Robust Detection of Hard Exudates for Diagnosis of Non-proliferative Diabetic Retinopathy Using Integrated Approach

Abstract

Diabetic retinopathy is seen in an individual suffering from diabetes for a long time. It is one of the main driving reasons for blindness. The severity of the disease will increase if it is not detected in the early stages. There will be formation of lesions in retina due to diabetes. Exudates are formed when the lipid is leaked from the retinal blood vessels. Exudates appear in the moderate stage of Diabetic retinopathy. Detection of exudates is an important step in the diagnosis of Non-proliferative diabetic retinopathy. The detection can be achieved by using the image processing algorithms. This paper deals with the usage of modified morphology algorithm which addresses the interference of the optic disc prior to the detection of exudates. The features of the detected exudates are then extracted and are fed to Support Vector machine to get the result. A Graphical user interface is created which helps in ease of usage of the computer aided algorithm. The sensitivity, specificity and accuracy achieved are 92.40%, 100% and 95.16% respectively for retinal images from standard databases and hospitals.