Study On Iterative CT Reconstruction Algorithm Based On Image Rotation

Computer tomography(Computed Tomography,CT)is currently the most widely used medical imaging technology.The current mainstream CT image reconstruction algorithm is still an analytical method,but this algorithm cannot effectively remove the artifacts caused by sparse projection data,so iterative methods that can be used for high-precision sparse reconstruction are getting more and more attention.The Total Variation(TV)algorithm is a classic iterative reconstruction algorithm based on compressed sensing,which can reconstruct images with high precision from sparse projections.However,the system matrix corresponding to the algorithm is often extremely large,and it cannot even be stored in the computer's memory.In order to avoid the system matrix,this paper proposes a TV algorithm based on image rotation,the specific content is as follows:(1)The classic Adaptive Steepest Descent Projection onto Convex Sets(ASD-POCS)algorithm including the system matrix is implemented,and the method of obtaining the system matrix and the effect of the algorithm parameters on the reconstruction are made In-depth study.(2)Designed an ART(Algebraic Reconstruction Technique)implementation method based on image rotation,and then constructed a new TV algorithm,focusing on the impact of different interpolation algorithms on reconstruction results.The best performance can be achieved in accuracy.The system compares this algorithm with traditional TV algorithm.It is found that the algorithm of this paper can achieve the accuracy equivalent to the traditional TV algorithm no matter whether it is no noise or no noise,and the reconstruction speed is increased by nearly 6 times,while the storage space is greatly reduced.(3)In order to further improve the reconstruction accuracy of the algorithm,based on the TV model based on image rotation,the NLTV model is used for reconstruction.The GPU is used to accelerate the calculation of the weight matrix of the NLTV model.The experimental results show that the NLTV model has higher reconstruction accuracy than the TV model,and the acceleration effect of the NLTV model based on GPU parallelization is obvious.Aiming at the shortcoming that the system matrix of traditional TV algorithm needs to occupy huge storage space,this paper designs a novel TV algorithm based on image rotation.This algorithm saves the memory consumption in the reconstruction process,improves the reconstruction speed,and can achieve the reconstruction accuracy equivalent to the traditional method.In the case of strict requirements on the use of computer memory,the algorithm has certain theoretical significance and practical value.