Benchmarking Self-Supervised Methods for Accelerated MRI Reconstruction

2025-02-19

AI Summary

To address the lack of systematic evaluation of self-supervised methods for accelerated MRI reconstruction under settings without fully-sampled ground-truth data, this paper establishes the first benchmark dedicated to unpaired self-supervised MRI reconstruction, uniformly evaluating 12 mainstream unsupervised loss functions. We propose MO-EI, a multi-operator equivariant imaging framework that jointly incorporates physical forward models and equivariance constraints within a deep unrolling architecture, yielding interpretable and generalizable reconstructions. Validated via the open-source DeepInverse library on standard datasets (e.g., FastMRI), MO-EI significantly outperforms existing self-supervised approaches—achieving ≥1.8 dB PSNR gain—and approaches supervised-learning performance. Our core contributions are threefold: (i) the first comprehensive no-ground-truth benchmark for self-supervised MRI reconstruction; (ii) the MO-EI framework integrating physics-guided modeling with learned equivariant priors; and (iii) empirical validation of the efficacy of such physically informed equivariant modeling.

Technology Category

• Machine Learning: Unsupervised & Self-Supervised Learning
• Computer Vision: Multi-modal Vision
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

Reconstructing MRI from highly undersampled measurements is crucial for accelerating medical imaging, but is challenging due to the ill-posedness of the inverse problem. While supervised deep learning approaches have shown remarkable success, they rely on fully-sampled ground truth data, which is often impractical or impossible to obtain. Recently, numerous self-supervised methods have emerged that do not require ground truth, however, the lack of systematic comparison and standard experimental setups have hindered research. We present the first comprehensive review of loss functions from all feedforward self-supervised methods and the first benchmark on accelerated MRI reconstruction without ground truth, showing that there is a wide range in performance across methods. In addition, we propose Multi-Operator Equivariant Imaging (MO-EI), a novel framework that builds on the imaging model considered in existing methods to outperform all state-of-the-art and approaches supervised performance. Finally, to facilitate reproducible benchmarking, we provide implementations of all methods in the DeepInverse library (https://deepinv.github.io) and easy-to-use demo code at https://andrewwango.github.io/deepinv-selfsup-fastmri.

Problem

• Benchmarking self-supervised MRI reconstruction
• Comparing performance without ground truth
• Proposing Multi-Operator Equivariant Imaging

Innovation

• Self-supervised MRI reconstruction methods
• Multi-Operator Equivariant Imaging framework
• DeepInverse library for reproducible benchmarking