The Study On X-ray Reconstruction Algorithm Acceleration By Using GPU

X-ray imaging technology is a kind of technology which can reconstruct the internal organization and structure based on projection. According to different reconstruction manners, X-ray reconstruction algorithms can be divided into to two categories:iterative and analytical methods. The most representative iterative method is Simultaneous Algebraic Reconstruction Technique (SART), while the most widely used analytical method is FDK algorithm which was proposed by Feldkamp, David and Kress. However, both projection and back projection process of SART algorithm as well as filtering and back projection process of FDK algorithm are time-consuming. As the requirements for real time and image quality in industrial and medical areas continue to increase, reconstruction speed has become one of the key factors that restrict their practical applications. Therefore, the acceleration research of these two reconstruction algorithms has significant meaning.In this paper, the content consists of two aspects. The first one is the research on filtering acceleration of FDK algorithm by using GPU based on Cg language. The second one is the research on Tomosynthesis reconstruction acceleration based on CUDA.For the first research, by analyzing the filtering process of FDK algorithm in detail, studying the most time-consuming part—FFT calculation process and using RGBA four channel texture technology and "Pingpang" technology, this paper implements the filtering acceleration process of FDK algorithm by using GPU to improve computational efficiency. Compared with the results of traditional CPU-based implementation of filtering process of FDK algorithm, GPU-based implementation can highly speed up the algorithm.For the second research, this paper accelerates SART algorithm to implement three kinds of reconstruction methods by using GPU. During reconstruction process, the method of ray-casting algorithm is applied to projection algorithm while the method of MBR is applied to back projection reconstruction algorithm, which both significantly improve the computa-tional efficiency of Tomosynthesis reconstruction algorithm. To solve layer interference problem of Tomosynthesis, TV algorithm is added which improves the quality of the reconstructed images. And this paper implements the reconstruction process with PCB X-ray projection and gets a good result which proves superiority as well as practicality of the proposed method. This paper also discusses some relevant knowledge of spatial resolution, including horizontal resolution and vertical resolution, and their influence on experimental results. At last, this paper does related research on the problem how the changes of projection angle affect the experimental results.