Retina Blood Vessels Segmentation and Classification with the Multi-featured Approach

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Retinal blood vessel segmentation is challenging due to small vessel irregularities and data complexity.
• Existing methods struggle with feature merging and spatial information loss, impacting faint vessel detection.
• Automated arterial and venous (A/V) classification is crucial for comprehensive retinal vessel analysis.

Purpose of the Study:

• To develop and validate an advanced model for accurate retinal blood vessel segmentation and A/V classification.
• To address challenges in detecting small, low-contrast vessels and improve overall segmentation performance.
• To enhance the resilience and perception of models in analyzing complex retinal vasculature.

Main Methods:

• Employed an ensemble filter approach with Bilateral and Laplacian edge detectors for image enhancement.
• Utilized a complete convolution network with attention operations for detailed vessel map generation.
• Incorporated an orientation map and attention module for refined blood vessel depiction and feature extraction.

Main Results:

• Achieved high accuracy across multiple datasets: 97.5% (DRIVE), 99.25% (STARE), 98.33% (INSPIRE-AVR), and 98.67% (HRF).
• Demonstrated superior performance in segmenting and classifying retinal blood vessels, including small and faint ones.
• The attention mechanism significantly improved the model's perception and resilience in segmentation tasks.

Conclusions:

• The proposed model offers a highly effective solution for accurate retinal blood vessel segmentation and A/V classification.
• The methodology successfully overcomes limitations in detecting intricate and low-contrast vascular structures.
• This approach holds significant potential for improving diagnostic capabilities in ophthalmology through advanced image analysis.