Constructive Analysis on Prediction and Detection of Diabetic Retinopathy (DR) using Machine Learning Algorithms - A Generic Survey

Abstract

Diabetic retinopathy is an eye disorder that affects people with diabetes. When high blood sugar levels harm the retina's blood vessels, the blood vessels may enlarge and leak or they may block the flow of blood. All these alterations may cause visual loss. In the current era, machine learning is very useful and plays a key role in medical applications. In this research, investigate the effectiveness and capability of different types of Machine Learning (ML) algorithms based diabetic retinopathy detection systems. The authors tested large quantities of retinal fundus images & gray scale pictures from Kaggle, Messidor, IEMRC and multiple available datasets. The application of an ensemble of machine learning classification algorithms to characteristics taken from the results of various retinal image processing algorithms is used to determine the presence of disease. The proposed goal is to anticipate the DR before it results in the worst scenario and prevent the patient from losing their visual acuity. The proposed method uses both supervised algorithms like Support Vector Machine (SVM), ResNet and DensNet, Naive Bayes, K-Nearest Neighbors, and Neural Networks (NN), as well as unsupervised algorithms like k-means clustering, hierarchical clustering, and Markov chains to classify a variety of characteristics in DR. The best overall result is displayed by ResNet and DensNet with the accuracy of 96.22%.