A Rapid Method For Dose Calculation Based On GPU

Radiotherapy is now one of the important means to treat malignant tumor. A rapid radiotherapytreatment planning is of great practical significance, in which dose calculation plays important role.The speed and accuracy of dose calculation have great effect on the efficiency and the quality ofradiation treatment planning.There are several common used dose calculating algorithm: Monte-Carlo algorithm, pencil beamalgorithm and convolution/superposition algorithm. Monte Carlo algorithm has a high accuracy butusually costs several hours; pencil beam algorithm has a faster calculating speed than Monte-Carloalgorithm, but the accuracy is relatively low. Convolution/superposition algorithm has a fastercalculating speed than Monte-Carlo algorithm and a higher accuracy than pencil beam algorithm. Animproved Convolution/superposition algorithm named Collapsed Cone convolution/superposition(CCCS) decreases the computational complexity and increases the calculating speed outstandingly,however its calculating speed still needs to be improved for practical application especially forintensity modulated radiotherapy (IMRT) planning.Spectra are one of the most important parameters in CCCS dose calculating, however it isdifficult to measure the spectra from a clinic accelerator directly. In this paper, simulated annealingmethod to reconstructed photon spectra is proposed. To evaluate the performance of the proposedmethod, the reconstructed photon spectra are compared with that calculated by Monte-Carlosimulation. The results show that it is reliable and effective to reconstructed spectra utilizingsimulated annealing method.For radiation treatment planning, dose calculating needs to be executed for many times inplanning optimization process. In order to improve the optimization speed of treatment planning, thecalculating speed of CCCS method needs to be improved in a great deal so as to meet practicalclinical requirements. Because of the high parallel level and high float calculating capability, graphicsprocessing units (GPU) has got great attention in different application fields, and has been used indose calculation and treatment plan optimizing in recent years. In this paper, GPU is proposed toaccelerate four core steps in CCCS dose calculation method with Compute Unified DeviceArchitecture (CUDA). The experimental environment is as follows: Window7operating system,Visual Studio2010development platform, CUDA4.1environment, NVIDIA Quadra FX380LP GPU,which has a1.2compute capability. In order to evaluate the performance of GPU-accelerated CCCS dose calculating method, theresults calculated by the proposed method are compared with those calculated by CCCS method basedon CPU in both computation time and accuracy. The experiment results show that CCCS dosecalculation algorithm with GPU acceleration could save much time while the dose calculationaccuracy can be kept. As the computation scale gets larger, the speed-up ratio gets more increasing.From these results, we can conclude that dose calculation by GPU is a reliable and rapid method: themethod could increase the CCCS dose calculation speed in a great deal while ensuring dosecalculation accuracy. With high performance and accuracy and low costs of GPU, the proposedmethod is much promising in clinical application in the future.