This website requires JavaScript.
DOI: 10.1016/

InDuDoNet+: A Deep Unfolding Dual Domain Network for Metal Artifact Reduction in CT Images

Hong WangYuexiang LiHaimiao ZhangDeyu MengYefeng Zheng
Dec 2021
During the computed tomography (CT) imaging process, metallic implants withinpatients often cause harmful artifacts, which adversely degrade the visualquality of reconstructed CT images and negatively affect the subsequentclinical diagnosis. For the metal artifact reduction (MAR) task, current deeplearning based methods have achieved promising performance. However, most ofthem share two main common limitations: 1) the CT physical imaging geometryconstraint is not comprehensively incorporated into deep network structures; 2)the entire framework has weak interpretability for the specific MAR task;hence, the role of each network module is difficult to be evaluated. Toalleviate these issues, in the paper, we construct a novel deep unfolding dualdomain network, termed InDuDoNet+, into which CT imaging process is finelyembedded. Concretely, we derive a joint spatial and Radon domain reconstructionmodel and propose an optimization algorithm with only simple operators forsolving it. By unfolding the iterative steps involved in the proposed algorithminto the corresponding network modules, we easily build the InDuDoNet+ withclear interpretability. Furthermore, we analyze the CT values among differenttissues, and merge the prior observations into a prior network for ourInDuDoNet+, which significantly improve its generalizationperformance.Comprehensive experiments on synthesized data and clinical datasubstantiate the superiority of the proposed methods as well as the superiorgeneralization performance beyond the current state-of-the-art (SOTA) MARmethods. . Code is available at\url{}.