Automated Retinal Blood Vessel Feature Extraction in Digital Fundus Images

Abstract

The retinal microvascular network manifests the well-being of other systems and organs as they are structurally and physiologically similar. It offers a unique window to assess numerous disorders such as hypertension, heart disease and nervous system illnesses. However, manually analysing retinal blood vessels in digital fundus images is challenging. In addition, the low contrast images limit the diagnosis of retinal blood vessel-related eye diseases. Thus, this work uses the digital image processing approach to automate the extraction and selection of significant blood vessel features, i.e., the width and pixel intensity of the artery and vein. The digital fundus images are collected from the Digital Retinal Images for Vessel Extraction (DRIVE) database, consisting of twenty 584×565-pixel digital fundus and ground truth images. The proposed method automatically extracts the retinal width and intensity based on the identified coordinates of the blood vessel's skeleton images. Using a one-way ANOVA statistical test computation, we found that the width and the green channel intensity pixel are significant features (p-value <0.005) that can be used to differentiate artery and vein in digital fundus images.