Related Experiment Video
Updated: May 21, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
448
Hybrid CNN-Mamba model for multi-scale fundus image enhancement
Xiaopeng Wang1, Di Gong2, Yi Chen2
1Academy of Artificial Intelligence, Beijing Institute of Petrochemical Technology, Beijing 102617, China.
Biomedical Optics Express
|March 20, 2025
Summary
A novel multi-scale approach enhances fundus images using Convolutional Neural Networks (CNN) and Mamba. This method significantly improves image quality and diagnostic accuracy, especially for high-resolution retinal images.
Area of Science:
- Medical Imaging
- Computer Vision
- Artificial Intelligence
Background:
- Fundus image quality is crucial for diagnosing eye diseases.
- Existing enhancement methods struggle with multi-scale features and high resolutions.
- Deep learning models offer potential for advanced image enhancement.
Purpose of the Study:
- To develop and evaluate a multi-scale fundus image enhancement method.
- To combine Convolutional Neural Networks (CNN) with Mamba for improved performance.
- To assess the model's effectiveness across various resolutions and its impact on diagnostic tasks.
Main Methods:
- A novel multi-scale architecture integrating CNN and Mamba was proposed.
- The model was trained and evaluated on public fundus image datasets.
- Performance was quantified using metrics like FID, KID, PSNR, SSIM, VSD, and IOU.
- Evaluations included assessments at different image resolutions, including 1024x1024.
Main Results:
- The proposed CNN-Mamba model demonstrated superior performance over existing benchmarks.
- Lowest FID and KID scores, highest PSNR and SSIM values were achieved.
- Performance consistently improved with increasing image resolution, peaking at 1024x1024.
- Excellent structural preservation and high VSD/IOU scores in segmentation tasks were observed.
Conclusions:
- The multi-scale CNN-Mamba approach is highly effective for fundus image enhancement.
- The model exhibits excellent scale generalizability and structural preservation.
- This technique offers a valuable tool for improving diagnostic accuracy in ophthalmology.
