【摘要】 卷积/积分(Convolution/Superposition,CS)算法是精度仅次于蒙特卡罗(Monte Carlo,MC)的光子线剂量计算算法。该算法的计算速度远远快于蒙特卡罗算法,但仍不能完全满足临床放射治疗要求。借助单颗图形显卡GPU(Tesla C1060)对CS算法进行加速后,与传统的CPU串行计算相比,计算速度可以提高60倍,单野计算时间达到1 min左右,能适用于简单的三维适形计划(3DCRT),但无法满足调强放射治疗计划(Intensity Modulated Radiation Therapy,IMRT)的速度要求。本文通过设计"CPU+多GPU"异构模型加速方案,探讨使用不同GPU个数的加速情况。结果表明:CS算法加速倍数与GPU使用个数并非呈线性关系,通过合理选择GPU的使用数量和程序代码优化可达到相关计算的速度要求;基于中高端的Tesla C2015 GPU,采用"CPU+7个GPU"模型的CS算法,单野计算时间缩减到9 s,与单用CPU相比能提高207倍,可满足临床调强计划设计要求。更多还原

【Abstract】 [Background] The accuracy of Convolution/Superposition(CS) algorithm is considered to be next to Monte Carlo algorithm(MC) for radiotherapy dose calculation algorithm. Although the calculating speed of this algorithm is much faster than that of MC, its calculating speed can not fully meet the clinical requirements. With the aid of a single graphics processing unit GPU(Tesla C1060), the CS algorithm can be accelerated to 60 times faster than the traditional CPU serial calculation. The calculating time for single field is about 1 min which can be used in some simple three dimensional conformal radiotherapy planning(3 DCRT), but this calculating speed does not satisfy the speed need for intensity modulated radiation therapy(IMRT) planning. [Purpose] This study aims to explore a faster calculating speed solution of CS algorithm applied to IMRT with multi GPU. [Methods] The acceleration scheme of CPU + multi GPU heterogeneous model was analyzed by using different number of GPUs. High-end GPU,i. e., Tesla C2015, was used for experimental test of CS algorithm executing under the compute unified device architecture(CUDA) platform. Speeds of different number GPUs combined with CPU were compared to find the suitable solution. [Results] The experimental results show that the speedup of CS algorithm is not completely linear with the number of GPUs. With reasonable number of GPUs and optimized program codes, the computing time of CS algorithm for single field radiotherapy dose calculation can be reduced to 9 seconds by using 7 high-end GPUs(Tesla C2015), 207 times faster than that of a single CPU. [Conclusions] With the implementation of multi GPU heterogeneous model and code optimization, the CS algorithm can be applied to clinical IMRT treatment planning.更多还原