Research On Fast Cone-beam CT Reconstruction Algorithms Using GPU

CT(Computed Tomography) is a technology used to obtain internal information of measured objects by the X-ray scanning. This technology is an important branch in the field of nondestructive testing and widely used in industrial, medical, aerospace, machinery and other fields. With the characteristics of fast scanning speed and high spatial resolution, the cone-beam CT technology has become the focus and hotspot of CT research currently.However, the great amount of computation and data transmission requirement of cone-beam CT 3D images reconstruction makes it time-consuming, and it is impossible to meet the requirement of 3D images reconstruction in real-time only using CPU. For nearly half a century, with the rapid development of GPU(Graphics Processing Unit), its powerful parallel computing ability has a revolutionary significance to the acceleration of CT reconstruction algorithm. Especially, the CUDA technology makes the GPU programming easier, which is suitable for developers to master quickly and could shorten the program development cycle. This provides a new technology for the acceleration of the cone-beam CT reconstruction algorithm.In this paper, we mainly study the cone-beam FDK algorithm. First of all, this paper briefly introduces the research status, purpose and significance of the CT technology and the CT reconstruction algorithm. Secondly, the composition of CT system, the properties of X-ray and the mathematical physics basis of CT imaging are summarized. Third, start from the FBP algorithm for 2D fan-beam CT, focus on the selection of the filter function in the process of the algorithm are analyzed, and then completed the research work of FDK algorithm in 3D cone-beam CT. Finally, using the skull model as the experimental data, the FDK algorithm of 3D cone-beam CT was realized using GPU.The results show that the selection of the filter function will affect the quality of the reconstructed image. By using the two-dimensional Shepp-Logan model for experimental data, puts forward several filtering function according to a certain weight integration method to construct the new mixed filter function. Based on this, the quality of the reconstructed image of CT in the case of noise free and noisy is studied respectively. In the case of no noise, the image quality of the RL-SL filter function CT is the highest. In the case of noise, FBP algorithm CT reconstruction image quality is generally poor. But RL-NEW filter function in this case, CT reconstruction image quality is slightly better than other filtering function. By using GPU, the FDK algorithm has a significant advantage in parallel operation. According to the model of human skull as experimental data, data reconstruction body size is 256*256*256 and 512*512*512 objects. Based on the CUDA architecture, compared reconstruction method and using CPU by GPU alone, were more than 100 times speedup. And the larger the size of the reconstruction data, the effect of acceleration is more obvious. Therefore, it can be concluded that the conclusion, with the CUDA technology continues to mature, continuous improvement of the performance of GPU, the speed of 3D cone beam CT reconstruction algorithm will also have a qualitative improvement, which can be better meet the CT reconstruction of real-time requirements.