Diabetic Retinopathy: Detection and Classification Using AlexNet, GoogleNet and ResNet50 Convolutional Neural Networks

Diabetic retinopathy (DR) is an ocular condition developed in diabetes patients. This eye disease is increasing worldwide and is considered one of the leading causes of blindness; for this reason, early detection and prompt treatment are essential. DR can be divided depending on its severity into five stages: i) no DR, ii) mild, iii) moderate, iv) severe, and v) proliferative. This pathology is almost undetectable in its early stages, and it can even take a long time for highly trained healthcare professionals to detect it. In this context, artificial intelligence has become a promising solution compared to manual detection methods. It offers an easy, fast, less expensive, and more efficient alternative. Convolutional Neural Networks (CNN) have been widely used for medical image analysis. This study used three CNN: AlexNet, GoogleNet, and ResNet50 to detect and classify the five different stages of DR. The best results were obtained using AlexNet getting an accuracy of 93.56%, and the lowest value was obtained using GoogleNet (89.43%).