摘要
基于CUDA编程技术,研究了如何将NVIDIA的GPGPU模型应用于剂量计算,并首次将该技术应用于基于点核卷积/迭加模型的三维放射治疗计划系统商业化产品.本工作对原有剂量计算模型做了改进,使其可以在device端进行并行处理.在程序架构设计中使用MFC导出类及动态库技术,避免了大量代码移植工作.对结果数据进行了比较与分析,确定了基于特定显卡效率最高的thread数目.结果表明:基于实际患者计划数据执行结果的评估,采用GPU技术加速,大大提高了系统剂量计算速度,使射野剂量计算速度在1 s以内,大大增强了产品市场竞争力.
The application of GPGPU model of NVIDIA in dose calculation based on CUDA program technique was studied.It was the first time that GPGPU method was used in commercial 3D radiation therapy planning system based on point kernel convolution/superposition model.The old dose calculation model was modified in order to make it be used in device for the parallel computing.At the same time the MFC DLL and export class technique was used to avoid mass code migrating work.The most efficient number of parallel threads was determined by analyzing the result data.The results showed that the dose calculation speed is improved remarkably and the single beam computing time is less than one second using GPU speedup method,which improved the competition ability of the product.
出处
《东北大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2012年第5期649-652,共4页
Journal of Northeastern University(Natural Science)
基金
国家自然科学基金资助项目(61071057)
