的方差在高剂量区域（D〉Dmax）小于0.5%,计量误差经过Dmax归一化之后,其和EGSnrc的误差小于1%的比率在占整个区域的90%以上。结论：通过此新程序表明,基于GPU的蒙特卡洛算法可以极大地提高蒙特卡洛程序的运行效率,与此同时,GPU版本的EGSnrc在最大程度上保持了EGSnrc的模拟精度。考虑到GPU版本的EGSnrc程序的速度以及精度优势,其在未来的临床应用中有着巨大的前景。

英文摘要：

Objective Monte Carlo simulation is generally considered to be the most accurate algorithm for radiation therapy dose calculation. However, it is very time consuming for Monte Carlo simulation to get a desirable accuracy, which makes it difficult to use in routine clinical application.EGSnrcis considered to be the most widely used Monte Carlo Simulation code in Medical Physics, the long execution time is the main barrier for its routine clinical application. To overcome this obstacle, a graphics processing units （GPU） based parallel computing version of EGSnrc was developed to provide a fast and low cost solution for accurate Monte Carlo simulation. Methods In this paper, we give a GPU-based Monte Carlo simulation program package using the CUDA 5.0 environment. We implement the EGSurc to the GPU platform, the EGSnrc＇s core physics interaction and particle transportation mechanism were maintained in the GPU version of EGSnrc, which will maintain the accuracy of EGSnrc, at the same time we accelerate its execution speed.The GPU based Monte Carlo simulation code is run on a Nvidia Tesla C2050 card. And the test phantom we use in this experiment is a pure water phantom, and the incident source we choose is a 6 MV photon. Results Our research results show that using a NVIDIA Tesla C2050 GPU card against an Intel Xeon CPU card, both simulates 2 billion phase-space histories, our GPU-based Monte Carlo program package will get a 43 speed-up factor at the same time maintain the accuracy of EGSnrc. To further test the agreement of the calculation result, we run a voxel by voxel test,which shows that the GPU based EGSnrc agree well with the EGSnrc.When the uncertainty is less than 0.5% everywhere in the high dose region （D〉0.5Dmax）, the dose difference normalized to Dmax between GEGS and EGSnrc is less than 1% for more than 90% of the voxels. Conclusion Our new GPU-based Monte Carlo program package could accelerate the Monte Carlo simulation in a great degree, at the same time, maintains the accuracy of EGSnrc.