Deep Learning Architectures for Ophthalmic Diagnosis: Evaluating CNN, VGG16, ResNet50, and Inception V3 in Myopia Detection

In the world, the most prevalent vision disorder is myopia, and it is desirable to prevent its early diagnosis for long-term effects and vision impairment. For automatic classification of retinal fundus images into myopic and normal cases, deep learning techniques were applied in the presented work. Four architectures: CNN, VGG16, ResNet50, and InceptionV3 were evaluated using benchmark dataset and its performance was accessed through various evaluation metrics including precision, recall, accuracy, Area Under Curve (AUC) nad F1-score. It was experimentally found that all, good predictive performance was achieved with models accuracies more than 98% and AUC values near 1. 0. The best accuracy was reported by ResNet50 among others of 99.65% and F1score of 0.9965, while VGG16 received the highest AUC of 0.9999, which confirms their increased discriminative capacity. The similarity in all models contribute to the stability of deep learning through performance in capturing myopia-related pathological features. These findings demonstrate that AI-based diagnosis equipment may serve ophthalmologists with great assistance for Promoting mass, rapid, and precise myopia screening. This study provides the foundation of the development of scalable clinical decision support systems and provides possibilities to future research with greater, more diverse data, and interpretable AI in order to build more trust and uptake in real-world healthcare applications.