A Transfer Learning–Based Deep Convolutional Neural Network Framework for Automated Multi-Class Eye Disease Classification in the USA Using Retinal Fundus Image

Abstract

Eye diseases are among the leading causes of visual impairment and blindness worldwide, and delayed diagnosis frequently results in irreversible vision loss. Early and accurate detection remains challenging because many ocular conditions exhibit subtle visual features and current diagnosis relies heavily on manual clinical assessment, which is time-consuming and subject to inter-observer variability. This study proposes an automated deep learning–based framework for multi-class eye disease classification using retinal fundus images. A transfer learning strategy is employed by fine-tuning multiple pre-trained convolutional neural network architectures, including VGG-16, VGG-19, ResNet-50, ResNet-152, and DenseNet-121. The proposed system is evaluated on a publicly available benchmark dataset comprising eight ocular disease categories. Image preprocessing and model optimization techniques are applied to enhance classification performance. Experimental results show that the fine-tuned VGG-19 model achieves the best performance, reaching an overall accuracy of 95% with balanced precision, recall, and F1-score. These results demonstrate that transfer learning significantly improves diagnostic accuracy while reducing computational complexity. The proposed framework provides a reliable, scalable solution for automated eye disease diagnosis and has strong potential to support clinical decision-making in ophthalmic screening and telemedicine systems.