Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Deform-Mamba Network for MRI Super-Resolution

Zexin Ji, Beiji Zou, Xiaoyan Kui, Pierre Vera, Su Ruan

TL;DR

The paper tackles MRI super-resolution under limited scanning resources by introducing Deform-Mamba, a multi-branch network that fuses content-adaptive deformable locality with a linear-complexity global model via the vision Mamba backbone. It combines a Deform-Mamba encoder, a multi-view context bottleneck, and a vision Mamba decoder, guided by a contrastive edge loss to better recover edges and local contrast. Empirical results on IXI and fastMRI demonstrate competitive PSNR/SSIM, with strong performance on fastMRI 4× and favorable efficiency due to the linear-complexity design. The work offers a practical CNN/Transformer alternative for MR SR and suggests future integration with diffusion-model-based SR.

Abstract

In this paper, we propose a new architecture, called Deform-Mamba, for MR image super-resolution. Unlike conventional CNN or Transformer-based super-resolution approaches which encounter challenges related to the local respective field or heavy computational cost, our approach aims to effectively explore the local and global information of images. Specifically, we develop a Deform-Mamba encoder which is composed of two branches, modulated deform block and vision Mamba block. We also design a multi-view context module in the bottleneck layer to explore the multi-view contextual content. Thanks to the extracted features of the encoder, which include content-adaptive local and efficient global information, the vision Mamba decoder finally generates high-quality MR images. Moreover, we introduce a contrastive edge loss to promote the reconstruction of edge and contrast related content. Quantitative and qualitative experimental results indicate that our approach on IXI and fastMRI datasets achieves competitive performance.

Deform-Mamba Network for MRI Super-Resolution

TL;DR

The paper tackles MRI super-resolution under limited scanning resources by introducing Deform-Mamba, a multi-branch network that fuses content-adaptive deformable locality with a linear-complexity global model via the vision Mamba backbone. It combines a Deform-Mamba encoder, a multi-view context bottleneck, and a vision Mamba decoder, guided by a contrastive edge loss to better recover edges and local contrast. Empirical results on IXI and fastMRI demonstrate competitive PSNR/SSIM, with strong performance on fastMRI 4× and favorable efficiency due to the linear-complexity design. The work offers a practical CNN/Transformer alternative for MR SR and suggests future integration with diffusion-model-based SR.

Abstract

In this paper, we propose a new architecture, called Deform-Mamba, for MR image super-resolution. Unlike conventional CNN or Transformer-based super-resolution approaches which encounter challenges related to the local respective field or heavy computational cost, our approach aims to effectively explore the local and global information of images. Specifically, we develop a Deform-Mamba encoder which is composed of two branches, modulated deform block and vision Mamba block. We also design a multi-view context module in the bottleneck layer to explore the multi-view contextual content. Thanks to the extracted features of the encoder, which include content-adaptive local and efficient global information, the vision Mamba decoder finally generates high-quality MR images. Moreover, we introduce a contrastive edge loss to promote the reconstruction of edge and contrast related content. Quantitative and qualitative experimental results indicate that our approach on IXI and fastMRI datasets achieves competitive performance.
Paper Structure (10 sections, 7 equations, 2 figures, 2 tables)

This paper contains 10 sections, 7 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Preliminaries
  4. Proposed Architecture
  5. Experiments
  6. Datasets and Evaluation Metrics
  7. Experimental Details
  8. Ablation Study
  9. Comparison with state-of-the-art methods
  10. Conclusion and Future Work

Figures (2)

  • Figure 1: Architecture of our Deform-Mamba. (A) The overall architecture mainly consists of a Deform-Mamba encoder, a multi-view context module, and a vision Mamba decoder. (B) An implementation of the Deform-Mamba module includes (B-1) modulated deform block, (B-2) vision Mamba block, and (B-3) 2D-Selective-Scan (SS2D) block. (C) Multi-view context block. (D) Contrastive edge loss (CELoss).
  • Figure 2: Qualitative results with different methods under fastMRI and IXI dataset