Related Experiment Video
Updated: Jul 15, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.8K
CCS-UNet: a cross-channel spatial attention model for accurate retinal vessel segmentation
Yong-Fei Zhu1, Xiang Xu1, Xue-Dian Zhang1
1Shanghai Key Laboratory of Contemporary Optics System, College of Optical-Electrical and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China.
Biomedical Optics Express
|October 4, 2023
Summary
A novel deep learning model, Cross-Channel Spatial Attention U-Net (CCS-UNet), enhances retinal vessel segmentation accuracy. This method improves diagnostic aid for retinal diseases by better recognizing vascular structures in medical images.
Area of Science:
- Medical Imaging
- Computer-Assisted Diagnosis
- Deep Learning
Background:
- Accurate retinal vessel segmentation is crucial for diagnosing eye diseases.
- Existing deep learning models struggle with scale variations and complex backgrounds in retinal images.
Purpose of the Study:
- To introduce a Cross-Channel Spatial Attention U-Net (CCS-UNet) for improved retinal vessel segmentation.
- To address limitations in scale variation and background complexity in current segmentation models.
Main Methods:
- Developed CCS-UNet incorporating a ResNeSt block for diverse feature extraction.
- Implemented soft attention for cross-channel information aggregation and an attention mechanism in skip connections for enhanced feature integration.
- Utilized a Feature Fusion Module (FFM) for refined segmentation maps.
Main Results:
- CCS-UNet achieved superior segmentation performance on five benchmark datasets (DRIVE, CHASEDB1, STARE, IOSTAR, HRF).
- Achieved high global accuracy (e.g., 0.9617 on DRIVE) and AUC (e.g., 0.9863 on DRIVE).
- Ablation studies confirmed the effectiveness of individual architectural components.
Conclusions:
- The proposed CCS-UNet demonstrates significant potential for accurate retinal vessel segmentation.
- This model can serve as a valuable tool for computer-assisted diagnosis of retinal diseases.
