Glaucoma, a leading cause of blindness, necessitates timely detection. This study develops and validates a generalized deep-learning algorithm for glaucoma screening using fundus images from 20 diverse global databases (18,468 images). The best-performing model (vgg19_bn) achieved high AUROC (0.9920), sensitivity, specificity, accuracy, precision, and F1-score (>0.9530) on the training dataset. External validation on the Drishti-GS dataset showed an AUROC of 0.8751 and accuracy of 0.8713. While accuracy decreased slightly on unseen data, highlighting dataset inconsistencies, the model demonstrates potential for population-level glaucoma screening, though further refinement with larger, diverse datasets is needed.

Abadh K. Chaurasia, Guei-Sheung Liu, Connor J. Greatbatch, Puya Gharahkhani, Jamie E. Craig, David A. Mackey, Stuart MacGregor, Alex W. Hewitt