Deep Learning in Glaucoma Detection and Progression Prediction: A Systematic Review and Meta-Analysis

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Glaucoma diagnosis and progression prediction are critical for preventing vision loss.
• Deep learning (DL) shows promise in analyzing medical images for various diagnostic tasks.

Purpose of the Study:

• To systematically evaluate the diagnostic and prognostic performance of DL algorithms in glaucoma detection and progression prediction.
• To assess DL performance using fundus photography and optical coherence tomography (OCT) imaging.

Main Methods:

• A comprehensive literature search was conducted across major databases up to October 2024.
• A meta-analysis using a bivariate random-effects model was performed to pool performance metrics.
• Key metrics included pooled sensitivity, specificity, likelihood ratios, and AUROC.

Main Results:

• DL demonstrated high accuracy in glaucoma diagnosis via fundus photography (AUROC 0.90) and OCT (AUROC 0.86).
• DL performance in predicting glaucoma progression was less robust compared to diagnostic tasks.
• Internal validation datasets yielded higher accuracy than external ones.

Conclusions:

• DL algorithms are highly effective for glaucoma diagnosis using fundus photography and OCT.
• Integrating multimodal data and extensive real-world validation are crucial for improving DL-based glaucoma progression prediction.