Iterative thresholding framework for row-action reconstruction from sparse projection data

Abstract

This paper presents a framework for row-action type iterative reconstruction from projection data measured over small number of projection views, or sparse projections. Image reconstruction from sparse projections usually suffers from streak artifacts that degrade the image quality. However, this imaging scenario becomes a hot topic of research due its possibilities to reduce patient dose and other benefits in several imaging applications. The motivation behind this work is the use of l 1 =l 0 norm distance to a reference image to select the sparse solution corresponding to the difference between the reconstructed image and the reference image. The concept of Lagrangian duality is used to derive the iterative thresholding framework which can be classified as an ART-like reconstruction method. The work presented here can be thought as a generalization of the previous work in (Li et al., 2004) which mainly focuses on the reconstruction of sparse objects, such as blood vessels. However, most of clinical applications consider imaging of non-sparse objects. The extension to the general case of x-ray computed tomography (CT) imaging, where the target object is non-sparse, is presented here. Experimental data indicates the power of the proposed method in reducing streak artifacts produced from data down-sampling