Multi-Level Attention Network for Retinal Vessel Segmentation.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Accurate retinal vessel segmentation is crucial for diagnosing cardiovascular and ophthalmologic diseases.
• Challenges include limited annotated data, varying vessel sizes, and complex structures.

Purpose of the Study:

• To propose a novel deep learning model, AACA-MLA-D-UNet, for accurate retinal vessel segmentation.
• To address the challenges of low-level detail utilization and feature integration in U-Net architectures.

Main Methods:

• Developed AACA-MLA-D-UNet based on U-Net architecture.
• Incorporated dropout dense blocks to preserve vessel information and prevent overfitting.
• Integrated an adaptive atrous channel attention module in the contracting path.
• Utilized a multi-level attention module in the expanding path to refine features.

Main Results:

• Validated on DRIVE, STARE, and CHASE_DB1 datasets.
• Achieved superior or comparable performance in retinal vessel segmentation.
• Demonstrated lower model complexity compared to existing methods.
• Showcased effectiveness in challenging cases and strong generalization ability.

Conclusions:

• The AACA-MLA-D-UNet model offers an effective solution for retinal vessel segmentation.
• The model's design enhances feature utilization and robustness.
• It holds potential for improved screening and diagnosis of related diseases.