Disease Detection from Retinal OCT Images

Abstract

In this study, three common Convolutional Neural Network (CNN) architectures—AlexNet, VGG16, and ResNet50—are compared to classify retinal OCT (Optical Coherence Tomography) images into four classes: CNV, DME, DRUSEN, and NORMAL. Each model is analyzed using evaluation metrics including accuracy, precision, F1 score, and ROC AUC. The main objective is to reveal the performance differences of various CNN architectures in medical image classification tasks and to determine their potential contribution to the computer-aided diagnosis of retinal diseases. Experimental findings show that all models achieve high success rates; in particular, architectures with deep and residual connections can learn complex textural structures more effectively. These results suggest that CNN-based models provide a reliable basis for advanced diagnostic systems to be developed for automated OCT analysis.