Retinal blood vessel localization to expedite PDR diagnosis

Abstract

Ophthalmologist relies on the retinal fundus image segmentation for accurate diagnosis of Diabetic Retinopathy caused due to prolonged deterioration in retinal blood vessels. Blood vessel and optical disc localization determines the vascular alterations helpful in identifying retinal diseases with accurate identification of pathologies like microaneurysms and exudates. This work comprises evaluation of proposed Optical Disc Segmentation and blood vessel localization techniques followed by a statistical analysis using three fractal dimensions; box count, information and correlation. Fractal dimensions explored are beneficial for Proliferative Diabetic Retinopathy (PDR) diagnosis as its value for vascular structures increases with increasing level of PDR. Two benchmark fundus image databases, DRIVE and STARE were evaluated by utilizing shape and fractal features for performance validation and average accuracies of 96.79% and 95.68% were achieved for extracted blood vessels using proposed approach.