InDuDoNet: An Interpretable Dual Domain Network for CT Metal Artifact Reduction

For the task of metal artifact reduction (MAR), although deep learning (DL)-based methods have achieved promising performances, most them suffer from two problems: 1) CT imaging geometry constraint is not fully embedded into network during training, leaving room for further performance improvement; 2) model interpretability lack sufficient consideration. Against these issues, we propose a novel interpretable dual domain network, termed as InDuDoNet, which combines advantages model-driven and data-driven methodologies. Specifically, build joint spatial Radon reconstruction utilize proximal gradient technique to design an iterative algorithm solving it. The optimization only consists simple computational operators, facilitate us correspondingly unfold steps modules thus improve interpretablility framework. Extensive experiments on synthesized clinical data show superiority our InDuDoNet. Code available in https://github.com/hongwang01/InDuDoNet.