Multi-class Classification Approach for Retinal Diseases

Abstract

—Early detection of the diagnosis of some diseases in the retina of the eye can improve the chances of cure and also prevent blindness. In this study, a Convolutional Neural Network (CNN) with different architectures (Scratch Model, GoogleNet, VGG, ResNet, MobileNet and DenseNet) was created to make a comparison between them and find the one with the best percentage of accuracy and less loss to generate the model for a better automatic classification of images using a MURED database containing retinal images already labeled previously with their respective disease. The results show that the model with the ResNet architecture variant InceptionResNetV2 has an accuracy of 49.85%.