Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Versatile and Efficient Medical Image Super-Resolution Via Frequency-Gated Mamba

Wenfeng Huang, Xiangyun Liao, Wei Cao, Wenjing Jia, Weixin Si

TL;DR

FGMamba tackles medical image SR by unifying global dependency modeling with high-frequency detail restoration in a lightweight framework. It introduces the Gated Attention-enhanced State-Space Module (GASM) and the Pyramid Frequency Fusion Module (PFFM) to achieve efficient long-range context capture and multiscale high-frequency fusion, respectively, while constraining the parameter count to under $<0.75M$. Across five modalities (ultrasound, OCT, MRI, CT, endoscopy), it outperforms CNN-, Transformer-, and prior Mamba-based SR methods in PSNR/SSIM, demonstrating strong generalization and a practical footprint for clinical deployment. The results validate frequency-aware state-space modeling as a scalable and accurate approach for medical image enhancement with potential impact on downstream tasks like segmentation and diagnosis.

Abstract

Medical image super-resolution (SR) is essential for enhancing diagnostic accuracy while reducing acquisition cost and scanning time. However, modeling both long-range anatomical structures and fine-grained frequency details with low computational overhead remains challenging. We propose FGMamba, a novel frequency-aware gated state-space model that unifies global dependency modeling and fine-detail enhancement into a lightweight architecture. Our method introduces two key innovations: a Gated Attention-enhanced State-Space Module (GASM) that integrates efficient state-space modeling with dual-branch spatial and channel attention, and a Pyramid Frequency Fusion Module (PFFM) that captures high-frequency details across multiple resolutions via FFT-guided fusion. Extensive evaluations across five medical imaging modalities (Ultrasound, OCT, MRI, CT, and Endoscopic) demonstrate that FGMamba achieves superior PSNR/SSIM while maintaining a compact parameter footprint ($<$0.75M), outperforming CNN-based and Transformer-based SOTAs. Our results validate the effectiveness of frequency-aware state-space modeling for scalable and accurate medical image enhancement.

Versatile and Efficient Medical Image Super-Resolution Via Frequency-Gated Mamba

TL;DR

FGMamba tackles medical image SR by unifying global dependency modeling with high-frequency detail restoration in a lightweight framework. It introduces the Gated Attention-enhanced State-Space Module (GASM) and the Pyramid Frequency Fusion Module (PFFM) to achieve efficient long-range context capture and multiscale high-frequency fusion, respectively, while constraining the parameter count to under . Across five modalities (ultrasound, OCT, MRI, CT, endoscopy), it outperforms CNN-, Transformer-, and prior Mamba-based SR methods in PSNR/SSIM, demonstrating strong generalization and a practical footprint for clinical deployment. The results validate frequency-aware state-space modeling as a scalable and accurate approach for medical image enhancement with potential impact on downstream tasks like segmentation and diagnosis.

Abstract

Medical image super-resolution (SR) is essential for enhancing diagnostic accuracy while reducing acquisition cost and scanning time. However, modeling both long-range anatomical structures and fine-grained frequency details with low computational overhead remains challenging. We propose FGMamba, a novel frequency-aware gated state-space model that unifies global dependency modeling and fine-detail enhancement into a lightweight architecture. Our method introduces two key innovations: a Gated Attention-enhanced State-Space Module (GASM) that integrates efficient state-space modeling with dual-branch spatial and channel attention, and a Pyramid Frequency Fusion Module (PFFM) that captures high-frequency details across multiple resolutions via FFT-guided fusion. Extensive evaluations across five medical imaging modalities (Ultrasound, OCT, MRI, CT, and Endoscopic) demonstrate that FGMamba achieves superior PSNR/SSIM while maintaining a compact parameter footprint (0.75M), outperforming CNN-based and Transformer-based SOTAs. Our results validate the effectiveness of frequency-aware state-space modeling for scalable and accurate medical image enhancement.

Paper Structure

This paper contains 18 sections, 14 equations, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Image Super-Resolution
  4. Mamba Architectures
  5. Method
  6. FGBlock
  7. Experiments
  8. Datasets
  9. Ultrasound Image Dataset
  10. OCT Image Dataset
  11. Endoscope Image Dataset
  12. CT Image Dataset
  13. MRI Image Dataset
  14. Impletement details
  15. Comparison with the State of the Arts
  16. ...and 3 more sections

Figures (2)

  • Figure 1: Overall architecture of the proposed FGMamba. It consists of an initial convolution, several FGBlocks, and a reconstruction module with pixel-shuffle upsampling. Each FGBlock contains multiple GASM (Gated Attention State Space Modules), a frequency-enhancing PFFM (Pyramid Frequency Fusion Module), and additional Mamba residual connections. The submodules are illustrated below: (1) GASM incorporates VSSM2D with gated spatial/channel attention, (2) GAU enhances feature selection via dual attention gating, and (3) PFFM extracts and fuses high-frequency components across multiple scales via FFT-based filtering and residual learning.
  • Figure 2: Visual comparison across five medical modalities: (a) Ultrasound ($\times$2), (b) CT ($\times$4), (c) OCT ($\times$4), (d) Endoscopic ($\times$4), and (e) MRI ($\times$4). Red boxes highlight diagnostically critical structures (vessels, lesions, tissue textures), where FGMamba achieves sharper, more detailed reconstructions to aid radiologist diagnosis and clinical decision-making.