Automatic Detection of Diabetic Retinopathy using Retinal Fundus Images Implementing Machine Learning Algorithms

Abstract

Diabetic retinopathy is one of the leading causes of loss of vision that has affected approximately 93 million people. An analysis considering 35 studies, all across the world, estimated that the global figure of DR among diabetes patients is in the range of 7.62%–47.1%. If Diabetic retinopathy is not identified in an initial stage, it can cause serious vision problems such as vitreous hemorrhage, retinal detachment, glaucoma, and even permanent total blindness. Currently, it is detected by trained ophthalmologists and examining and evaluating the fundus photographs require a lot of time. This leads to delayed follow-ups and hence delayed treatment. Since we are aware that the population increases every day and so does the number of diabetic patients, the current infrastructure and manual method are insufficient. Thus there is a requirement for automatic and effective diabetic retinopathy detection. There have been previous attempts made at this and have even provided good progress with the classification and pattern recognition in the image and machine learning, we still require a method that can have potential as close as to the realistic clinical examination method. So in this paper, we have proposed a prompt method to detect diabetic retinopathy using retinal fundus images. Our model makes the prediction whether a person has diabetic retinopathy using Support Vector Machine with radial basis function (SVM-rbf) and also with the help of K-Nearest Neighbors (KNN) which are machine learning algorithms and we received an accuracy of about 96.62% and 94.38% using SVM-rbf and KNN respectively.