Keywords: Diffusion models, Regularization by denoising (RED), MRI reconstruction
TL;DR: We propose regularization by denoising diffusion processes for MRI reconstruction, which improves reconstruction quality with 3x faster inference.
Abstract: Diffusion models have recently delivered state-of-the-art performance for MRI reconstruction with improved robustness. However, these models still fail when there is a large distribution shift, and their long inference times impede their clinical utility. In this paper, we present regularization by denoising diffusion processes for MRI reconstruction (RED-diff). RED-diff formulates sampling as stochastic optimization, and outperforms diffusion baselines in PSNR/SSIM with 3x faster inference while using the same amount of memory.