A Comprehensive Study of Retinal Vessel Classification Methods in Fundus Images

Area of Science:

• Ophthalmology
• Medical Imaging
• Biomedical Engineering

Background:

• Changes in retinal vessel structure are linked to adult cardiovascular diseases and infant retinopathy of prematurity.
• Retinal fundus images offer non-invasive visualization of vasculature, enabling early disease diagnosis.
• The arteriolar-venular ratio, calculated from fundus images, is a key indicator for hypertension, diabetes, and cardiovascular diseases.

Purpose of the Study:

• To present a comprehensive review of existing methods for classifying retinal arteries and veins in fundus images.
• To address the challenge of comparing diverse classification methods due to varied datasets and evaluation criteria.
• To analyze classification methods from a modeling perspective to identify areas for improvement.

Main Methods:

• A comprehensive review of proposed methods for artery and vein classification in retinal fundus images.
• Evaluation of classification methods based on their underlying modeling approaches.
• Analysis of the focus on statistical and geometric models versus transform domain models.

Main Results:

• Most existing methods primarily utilize statistical and geometric models in the spatial domain.
• Transform domain models, particularly data-adaptive ones, have received less attention in current approaches.
• A fair performance comparison of existing methods is hindered by differing datasets and evaluation metrics.

Conclusions:

• Accurate artery and vein classification is essential for reliable measurement of the arteriolar-venular ratio and disease diagnosis.
• The current focus on statistical and geometric models may limit the precision of retinal vessel classification.
• Future research should explore the potential of transform domain models, especially data-adaptive techniques, for enhanced fundus image classification.