Grading of Diabetic Retinopathy Using Machine Learning Techniques

Abstract

AbstractDiabetic retinopathy (DR) is a vision-threatening eye disease caused by blood vessel damage. Diabetes patients are commonly affected by DR, and early detection is essential to avoid vision loss. The proposed system uses Indian diabetic retinopathy image dataset (IDRiD) and enhances it using Partial Differential Equation (PDE). Morphological operations are used to detect lesions like microaneurysms, exudates, and haemorrhages, and the clinical features include the area of blood vessels, area and count of the lesions are extracted. The Grey Level Co-occurrence Matrix is used to extract statistical features (GLCM). The extracted 7 clinical and 11 statistical features are fed into machine learning classifiers such as feed forward neural network (FFNN), support vector machine (SVM), and k-nearest neural network (KNN) for DR classification with two and five classes. The performance metrics sensitivity, specificity, accuracy, negative predictive value, and positive predictive value are calculated, and the accuracy obtained for two class classification for FFNN, SVM, and KNN are 95, 95, and 90%.KeywordsDiabetic retinopathyPartial differential equationExudatesMicroaneurysmsHaemorrhagesFeed forward neural network