Research On Imaging Reconstruction Algorithms For Short-scan Cone-beam CT

Cone-beam Computational Tomography(CT)technology can obtain the internal structure information of the object without damage,and has the advantages of fast imaging speed,high spatial resolution and high radio efficiency.Short scan is an important research direction in the field of CT,which can shorten the imaging time and reduce the radiation dose compared with the traditional full scan,because it only requires ? plus fan angle of the scanning range.Image quality and imaging speed have always been the focus of the field of CT,so the study of fast and effective reconstruction algorithm for short scan imaging is of great significance.This paper focuses on the problem of image reconstruction under continuous sampling and sparse angle sampling for short scan cone beam CT.The short scan projections contain redundant part under continuous sampling,causing that it is important to study an efficient analytic reconstruction method to improve the image quality and reconstruction efficiency for short scan cone beam CT.At the same time,sparse angle sampling can be carried out in short scan,and designing an efficient optimization reconstruction algorithm has great significance to reduce the radiation dose and expand the application range.In view of the above problems,this paper mainly focuses on solving the redundancy problem of projection data,sparse sampling image reconstruction and asynchronous parallel iterative reconstruction respectively.The major achievements are as follows:1.A novel filter backprojection reconstruction algorithm based on selective backprojection(S-FDK)is proposed to solve the data redundancy problem in the short scan cone-beam CT.Compared with the traditional short scan filter backprojection reconstruction algorithm,this paper utilizes the geometric relations of projection rebinning and forward projection to derive a selective backprojection strategy,which can avoid the influence of redundant data and reduce the data range of backprojection reconstruction,leading to the improvement of efficiency.The experimental results show that the algorithm can effectively eliminate the influence of redundant data on the reconstruction results,and improve the reconstruction efficiency with high quality of reconstruction results as traditional short scan reconstruction algorithm.2.Total Variation(TV)minimization model is commonly used in sparse optimization reconstruction algorithm,nevertheless it is not optimal for image sparsity.A weighted difference of Li and L2(L1-?L2)on the gradient based on alternating direction method(WDG-ADM)is proposed in this paper to solve the above problem,which can obtain a sparse description closer to the LO norm.The alternating direction method is an efficient method to solve the proposed algorithm.The solving process of the subproblem is improved by linearization and approximate point technique,which improves the reconstruction efficiency of the algorithm.Finally,the stopping criterion of the algorithm is given by using KKT condition.The experimental results show that the proposed algorithm has a more significant advantage over the convergence speed and the accuracy of the proposed method than TV minimization algorithms under sparse angle sampling.3.Optimization reconstruction algorithm takes a long time to deal with cone-beam CT image reconstruction,which is commonly accelerated by synchronous parallel computation.However,the acceleration effect is limited by the calculation ability of the system's slowest nodes.Aiming at this problem,an asynchronous parallel alternating direction method for TV minimization reconstruction(Async-ADTVM)is proposed in this paper,which is implemented on a Multi-GPU system to maximize reconstruction efficiency by optimizing the communications between each node and the parallel computing within the node.The algorithm transforms TV minimization reconstruction problem into the generic fixed-point iteration problem,which satisfies the asynchronous parallel computing framework.It ensures the convergence of the algorithm,and can obtain higher speedup when the performance of the nodes is quite different in system.The experimental results show that the algorithm can achieve the same reconstruction precision as the traditional algorithm under the same iteration number,and can effectively reduce the influence of the performance difference between the nodes in the parallel acceleration system.