基于OpenCL的DRR算法优化研究

摘要

放射治疗计划系统(TPS)是为放疗患者制定放疗计划的特殊系统,对放射治疗的精度和效果有重要影响,是放射治疗的核心技术之一.数字重建放射影像算法(DRR)是TPS中的关键算法,广泛用于实现射野验证、病人摆位等.针对DRR算法性能不能满足交互式和实时性的要求,提出了一种使用OpenCL技术对DRR算法进行并行加速的计算方法.首先介绍了DRR算法在TPS系统中的重要性和OpenCL框架,接着在DRR算法的优化过程中以X射线衰减理论公式为基础,构建了DRR算法的串行版本作为算法优化的基准,分析了基于光线跟踪的DRR算法的特点,给出了算法的并行化方案,并结合OpenCL的存储器特性对并行算法进行了性能调优.在NVIDIA平台对算法进行了实验和评测,结果显示, OpenCL并行优化版本相对其串行版本加速约36倍左右,满足了系统的性能要求.%Radiotherapy treatment planning system(TPS),as one of core technologies of radiotherapy,is a special system for radiotherapy patients,which has important influence on the accuracy and effectiveness of radiotherapy.Digitally reconstructed radiograph(DRR) is a key algorithm in TPS and widely used in field validation,patient placement and so on.Concerning the performance of DRR can't meet the interactive and real-time requirement,we propose a new approach to accelerate the DRR using open computing language(OpenCL). Firstly,we introduce the importance of DRR in TPS and OpenCL framework.Then in the optimization,according to the formula of X-ray attenuation,we develop the serial version of the DRR as the benchmark of optimization,and analyze the characteristics of DRR based on ray tracing.Finally the parallel scheme about the DDR is established and done performance tuning according to OpenCL memory architec-ture.The experiment and test on NVIDIA platform show that the parallel optimization of OpenCL is about 36 times faster than its serial version,which can meet the requirements of the system performance.