Computed tomography (CT), as a non-destructive imaging modal, has become one of the most prevalent tools in the clinical, pre-clinical as well as industry. With the development of hardware, current CT scanners have faster and faster scanning speed, and generate huge amount of data. All of these new features put forward new requirement on the reconstruction speed. In this thesis, the fast CT reconstruction algorithms have been studied, involving the acceleration of fan beam reconstruction, cone beam reconstruction in term of circle orbit, spiral orbit, as well as circle-line orbit. The main contributions of this thesis include following issues: 1. Accelerate FDK reconstruction algorithm using GPU hardware, and propose some acceleration technologies such as cyclic render-to-texture (CRTT). The fast reconstruction algorithm has been applied to our Micro-CT system. 2. Review the development of the exact reconstruction algorithms, and accelerate Katsevich algorithms in term of spiral orbit and circle-line orbit. In spiral orbit an overscan formulation is derived which is helpful in practical CT scanning. Based on the overscan formulation, a volume blocking method in GPU memory is proposed, which saves the limited GPU memory. 3. CT reconstruction framework into MITK toolkit. Fan beam reconstruction by CPU and GPU acceleration, circle orbit cone beam reconstruction by CPU and GPU acceleration, cone beam reconstruction in term of spiral and circle-line orbit by GPU acceleration have been added into the framework, and the framework is convenient to extend other algorithms.
修改评论