$MC^2$-Net

Abstract

Motions in MRI cause motion artifacts in MR images. Since motion artifacts degrade image qualities and hinder an accurate diagnosis, the artifacts are subject to be corrected. However, previous conventional motion correction methods have a lot of limitations for real-world applications. To overcome these issues, in this paper, we propose a motion correction method using deep learning. Especially, in brain MRI, it is clinically common to apply a multi-contrast imaging. Motivated by this environment, we compose a neural network to correct motion artifacts in a single contrast image with helps of motion-less images from other contrasts. In results, the multi-contrast motion correction network called $MC^2$-Net outperforms other single-contrast motion correction networks and multi-contrast imputation network for both simulated data and in-vivo data.