Deep learning-based glaucoma classification from fundus images: Comparative analysis of CNN architectures

Glaucoma represents a leading cause of irreversible blindness worldwide, characterized by progressive optic nerve damage that often progresses asymptomatically until significant vision loss occurs. Early detection remains critical for preventing permanent visual impairment, yet manual diagnosis by ophthalmologists is time-consuming, subjective, and inaccessible in resource-limited regions. This study presents a comparative evaluation of three convolutional neural network (CNN) architectures—DenseNet121, VGG16, and ResNet50—for automated glaucoma classification from retinal fundus images. Using a dataset of 705 fundus images (564 training, 141 testing) from publicly available sources, we evaluate each model's performance using accuracy, precision, recall, F1-score, and ROC-AUC metrics. Our results demonstrate that ResNet50 achieved the highest classification accuracy of 98.56% on the test set, while VGG16 attained 95.00% accuracy and DenseNet121 achieved 94.33% accuracy on the test set. These findings underscore the effectiveness of transfer learning approaches in medical image analysis and suggest that deep learning systems can serve as reliable clinical decision support tools. However, we acknowledge that evaluation on larger, multi-institutional datasets is essential to validate generalization capabilities and clinical applicability. This work contributes to the growing body of evidence supporting automated glaucoma screening systems as scalable solutions for early diagnosis in both developed and developing healthcare settings.