The Study On Cone-beam CT Reconstruction Algorithm Acceleration

CT reconstruction algorithm is one of the cores of the CT system, and the speed of reconstruction directly decides the possibility of CT application. With the wide application of the third generation cone-beam CT scanning system, to find a suitable method which can improve efficiency of computing of the cone-beam reconstruction algorithm has a vary important value both in academic research and application. The cone-beam CT reconstruction algorithm can be divided into to categories:iterative methods and analytical methods. The most representative iterative method is Algebraic Reconstruction Technique, while FDK algorithm proposed by Feldkamp, David and Kress is one of the most widely used analytical methods. The projection operation which is contained in both iterative methods and analytical methods is the most time-consuming process. Therefore, the key of cone-beam CT reconstruction algorithm acceleration is to improve the speed of the projection operation.In this paper, the content consists of three aspects. The first one is the research on cone-beam CT forward projection algorithm based on ray-casting algorithm. The second one is the research on FDK algorithm acceleration by using GPU based on Cg language. The third one is the research on the fast ART algorithm base on Intel SIMD instruction set.For the first research, by analyzing the similarity between the process of ray casting and the process of CT forward projection, the idea of ray casting algorithm is applied to the CT forward projection algorithm, and a CT forward projection algorithm in circular cone-beam scanning mode based on ray-casting algorithm is proposed in the paper and accelerated by using GPU and CUDA technology. Compared with the results of the traditional forward projection algorithm, the proposed algorithm can improve the quality of projection images. What's more, it has the advantages of higher computational efficiency.For the second one, by analyzing the similarity between the back-projection process of the FDK algorithm and the process of perspective texture mapping in Computer Graphics, the back-projection process which is the most time-consuming part of FDK algorithm is accelerated by GPU hardware to achieve FDK algorithm acceleration. Then, the implementation of the idea is show in the last past of this section, and comparing to the results of the traditional CPU-based implementation of FDK algorithm, the new implementation accelerated by GPU can highly speed up the algorithm by more than ten times, and improve the quality of the reconstructed imagesFor the third one, Intel SIMD (Single Instruction Multiple Data) technology is developed by Intel Corporation for multimedia processing software with large scale data computation. It is a useful tool for parallel computation. In this paper, the technology is applied to the implementation of ART algorithm, and comparing to results of the traditional implementation, this method can highly speed up the algorithm, while maintaining the reconstruction precision.