Detection of Hypertensive Retinopathy throuh Color Fundus Images

Abstract

Hypertensive retinopathy is a condition characterized by several pathological changes in the retinal vasculature in response to very high blood pressure. The present work proposes an effective methodology for the detection of hypertensive retinopathy by (i) designing an efficient method to segment blood vasculature and (ii) performing extensive measurements on the segmented blood vasculature to confirm the condition of hypertensive retinopathy. The proposed vessel segmentation method is designed by rotating top-hat transform at every 22.5° over the entire image to enhance the blood vessels irrespective of their size and direction, and afterwards iterative thresholding is applied on the combination of bit plane slices containing visually significant information related to blood vasculature of a retinal fundus image. The confirmation of hypertensive retinopathy is assured by measuring different diagnostically important performance measures, viz., arteriolar-to-venular ratio, bifurcation points and tortuosity on the segmented blood vasculature. The experimental evaluation of the proposed segmentation method on publicly available databases, viz., DRIVE and STARE shows the accurate extraction of retinal blood vasculature having proximity with the manual segmentation rates provided by the second observer with average accuracy of 94.89% and 94.56% respectively. The proposed segmentation method and related diagnostic measurements are thoroughly evaluated with four publicly available databases; DRIVE, STARE, INSPIRE-AVR and RET-TORT. The proposed segmentation method and bifurcations are analyzed on two publicly available databases DRIVE, STARE and achieved an average accuracy of 94.89% and 94.56% respectively. The AVR is measured on INSPIRE-AVR with a high accuracy of 97.44%, 98.43% with respect to expert 1 and 2. Various measurements based on curvatures are calculated on the torturous vessels using RET-TORT database. A comparison with several state-of-the-art methods shows that the proposed method represents a significant and competitive improvement in terms of quantitative performance measures.