研究一种医学动态锥束CT重建的计算框架，能够有效地解决目前医学临床锥束CT和医学预临床实验的小动物动态锥束CT面临的低剂量、稀疏角等投影数据不完备问题。研究的主旨是结合自然图像信息处理技术，在重建中尽可能多地利用相邻投影数据和相邻图像之间相关性，以此提高重建算法的效能。主要研究内容包括：基于投影数据一致性约束的运动信号检测方法；基于三维字典学习和稀疏表示的锥束CT迭代重建算法，以及基于自然图像统计模型的正则化参数的自适应最优选择方法；基于LOG和梯度模值联合表示的DVF计算方法，更加有效地分析相邻时间间隔的CT图像之间的相关性；设计所提出算法的并行计算结构和其中的数据交换方法，充分利用GPU中各级存储器的空间大小、读写操作延时等优势，基于CUDA并行计算体系，开发出一种满足临床/预临床设备要求的动态锥束CT重建算法。

This project aims to develop a novel reconstruction framework for 4D-CBCT to solve the insufficient data reconstruction problems, such as low-dose, few-view and so on, for clinical and preclinical applications. Our idea is to combine advanced image and signal processing techniques in natural images into CT reconstruction procedure and utilize the correlations between neighboring projections and slices in 4D-CBCT to improve the performance of reconstruction. The project is composed of the following aspects: 1) projection correlations based movement signal detection; 2) 3D-dictionary learning based CBCT reconstruction method and the natural image statistical model based regularization parameters selection; 3) DVF computation based on joint representation of LOG signal and gradient magnitude, which is used to analyze the relationship between neighboring phase data; 4) parallel computation framework of the proposed methods by using GPU and CUDA to realize real time 4D-CBCT reconstruction method for clinical and preclinical applications.