Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Self-Supervised MRI Reconstruction with Unrolled Diffusion Models

Yilmaz Korkmaz, Tolga Cukur, Vishal M. Patel

TL;DR

This work addresses the slow acquisition in MRI by introducing SSDiffRecon, a self-supervised diffusion-based reconstruction that unrolls a conditional diffusion process with cross-attention transformers and data-consistency blocks. It trains end-to-end from undersampled k-space using a k-space masking strategy, eliminating the need for fully-sampled targets. Empirically, SSDiffRecon matches supervised baselines while offering faster inference than prior diffusion approaches and surpasses self-supervised methods in both speed and fidelity. The approach offers a practical, data-efficient path toward high-quality, accelerated MRI reconstructions in clinical settings.

Abstract

Magnetic Resonance Imaging (MRI) produces excellent soft tissue contrast, albeit it is an inherently slow imaging modality. Promising deep learning methods have recently been proposed to reconstruct accelerated MRI scans. However, existing methods still suffer from various limitations regarding image fidelity, contextual sensitivity, and reliance on fully-sampled acquisitions for model training. To comprehensively address these limitations, we propose a novel self-supervised deep reconstruction model, named Self-Supervised Diffusion Reconstruction (SSDiffRecon). SSDiffRecon expresses a conditional diffusion process as an unrolled architecture that interleaves cross-attention transformers for reverse diffusion steps with data-consistency blocks for physics-driven processing. Unlike recent diffusion methods for MRI reconstruction, a self-supervision strategy is adopted to train SSDiffRecon using only undersampled k-space data. Comprehensive experiments on public brain MR datasets demonstrates the superiority of SSDiffRecon against state-of-the-art supervised, and self-supervised baselines in terms of reconstruction speed and quality. Implementation will be available at https://github.com/yilmazkorkmaz1/SSDiffRecon.

Self-Supervised MRI Reconstruction with Unrolled Diffusion Models

TL;DR

This work addresses the slow acquisition in MRI by introducing SSDiffRecon, a self-supervised diffusion-based reconstruction that unrolls a conditional diffusion process with cross-attention transformers and data-consistency blocks. It trains end-to-end from undersampled k-space using a k-space masking strategy, eliminating the need for fully-sampled targets. Empirically, SSDiffRecon matches supervised baselines while offering faster inference than prior diffusion approaches and surpasses self-supervised methods in both speed and fidelity. The approach offers a practical, data-efficient path toward high-quality, accelerated MRI reconstructions in clinical settings.

Abstract

Magnetic Resonance Imaging (MRI) produces excellent soft tissue contrast, albeit it is an inherently slow imaging modality. Promising deep learning methods have recently been proposed to reconstruct accelerated MRI scans. However, existing methods still suffer from various limitations regarding image fidelity, contextual sensitivity, and reliance on fully-sampled acquisitions for model training. To comprehensively address these limitations, we propose a novel self-supervised deep reconstruction model, named Self-Supervised Diffusion Reconstruction (SSDiffRecon). SSDiffRecon expresses a conditional diffusion process as an unrolled architecture that interleaves cross-attention transformers for reverse diffusion steps with data-consistency blocks for physics-driven processing. Unlike recent diffusion methods for MRI reconstruction, a self-supervision strategy is adopted to train SSDiffRecon using only undersampled k-space data. Comprehensive experiments on public brain MR datasets demonstrates the superiority of SSDiffRecon against state-of-the-art supervised, and self-supervised baselines in terms of reconstruction speed and quality. Implementation will be available at https://github.com/yilmazkorkmaz1/SSDiffRecon.
Paper Structure (19 sections, 12 equations, 4 figures, 5 tables)

This paper contains 19 sections, 12 equations, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Accelerated MRI Reconstruction
  4. Denoising Diffusion Models
  5. SSDiffRecon
  6. Self-Supervised Training:
  7. Inference:
  8. Unrolled Denoising Network $R_{\theta}(.)$
  9. Mapper network
  10. Unrolled Denoising Blocks
  11. Implementation Details
  12. Experimental Results
  13. Datasets
  14. Competing Methods
  15. Experiments
  16. ...and 4 more sections

Figures (4)

  • Figure 1: Overall training scheme and network architecture. SSDiffRecon utilizes an unrolled physics-guided network as a denoiser in the diffusion process while allowing time index guidance through the Mapper Network via cross-attention transformer layers (shown in green). After two transformer layers, it performs data-consistency (shown in orange). Corresponding noisy input undersampled ($x_t^{u,p}$) and denoised reconstructed images ($\hat{x_0}^{u}$) are shown during training. $L_1$ difference between k-space points in pre-allocated locations ($M_r$) has been utilized as the loss function.
  • Figure 2: Inference scheme. We start from zero-filled reconstruction of undersampled acquisitions and inject a lightweight noise ($\epsilon_{low}$) into them while performing data consistency in the backward diffusion steps to allow a more gradual denoising.
  • Figure 3: Reconstructions of T2- weighted images from the IXI dataset, along with the zoomed-in regions on the top and the corresponding error maps underneath.
  • Figure 4: Reconstructions of T1- weighted images from fastMRI, along with the zoomed-in regions on the top and the corresponding error maps underneath.