Deep Learning-Based Computer-Aided Diagnosis of Glaucoma from Fundus and OCT Imaging: A Structured Review

Abstract

Glaucoma remains a leading cause of irreversible visual disability worldwide, largely because structural damage to the optic nerve can progress substantially before a patient notices any vision change. Computer-aided diagnosis (CAD) using deep learning has therefore been explored as a strategy to support earlier identification, drawing on fundus photography, optical coherence tomography (OCT), visual field testing, and combinations of these modalities. This structured review discusses recent studies on AI-assisted detection in glaucoma management and considers their practical significance for clinical care. According to the literature reviewed, convolutional neural networks (CNNs) play a clear role in glaucoma screening, explicit segmentation models support assessment of the optic disc and cup, and OCT-based methods represent structural change more directly than surface imaging alone. Multimodal approaches, three-dimensional OCT analysis, transformer architectures, and progression-tracking models represent active and promising directions. However, clinical translation remains constrained by biased or geographically narrow datasets, device-specific variation, limited external validation, and unresolved questions about the reliability of individual predictions. The literature therefore supports AI as a developing decision-support tool that complements, rather than replaces, ophthalmic assessment. Further research should prioritise validation of AI-based tools in multi-centre settings, transparent documentation of errors and uncertainty, and prospective testing of systems intended for real-time clinical use.