Accelerated Research And Implementation Of X-ray CT Reconstruction Algorithm

In CT reconstruction,the reconstruction algorithm plays an extremely important role,and the reconstruction speed is inseparable from the demand in practical applications.In particular,with the improvement of the resolution of the area array detector and the resolution of the reconstructed image by the project,the amount of reconstructed data suddenly increases,and the execution efficiency of the existing reconstruction algorithm cannot meet the application requirements.Secondly,in order to realize high-quality reconstruction with limited angle,sparse angle and multi-energy spectrum,the iterative reconstruction algorithm has advantages,but it has high computational complexity,slow reconstruction speed,and limited engineering applications.Therefore,this article starts with the hardware and software acceleration methods of commonly used reconstruction algorithms and does research on the acceleration of reconstruction algorithms:1.Hardware acceleration is mainly reflected in the CUDA architecture.The current GPU parallel architecture performs well in many reconstruction accelerations,but in the reconstruction of high-resolution images,there are problems of insufficient display and frequent data exchange between the CPU and the GPU,causing the reco nstruction time to increase.This paper has optimized the relevant data transmission thread for the above problems.Open up specialized data transmission threads to solve the problem of huge memory shortages.C hange the existing CPU and GPU serial data access modes to parallel,and achieve parallel data transmission and CT reconstruction,effectively improving the speed and efficiency of algorithm reconstruction.2.Software acceleration is mainly manifested in optimizing the cumbersome operations and reducing the complexity of the algorithm.Real-time imaging and low doses have been hot spots in CT technology.Although the iterative algorithm has advantages in reducing the X-ray dose and reducing the scanning time,it also suffers from the problems of long reconstruction iteration time and slow convergence.To solve the problem,based on the accelerated OS-SPS algorithm,this paper studies the method of relaxation linearization augmented Lagrangian.Through simulation and verification,compared with OS-SPS algorithm,the improved reconstruction method of the paper starts to converge at iteration 4,which effectively improves the convergence speed of the algorithm.