Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Deep Unfolding Convolutional Dictionary Model for Multi-Contrast MRI Super-resolution and Reconstruction

Pengcheng Lei, Faming Fang, Guixu Zhang, Ming Xu

TL;DR

This work tackles the problem of accelerating MRI by leveraging multi-contrast information for SR and reconstruction with interpretability. It introduces MC-CDic, a deep unfolding network built from a convolutional dictionary learning formulation that explicitly separates common and unique features across contrasts and uses a joint observation model. The method integrates multi-scale dictionaries and learnable proximal networks to replace regularizers, achieving state-of-the-art performance on guided-SR and guided reconstruction while maintaining interpretability. The approach offers practical impact by improving reconstruction quality and providing a principled framework that can extend to other multi-modal imaging tasks.

Abstract

Magnetic resonance imaging (MRI) tasks often involve multiple contrasts. Recently, numerous deep learning-based multi-contrast MRI super-resolution (SR) and reconstruction methods have been proposed to explore the complementary information from the multi-contrast images. However, these methods either construct parameter-sharing networks or manually design fusion rules, failing to accurately model the correlations between multi-contrast images and lacking certain interpretations. In this paper, we propose a multi-contrast convolutional dictionary (MC-CDic) model under the guidance of the optimization algorithm with a well-designed data fidelity term. Specifically, we bulid an observation model for the multi-contrast MR images to explicitly model the multi-contrast images as common features and unique features. In this way, only the useful information in the reference image can be transferred to the target image, while the inconsistent information will be ignored. We employ the proximal gradient algorithm to optimize the model and unroll the iterative steps into a deep CDic model. Especially, the proximal operators are replaced by learnable ResNet. In addition, multi-scale dictionaries are introduced to further improve the model performance. We test our MC-CDic model on multi-contrast MRI SR and reconstruction tasks. Experimental results demonstrate the superior performance of the proposed MC-CDic model against existing SOTA methods. Code is available at https://github.com/lpcccc-cv/MC-CDic.

Deep Unfolding Convolutional Dictionary Model for Multi-Contrast MRI Super-resolution and Reconstruction

TL;DR

This work tackles the problem of accelerating MRI by leveraging multi-contrast information for SR and reconstruction with interpretability. It introduces MC-CDic, a deep unfolding network built from a convolutional dictionary learning formulation that explicitly separates common and unique features across contrasts and uses a joint observation model. The method integrates multi-scale dictionaries and learnable proximal networks to replace regularizers, achieving state-of-the-art performance on guided-SR and guided reconstruction while maintaining interpretability. The approach offers practical impact by improving reconstruction quality and providing a principled framework that can extend to other multi-modal imaging tasks.

Abstract

Magnetic resonance imaging (MRI) tasks often involve multiple contrasts. Recently, numerous deep learning-based multi-contrast MRI super-resolution (SR) and reconstruction methods have been proposed to explore the complementary information from the multi-contrast images. However, these methods either construct parameter-sharing networks or manually design fusion rules, failing to accurately model the correlations between multi-contrast images and lacking certain interpretations. In this paper, we propose a multi-contrast convolutional dictionary (MC-CDic) model under the guidance of the optimization algorithm with a well-designed data fidelity term. Specifically, we bulid an observation model for the multi-contrast MR images to explicitly model the multi-contrast images as common features and unique features. In this way, only the useful information in the reference image can be transferred to the target image, while the inconsistent information will be ignored. We employ the proximal gradient algorithm to optimize the model and unroll the iterative steps into a deep CDic model. Especially, the proximal operators are replaced by learnable ResNet. In addition, multi-scale dictionaries are introduced to further improve the model performance. We test our MC-CDic model on multi-contrast MRI SR and reconstruction tasks. Experimental results demonstrate the superior performance of the proposed MC-CDic model against existing SOTA methods. Code is available at https://github.com/lpcccc-cv/MC-CDic.
Paper Structure (27 sections, 29 equations, 8 figures, 3 tables)

This paper contains 27 sections, 29 equations, 8 figures, 3 tables.

Table of Contents

  1. Introduction
  2. MC-CDic Model for MR Images
  3. Model Formulation
  4. Optimization Algorithm
  5. Updating $U$.
  6. Updating $V$.
  7. Updating $C$.
  8. Deep Unfolding MC-CDic Model
  9. Network Design
  10. (1) Feature Initialization Module.
  11. (2) MC-CDic Iteration Module.
  12. (3) Reconstruction Module.
  13. Multi-Scale Dictionaries
  14. Loss Function
  15. Experiments
  16. ...and 12 more sections

Figures (8)

  • Figure 1: The visual comparison of the multi-contrast images, i.e. T2 and PD, on IXI dataset. They share some common textures (in green), and they also have their unique contrast information and some inconsistent structures (in red).
  • Figure 2: The overall structure of the proposed multi-contrast convolutional dictionary (MC-CDic) model for the guided-SR task.
  • Figure 3: The structure of the proposed multi-scale dictionaries (for updating $U^t$). (a) multi-scale CDic encoder, (b) multi-scale CDic decoder, and (c) multi-scale proximal network. $U_t=\{U_0^{t},U_1^{t},U_2^{t}\}$ contains a list of multi-scale representations.
  • Figure 4: Error maps of different single-contrast SR (the first three methods) and guided-SR (the next six methods) with the scale factor of $\times4$ on the IXI (PD guides T2) and BrainTS (T1 guides T2) testing sets.
  • Figure 5: Visualization of the decomposed common and unique components of our MC-CDic model on the guided-SR task.
  • ...and 3 more figures