Unified Motion Correction Model for Multi-modal MRI

Motivation: Motion artifacts in multi-modal MRI impair image quality and diagnostic accuracy. Traditional methods handle specific image types but struggle with contrast variations and differing motion patterns [1]. Goal(s): We aim to develop a unified framework to effectively correct motion artifacts in multi-modal MRI. Approach: We pretrain a CLIP model using MRI volumes and metadata, extracting features that decouple contrast and retain content. These features feed a Transformer to predict motion degradation scores [2]. A Mixture of Experts model allocates weights to specialized networks for final correction [5]. Results: Our method significantly reduces motion artifacts across various MRI, outperforming existing techniques. Impact: This framework enhances multi-modal MRI by effectively correcting motion artifacts, leading to improved image quality and diagnostic confidence. It holds potential for widespread clinical adoption, benefiting patient care and advancing research involving diverse MRI.