Eye Disease Classification System Based on Fundus Images Using the InceptionV3 Architecture

Abstract

This study aims to develop an automated eye disease classification system based on retinal fundus images using the InceptionV3 deep learning architecture. The dataset consists of four classes: cataract, diabetic retinopathy, glaucoma, and normal, collected from public sources and clinical data. The proposed method applies several preprocessing techniques, including background segmentation, data augmentation, data normalization, and an 80:20 data split to improve model performance and generalization. Transfer learning is implemented by utilizing pretrained ImageNet weights and modifying the final layers to suit the classification task. The model is trained using the Adam optimizer with a learning rate of 0.001 and categorical cross-entropy loss function. Evaluation results show that the model achieves an accuracy of 96%, with average precision, recall, and F1-score values of 0.97, 0.96, and 0.97, respectively. The confusion matrix analysis indicates that most predictions are correctly classified, demonstrating strong performance across all classes. Furthermore, the model is successfully integrated into a web-based system that enables users to upload fundus images and obtain classification results automatically. These findings indicate that the proposed system can effectively assist in early detection of eye diseases and support clinical decision-making.