Study On Accelerated Ray Tracing Method In PET Image Reconstruction With GPU

PET image reconstruction has always been a hot PET studies, due to the hugenumber of rays generated, and therefore very large amount of computation, especiallyiterative algorithms for image reconstruction, the use of a longer time. NVIDIACUDAarchitecture introduced in2007to implement a lot of data in large-scaleparallel computing on the GPU. CUDAcomputing architecture is an advantage of thecomprehensive utilization of the CPU and the GPU, the CPU utilization of serialarithmetic logic, while the GPU is employed to perform high-density calculationsection parallel, this can take advantage of the respective CPU and GPU advantage,improve operational efficiency.This paper summarizes the research and CUDA features, application architecture,development environment, program execution models, etc., and then use CUDAinPET image reconstruction to accelerate operations, focusing on the system responsematrix in the design and implementation of CUDA.The main contents of this paper are:1.Proposed a PET system response matrix iterative algorithm design method,which is based on ray tracing algorithm, the first element of the plane is divided into anumber of pixel cells, and then use ray tracing to calculate the pixel cell line througha LOR, and calculated based on the pixel unit through the system response matrix.2.GPU hardware combined cases, PET collected data into the appropriate subset,and use an iterative algorithm for image reconstruction, the use of CUDAin thisprocess to accelerate operations, including ray tracing algorithm, the system responsematrix calculation, image iterative calculations are suitable for parallel computing toaccelerate the GPU, while the CPU is responsible for the serial and logic controlsection. From the experimental results, achieved satisfactory speedup reconstructedimage quality is also good.