CROSS: Cross-Domain Residual-Optimization-Based Structure Strengthening Reconstruction for Limited-Angle CT
Dianlin Hu, Yikun Zhang, Guotao Quan, Jun Xiang, Gouenou Coatrieux, Shouhua Luo, Jean-Louis Coatrieux, Xu Ji, Hongbin Han, Yang Chen
Abstract
Limited-angle computed tomography (CT) is an effective way for practical scenarios due to its flexibility in various complex scanning conditions. However, incomplete projection data will lead to severe wedge artifacts and degraded images, which significantly lower the diagnostic values. To overcome this problem, we propose a novel method termed cross-domain residual-optimization-based structure strengthening (CROSS) reconstruction for limited-angle CT. The proposed CROSS framework consists of three steps, which are conducted on the image domain and measurement domain alternatively. Differing from traditional dual-domain-based algorithms, our CROSS method not only regularizes the reconstruction results on the image space but also the residual-error space, which boosts organ recovery where the area has a larger attenuation coefficient. Besides, the structure-strengthening network is adopted to enhance tissue preservation. Simulated and preclinical datasets are conducted to evaluate the proposed CROSS method. Experiments show that the proposed framework could produce a better performance in artifact removal and edge preservation.