Comparative analysis of deep learning models for the detection and classification of Diabetes Retinopathy

In recent years, diabetic retinopathy (DR), particularly in the elderly, has gained widespread recognition as a cause of blindness. The DR, which comes in a variety of forms and also has a variety of causes, is easily curable with early detection. Early detection of DR is challenging when manual medical approaches are used, and results are frequently inaccurate despite how long they take to complete. Therefore, a better approach to DR detection and prediction is required. Therefore, the purpose of this paper is to detect and classify diabetic retinopathy in patients using deep learning and compare different machine learning models to determine the one that performs best. The models employed are Convolutional Neural Network (CNN) that uses a four-layer VGG net plus an additional neural network to make it a custom five-layer network, K Nearest Neighbour (KNN) and Support Vector Machine (SVM). The IDRID which is the Indian Diabetic Retinopathy Image Dataset is where the dataset was acquired. When compared to other deep learning systems like KNN and SVM, which had an accuracy of 86% and 66% respectively, CNN attained an accuracy of 92%.