A Self-supervised Diffusion Bridge for MRI Reconstruction

2025-01-06

AI Summary

To address MRI reconstruction without paired high-quality reference images, this paper proposes Self-Supervised Diffusion Bridge (SelfDB), a diffusion-based method trained solely on undersampled k-space measurements—without requiring fully sampled ground-truth images. Methodologically, SelfDB constructs an invertible degradation path within the measurement domain via二次 subsampling and employs the original undersampled measurements as supervision, enabling diffusion modeling and inversion directly in k-space. Its core contribution is the first adaptation of the diffusion bridge paradigm to a fully self-supervised, measurement-domain-driven framework—eliminating reliance on fully sampled images entirely. Experiments demonstrate that, at an acceleration factor of 8, SelfDB achieves a 2.1 dB PSNR improvement over denoising diffusion models, yielding more faithful reconstructions and higher sampling efficiency. This work establishes a novel unsupervised paradigm for MRI reconstruction.

Technology Category

• Machine Learning: Unsupervised & Self-Supervised Learning
• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Content-based information diffusion

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.

Problem

• MRI image reconstruction
• diffusion model
• self-learning image processing

Innovation

• SelfDB
• MRI image reconstruction
• Compressive sensing MRI