RFARN: Retinal vessel segmentation based on reverse fusion attention residual network
Wenhuan Liu1, Yun Jiang1, Jingyao Zhang1
1College of Computer Science and Engineering, Northwest Normal University, Lanzhou Gansu, China.
Plos One
|December 3, 2021
Summary
This study introduces an enhanced framework for retinal vascular segmentation using multiscale retinex with color restoration (MSRCR) and a novel Reverse Fusion Attention Residual Network (RFARN). The method significantly improves the accuracy and integrity of blood vessel segmentation in fundus images.
Area of Science:
- Medical Imaging
- Computer Vision
- Ophthalmology
Background:
- Accurate retinal vessel segmentation is crucial for diagnosing and treating ocular diseases.
- Challenges include poor image contrast, inhomogeneous backgrounds, and complex vascular structures in fundus images.
Purpose of the Study:
- To propose an effective framework for enhanced retinal vascular segmentation.
- To improve the accuracy and integrity of segmented retinal blood vessels.
Main Methods:
- Image enhancement using multiscale retinex with color restoration (MSRCR) to suppress noise and highlight vessels.
- Utilizing a Reverse Fusion Attention Residual Network (RFARN) with Reverse Channel Attention Module (RCAM) and Reverse Spatial Attention Module (RSAM) for segmentation.
- RFARN fuses deep local features with shallow global features for vessel continuity.
Main Results:
- Achieved high performance across DRIVE, STARE, and CHASE datasets with accuracy (Acc) up to 0.9822, sensitivity (Se) up to 0.8874, specificity (Sp) up to 0.9891, AUC up to 0.9952, and F1-Score up to 0.8707.
- Demonstrated superior vessel segmentation performance compared to existing methods like UNet, R2UNet, and FANet.
Conclusions:
- The proposed framework effectively enhances retinal vascular segmentation.
- The RFARN model, incorporating attention mechanisms, ensures accurate and continuous vessel segmentation, outperforming current state-of-the-art methods.
