An efficient and modified median root prior based framework for PET/SPECT reconstruction algorithm

Abstract

Bayesian statistical algorithm plays a significant role in the quality of the images produced by Emission Tomography like PET/SPECT, since they can provide an accurate system model. The major drawbacks associated with this algorithm include the problem of slow convergence, choice of optimum initial point and ill-posedness. To address these issues, in this paper a hybrid-cascaded framework for Median Root Prior (MRP) based reconstruction algorithm is proposed. This framework consists of breaking the reconstruction process into two parts viz. primary and secondary. During primary part, simultaneous algebraic reconstruction technique (SART) is applied to overcome the problems of slow convergence and initialization. It provides fast convergence and produce good reconstruction results with lesser number of iterations than other iterative methods. The task of primary part is to provide an enhanced image to secondary part to be used as an initial estimate for reconstruction process. The secondary part is a hybrid combination of two parts namely the reconstruction part and the prior part. The reconstruction is done using Median Root Prior (MRP) while Anisotropic Diffusion (AD) is used as prior to deal with ill-posedness. A comparative analysis of the proposed model with some other standard methods in literature is presented both qualitatively and quantitatively for a simulated phantom and a standard medical image test data. Using cascaded primary and secondary reconstruction steps, yields significant improvements in reconstructed image quality. It also accelerates the convergence and provides enhanced results using the projection data. The obtained result justifies the applicability of the proposed method.