Classification of retinal blood vessels into arteries and veins using CNN and likelihood propagation

Abstract

Retinal Vasculature's altered artery and vein tree structure serves as a clear indicator of many health issues. Therefore, dividing blood vessels into arteries and veins is a crucial and necessary stage in the analysis and diagnosis of many disorders. This study presents a method for classifying arteries and veins in fundus pictures using CNN and likelihood propagation. To improve performance, the suggested strategy combines deep learning and graph analysis techniques. In this procedure, initial labelling is done using CNN, and then the labels are refined using likelihood propagation. The results are compatible with modern clustering or graph-based approaches. The proposed methodology attempts to classify major as well minor vessels and achieves an average value of sensitivity of 0.899 by retaining the value of specificity 0.911 and accuracy 0.914 for VICAR dataset images.