GPU加速剂量计算中微分卷积/积分算法的实现

GPU-based ultra fast dose calculation using differential convolution/superposition algorithm

微分卷积/积分算法是计算精度较高的光子线剂量计算算法,较长的计算时间限制了该算法在临床上的使用。本文对微分卷积/积分算法中最耗时的部分实现了基于GPU的并行化计算,与基于CPU的计算相比,在Tesla C1060上计算速度提高可达30X-60X。利用γ因子对计算结果的准确性进行了分析,结果显示,无论是均匀水模还是非均匀头模,在单照射野还是多照射野情况下,加速后的结果都与CPU的计算结果有相同的准确性。通过GPU并行加速,微分卷积/积分算法能成为日常的剂量计算算法。

Background： Dose calculation plays a key role in treatment planning for radiotherapy, its performance and accuracy are crucial to the quality of treatment plans. Differential convolution/superposition algorithm is considered as an accurate algorithm for photon dose calculation; however, improvement on its computational efficiency is still desirable for such purpose as real time treatment planning. Purpose： The goal of this work is to boost the performance of differential convolution/superposition algorithm by devising a graphics processing unit （GPU） implementation so as to make the method practical for daily usage. Methods： In this work, we implemented a GPU-based version of the differential convolution/superposition algorithm, by which the most time-consuming parts are implemented on GPU. In order to fully utilize the GPU computing power, the algorithm is modified to match the GPU hardware architecture. Results： Compared with the algorithm completely nmning on CPU, the GPU-based algorithm can speed up 30-60 times on a Tesla C1060 with higher values corresponding to larger field size. Finally, we use y index to analyze the accuracy of calculation results, no matter one field or multi-field, homogeneous phantom or inhomogeneous phantom, the GPU implementation has the same accuracy as the CPU implementation. Conclusions： GPU is a useful solution for satisfying the increasing demands on computation speed and accuracy of dose calculation. The GPU-based differential eonvolution/superposition can be feasible and cost-efficient for satisfying the increasing demands for either computation speed or accuracy by advanced radiation therapy technologies.