Deep Learning Classification of Retinal Images for the Early Detection of Diabetic Retinopathy Disease

Abstract

The health-related issues of diabetes, inflammation, ageing, macular degeneration, and fungal infections can damage the retina and macula of the eye, leading to permanent vision loss. The most common cause of diabetic eye blindness is diabetic retinopathy (DR). As a result, detecting DR at an early stage is critical. This article investigated a deep learning-based approach for the early detection of DR from retinal images. The primary objective of this paper is to investigate diabetic detection using retinopathy images. Deep learning algorithms that used patient retinal images, convolutional neural network (CNN), and EfficientNet are used to diagnose the severity of DR. The images are divided into five categories based on the severity of the disease. CNN could be a machine learning branch that can accurately classify images. This research attempted to develop and validate a deep learning algorithm for the detection of DR in retinal fundus photographs. To further improve the result, the proposed system used an EfficientNet as well as an optimizing threshold. In this DR classification, EfficientNet achieved a higher accuracy of 99.3% than the CNN, which achieved a lower accuracy of 93.27%.