Universal MR Image Restoration with Diffusion Models as Plug-and-Play Priors

Motivation: An all-in-one universal MR image restoration framework can reduce scan times while preserving image resolution and signal-to-noise ratio (SNR) across various clinical and technical scenarios. Goal(s): Combine the traditional plug-and-play (PnP) method with the diffusion sampling framework to restore complex MRI data accurately and robustly with a reasonable inference time. Approach: A powerful MRI model is trained on diverse and extensive complex-valued MRI datasets and then integrated into the PnP method for universal MR image restoration tasks. Results: Experimental findings indicate that our method provides accurate reconstructions for different MR inverse problems and demonstrates improved generalizability to cases outside the training data distribution. Impact: The Proposed Diffusion PnP method enables fast and accurate MRI reconstructions using a pre-trained diffusion prior, without the need for fine-tuning or retraining. This approach demonstrates strong potential for diverse clinical applications in MRI.