Vessels Segmentation In Diabetic Retinopathy By Adaptive Median Thresholding

Abstract

Diabetic retinopathy (DR) causes blindness to millions in the world. To prevent the blindness from DR, the diabetes patients are referred to eye specialists. But a significant number of patients (approximately 60-70%) do not have DR and a labour intensive, time-consuming process involves a dozen or so screener's. Automating this task, reduce the number of retinal images that the clinician needs to review, by more than 60%. This paper proposes, a user friendly MATLAB based graphical user interface (GUI) that segments the blood vessels using adaptive median thresholding. From the segmented image, features of blood vessels such as area, mean, standard deviation, energy and histogram are calculated, for classification of image as normal or abnormal. Also accuracy, specificity and sensitivity are calculated with respective to ground truth, for the performance evaluation. The GUI is implemented satisfactorily using MATLAB and the feature parameters are calculated. The average accuracy, average specificity and average sensitivity for 40 drive images found, is 0.91, 0.96 and 0.70 respectively for adaptive median thresholding.