Research On Iterative Reconstruction Algorithms Of Computed Tomogoraphy

At present, the analytical technique and the iterative technique are two of the major image reconstruction methods in CT. Convolution back-projection algorithm is commonly used in analytical technique. The reconstruction speed of this method is fast. The quality of reconstructed image is better, but with heavy artifacts. In addition, it needs complete and equiangular projection datasets, and the integral path (the light) must be linear. The typical form of iterative method is algebraic reconstruction method. It is available to either complete or incomplete projection data. The major disadvantage of ART is its low convergent speed and largeness of its calculation which affects the application in practice.This paper mainly researches iterative method of CT imaging. First, based on the physical and mathematic principles, we studied the algebraic reconstruction technology (ART) and particularly analyzed the main factors that influence on the convergence speed or divergent of reconstruction method. Secondly, we studied the simultaneous algebraic reconstruction method (SART), and it was improved. The improved method (MSART) solved the marginal issues by SART. Meanwhile, we studied ordered subsets method and gave the method of selecting the sequence of subsets. We analyzed the influence on convergence speed and image quality by different methods. Then we analyzed the convergence of different projection orders and subset levels in order subsets. We conclude that it is better with different projection orders. We introduce the transcendental value and filters when OS-SART may not convergence at the situation of high subset level. In order to improve the speed of the algorithm, with symmetrical grid method, we introduce a fast method about projection coefficient. Then we got an improved method about projection coefficient.In this paper we use the software of MATLAB to simulation and analyzed the results of those experiments. The results show that the speed and quality had been impacted by the choice of relaxation parameter, the fashion of data access and the choice of the initial value. The convergence speed and image quality is better when the relaxation parameter is 0.5, the projection form is disorder, the initial value is the mean value of projections. MSART can be used in image reconstruction with limited views for that it had already solved the marginal issues.