JoVE - Journal of Visualized Experiments
JoVE Visualize
Contact Us

CNN-Transformer gated fusion network for medical image super-resolution

  • 0Department of Artificial Intelligence and Data Science, Guangzhou Xinhua University, Dongguan, 523133, Guangdong, China. qinjuanjuan9806@xhsysu.edu.cn.
Scientific Reports +

|

May 2, 2025

|

May 2, 2025

View abstract on PubMed

Summary

This summary is machine-generated.

This study introduces a novel dual-branch network (CTGFSR) for medical image super-resolution, enhancing detail and global information utilization. CTGFSR improves image quality and detail restoration compared to existing methods.

Area Of Science

  • Medical imaging
  • Artificial intelligence
  • Computer vision

Background

  • Existing medical image super-resolution methods struggle with detail blurring and global information.
  • There's a need for advanced algorithms to improve the quality and diagnostic value of medical images.

Purpose Of The Study

  • To propose a novel dual-branch fusion network, CTGFSR, for enhanced medical image super-resolution reconstruction.
  • To address limitations in detail blurring and global information utilization in current super-resolution techniques.

Main Methods

  • A dual-branch network combining a residual Transformer network (global branch) and a dynamic convolutional neural network (local branch).
  • The global branch leverages self-attention for large-scale information, while the local branch uses dynamic convolution for multi-scale feature extraction.
  • Residual skip connections and a bidirectional gated attention mechanism are employed for detail preservation and branch fusion.

Main Results

  • CTGFSR demonstrates superior overall performance on ACDC (abdominal MR) and L2R2022 (lung CT) datasets.
  • Significant improvements in Structural Similarity Index (SSIM) and Peak Signal-to-Noise Ratio (PSNR) were observed at 2x and 4x magnification.
  • Outperformed mainstream super-resolution algorithms including CFIPC, PDCNCF, ESPCN, FSRCNN, VDSR, ESRT, and SwinIR.

Conclusions

  • The proposed CTGFSR network effectively enhances medical image super-resolution reconstruction.
  • The dual-branch architecture successfully integrates global context and local details, improving image quality and diagnostic potential.
  • CTGFSR offers a promising advancement in medical image processing for improved clinical applications.

Keywords: