Sparsity-induced dynamic guided filtering approach for sparse-view data toward low-dose x-ray computed tomography

Abstract

Iterative reconstruction (IR) methods that can incorporate filtering or regularization techniques have received widespread attention in many situations. Total variation (TV) regularization has proven to be a powerful tool to suppress streak artifacts and noise for sparse-view computed tomography (CT) reconstruction over 360°. However, with under-sampled projection data from limited-view (e.g. half-view) CT scanning, where the projections are further reduced, the edge structures are partly blurred, and some artifacts (such as blocky artifacts) are not effectively suppressed in TV-based results. To further improve the quality of the reconstructed image, a sparsity-induced dynamic guided image filtering reconstruction (SIDGIFR) method is proposed. Intermediate reconstruction results constrained by total difference (TD) minimization are taken as the guidance image to filter the results of projection onto convex sets (POCS) by guided image filtering (GIF). In the SIDGIFR algorithm, the guidance image is dynamically updated, which can transfer the important features (such as edge and small details) to the filtered image during the iterative process. To confirm the efficiency and feasibility of the SIDGIFR algorithm, simulated experiments and real data studies are performed. The quantitative evaluation shows that the proposed SIDGIFR method has better performance than other classical IR methods. What’s more, the proposed SIDGIFR algorithm can better preserve the edge structures, and suppress noise and artifacts, than the existing IR methods.