Quality Changes of Image from Total Variation to Nonlinear Sparsifying Transform for Sparse-view CT Reconstruction

Abstract

Sparse-view CT has been widely studied as an effective strategy for reducing radiation dose to patients. Total variation (TV) minimization, which is most extensively studied among the existing compressed sensing (CS) techniques, has been recognized as a powerful tool for dealing with the inverse problem of sparse-view image reconstruction. However, in recent years, the drawbacks of TV are being increasingly reported, such as appearance of patchy artifacts, depict of incorrect object boundaries, and loss in image textures. In order to address these drawbacks, a series of advanced algorithms using nonlinear sparsifying transform (NLST) have been proposed very recently. The NLST-based CS is based on a different framework from the TV, and it achieves an improvement in image quality. Since it is a relatively newly proposed idea, within the scope of our knowledge, there exist few literatures that discusses comprehensively how the image quality improvement occurs in comparison with the conventional TV method. In this study, we investigated the image quality differences between the conventional TV minimization and the NLST-based CS, as well as image quality differences among different kinds of NLST-based CS algorithms in the sparse-view CT image reconstruction. More specifically, image reconstructions of actual CT images of different body parts were carried out to demonstrate the image quality differences. Through comparative experiments, we conclude that the NLST-based CS method is superior to the TV method in the task of image reconstruction for sparse-view CT.