A Self-supervised Diffusion Bridge for MRI Reconstruction

TL;DR

This paper introduces SelfDB, a self-supervised diffusion bridge method for MRI reconstruction that trains directly on noisy measurements, eliminating the need for high-quality reference images and outperforming existing diffusion models.

Contribution

SelfDB is a novel self-supervised approach that enables diffusion models to be trained directly on noisy data for MRI reconstruction, broadening their applicability.

Findings

SelfDB outperforms denoising diffusion models in MRI reconstruction.

SelfDB does not require high-quality reference images for training.

SelfDB demonstrates superior performance on compressed sensing MRI.

Abstract

Diffusion bridges (DBs) are a class of diffusion models that enable faster sampling by interpolating between two paired image distributions. Training traditional DBs for image reconstruction requires high-quality reference images, which limits their applicability to settings where such references are unavailable. We propose SelfDB as a novel self-supervised method for training DBs directly on available noisy measurements without any high-quality reference images. SelfDB formulates the diffusion process by further sub-sampling the available measurements two additional times and training a neural network to reverse the corresponding degradation process by using the available measurements as the training targets. We validate SelfDB on compressed sensing MRI, showing its superior performance compared to the denoising diffusion models.