以CBCT为核心的图像引导放疗中，放疗患者多次行CBCT扫描以实现放疗精确摆位，但同时引起了额外的辐射剂量，减少CBCT扫描帧数是降低CBCT辐射剂量直接有效的途径。仅利用部分投影帧数据进行CBCT高质量重建，并构建CPU+多GPU构架将系统重建速度提高到临床可接受水平，是本项目的核心问题。. 本项目将基于压缩感知理论和字典学习算法，将少数据CBCT重建问题转化为最优化问题求解，充分利用同一放疗部位的先验CBCT图像，构建具有结构、特征、精度等先验信息的双精度字典，通过迭代算法中高低精度字典的替换迅速提高CBCT图像重建质量。同时，搭建CPU+多GPU异构并行编程模式加速重建过程，构造临床适用的图像引导放疗系统。. 模拟实验表明，在保证CBCT图像重建质量的前提下，本系统能够将CBCT扫描剂量降低至全扫描情况下的30%，显著减少患者的CBCT扫描辐射危害。

In image-guided radiotherapy (IGRT) based on Cone Beam CT (CBCT), daily CBCT imaging procedures correct the set-up errors of patients, but also increase additional dose to patients' normal tissues. Minimizing imaging views of CBCT is an direct and effective way to reduce extra radiation exopsure. Our goal is to establish a novel high-quality CBCT reconstruction system with limited CBCT imaging viewes at a high reconstruction speed based on CPU+MultiGPU configuration avilable for image-guided radiotherapy clincial practices..We will complete two tasks in our project. One is to further establish a high-quality recosntruction algorithm with limited number of CBCT imaging views to solve the CBCT reconstruction model mathmatically. The second task is to construct a CPU+multiGPU heterogeneous programming mode to accelerate the reconstruction process and establish an IGRT CBCT system suitable for clinical application..Our inital experimental simulation shows that our system is promising to reduce the CBCT radiation dose to 30% in clinical IGRT practice and significantly reduce the radiation hazards produced by IGRT.