Related Experiment Video
Updated: Jun 30, 2025
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
396
A retinal vessel segmentation network with multiple-dimension attention and adaptive feature fusion
Jianyong Li1, Ge Gao2, Lei Yang2
1College of Computer and Communication Engineering, Zhengzhou University of Light Industry, Zhengzhou, Henan Province, 450002, China.
Computers in Biology and Medicine
|March 19, 2024
Summary
This study introduces a novel deep learning network for precise retinal vessel segmentation in fundus images. The method improves accuracy in detecting thin vessels, crucial for diagnosing blinding eye diseases.
Area of Science:
- Ophthalmology and Medical Imaging
- Artificial Intelligence in Healthcare
Background:
- Accurate segmentation of retinal structures in fundus images is vital for detecting blinding eye diseases.
- Existing segmentation methods struggle with accurately segmenting thin retinal vessels.
- Deep learning models often lack detailed edge information due to repetitive convolution and pooling, limiting segmentation accuracy.
Purpose of the Study:
- To develop a pixel-level retinal vessel segmentation network that overcomes limitations of existing methods.
- To enhance the representation of local edge information and global contexts for improved segmentation.
- To effectively fuse features from different decoding stages for superior performance.
Main Methods:
- Proposed a novel deep learning network for pixel-level retinal vessel segmentation.
- Introduced a Multiple Dimension Attention Enhancement (MDAE) block for local edge information acquisition.
- Developed Deep Guidance Fusion (DGF) and Cross-Pooling Semantic Enhancement (CPSE) blocks for global context.
- Implemented an Adaptive Weight Learner (AWL) unit for optimal feature fusion.
Main Results:
- The proposed network significantly enhances retinal blood vessel segmentation performance.
- Achieved high Area Under the Curve (AUC) scores: 98.30% (DRIVE), 98.75% (CHASE_DB1), and 98.71% (STARE).
- Exceeded 83% F1 score on all three public fundus image datasets.
Conclusions:
- The developed network effectively segments retinal blood vessels, improving upon existing deep learning approaches.
- The integration of attention mechanisms and adaptive fusion strategies leads to enhanced segmentation accuracy.
- This method holds promise for improved clinical detection and diagnosis of eye diseases related to retinal morphology changes.
