Artery/Vein Classification of Retinal Vasculature

Abstract

Blood vessels in the human retina are the only non-invasive window of human blood circulatory system. In recent years, researchers have found that changes in arteries and/or veins in the retinal vasculature are associated with various systemic diseases, such as hypertension, diabetes, cardiovascular or cerebral disorders. Classification of retinal blood vessels into arteries and veins is the prerequisite for the assessment of vascular changes for automatic detection of particular systemic disease. An ensemble classification based approach has been employed in this study to accurately discriminate between arteries and veins in retinal vasculature. The methodology is evaluated on a publically available dataset CHASE_DB1. It consists of 387 retinal vessels (193 arterioles, 194 veins) from 28 retinal images of multi-ethnic school going children in England. A comparative analysis of different classifiers is performed and showed that ensemble based classifier offers preeminent accuracy and used for analysis in this study.