PROJECT TITLE :

Automated Method for Retinal Artery Vein Separation via Graph Search Metaheuristic Approach

ABSTRACT:

Identifying retinal biomarkers linked with systemic and neurodegenerative illnesses requires the separation of the vascular tree into arteries and veins. Using a graph search metaheuristic, we show how to automatically separate arteries and veins from colour fundus images. Vascular subtrees are labelled arteries and veins in our method based on local information that is combined with global information. Vessel network graphs are generated using binary vessel maps, which depict topological and geographic connection of the vessels. Vascular trees are divided into subtrees based on the anatomical distinctiveness of vessel crossing and branching locations. Based on hand-crafted characteristics trained with a Random Forest classifier, subtrees of vessels are labelled with A/V. Retinal datasets from AV-DRIVE and CT-DRIVE have been used to test the proposed method's accuracy with an average accuracy of 94.7 per cent, 93.2 per cent; 96 per cent; and 90.2 per cent accordingly among the four datasets. As can be seen from the data, our proposed strategy outperforms current best practises in A/V separation.