Gpu Implementation Of The Vision Cone Beam Ct Image Reconstruction Method

As a non-destructive testing technology, Computerized Tomography (CT) has been widely used in multitudinous fields. Compared with traditional 2D CT, cone-beam CT is of several advantages, such as efficient utilization of X-ray photons, fast scanning speed and high spatial resolution.However, there are several problems restricting the applications of cone-beam CT, including field of view and reconstruction speed. In order to enlarge the field of view of cone-beam CT, helical scan mode and off-centered scan mode were used. In order to accelerate the reconstruction algorithm, specialized high-performance hardware were used for accelerating, including FPGA (Field Programmable Gate Arrays), Cell Broadband Engine processor and Graphics Processing Unit (GPU).The computation ability of the Graphics Processing Unit (GPU) is increasing much faster than that of the Central Processing Unit (CPU) in computer systems. At the same time, the highly processing power, parallelism and programmability nowadays of the current GPU makes the general-purpose computation available. Because of the differences between GPU instruction set and CPU instruction set, the key to using GPGPU is reengineering the computationally expensive algorithms to take advantage of this architecture as well as making use of rendering optimizations built into the programmable graphics pipeline.In this paper, we use RT (Rotation-Translation) multi-scan mode which enlarge the horizontal field of view of cone-beam CT and realize BPF (BackProjection Filtration) reconstruction algorithm on GPU. BPF algorithm is able to reconstruct image directly without rebinning projection data, so it does not decrease resolution of reconstructed image. The numerical experiment results show that the speed is increased by more than 10~2 times. In addition, we optimize the storage of texture in the algorithm, which saves video memory bandwidth, and is more suitable for reconstruction of multi-scan mode.