A generalised computer vision model for improved glaucoma screening using fundus images

Worldwide, glaucoma is a leading cause of irreversible blindness. Timely detection is paramount yet challenging, particularly in resource-limited settings. Herein, we sought to develop and validate a generalised deep-learning-based algorithm for screening glaucoma using fundus images. We collected glaucomatous data from 20 publicly accessible databases worldwide and selected the best-performing model from 20 pre-trained models. The top-performing model was further trained for classifying healthy and glaucomatous fundus images using Fastai and PyTorch libraries. Gradient-weighted class activation mapping was used to visualise significant areas of fundus images for model decision-making. The best-performing model was validated on 1,364 glaucomatous discs and 2,047 healthy discs. Validation performance metrics indicate robust discriminative ability, with an Area Under the Receiver Operating Characteristic (AUROC) of 0.9920 (95% CI: 0.9920 to 0.9921) for glaucoma and 0.9920 (95% CI: 0.9920 to 0.9921) for healthy class. The model performed well on an external validation (unseen) set of the Drishti-GS dataset, with an AUROC of 0.8751 and an accuracy of 0.8713. Although the model's accuracy slightly decreased when evaluated on unseen data, this study highlighted the potential of computer vision to assist in glaucoma screening.