New AI method corrects MRI motion artifacts with 0.75 dB PSNR gain

Motion artifacts remain a major challenge in MRI, degrading image quality and diagnostic reliability. Existing deep learning methods are contrast-specific and fail to generalize across different MRI modalities or varying artifact severities. A team of researchers from Shanghai Jiao Tong University, Duke University, and other institutions tackles this with a parameter-informed disentanglement and adaptive experts framework. They pretrain a model called ScanCLIP on over 30,000 MRI text-image pairs to learn contrast embeddings from acquisition parameters, effectively disentangling contrast style from anatomical content. This yields contrast-free features that can be processed uniformly.

The system then employs a Vision Transformer to estimate motion severity from these features, routing them through a Mixture-of-Experts (MoE) network that applies targeted correction based on artifact severity. A dual-pathway decoder reconstructs both the clean image and a residual artifact map, enforcing image-space consistency. On IXI and HCP benchmarks, the method achieves 0.75 dB higher PSNR and up to 0.0279 higher SSIM than state-of-the-art approaches, with larger gains at higher artifact severities. Critically, it demonstrates robust zero-shot generalization on real-world clinical data acquired with unseen scanning parameters, where prior methods either fail to remove artifacts or introduce additional distortions.