DEMO: A Flexible Deartifacting Module for Compressed Sensing MRI

Yilang Zhang, Xiaojun Mao, Jian Wang, Weidong Liu

Abstract: Compressed sensing (CS) has been a novel technique for fast reconstruction of magnetic resonance (MR) images from their partial <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$k$</tex-math></inline-formula> -space measurements. However, the quality of the reconstructed images could be severely affected by artifacts from various sources. In this paper, we propose a deartifacting module (DEMO) that can effectively remove the artifacts by eliminating sparse outliers in the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$k$</tex-math></inline-formula> -space. Specifically, DEMO augments the measurements in the original loss function to approximate a new loss that is robust to outliers. Since DEMO is developed independently of any backbone algorithm to perform with, it can be flexibly incorporated into a broad range of CS-MRI methods, including both model-based methods and unrolling deep neural networks. Extensive experiments under various settings demonstrate the effectiveness and robustness of DEMO.