Related Experiment Video
Updated: Aug 12, 2026
09:17
Using Retinal Imaging to Study Dementia
Published on: November 7, 2017
Prediction of structural glaucoma progression from baseline fundus photographs using deep learning: a retrospective
Ruben Hemelings1, Damon W Wong2, Jacqueline Chua3
1Singapore Eye Research Institute, Singapore National Eye Centre, Singapore; SERI-NTU Advanced Ocular Engineering (STANCE) Program, Singapore; Research Group Ophthalmology, Department of Neurosciences, KU Leuven, Leuven, Belgium.
The Lancet. Digital Health
|August 10, 2026
Summary
A new deep learning model, G-PROG, accurately predicts glaucoma progression over 2-5 years using colour fundus photographs. This tool aids in identifying at-risk patients for timely intervention and vision preservation.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Early identification of glaucoma patients at risk of rapid progression is vital for preventing vision loss.
- Glaucoma progression prediction models are essential for proactive patient management.
Purpose of the Study:
- To develop and externally validate G-PROG, a deep learning model for predicting 2-5 year glaucoma progression.
- To assess the model's performance using baseline colour fundus photographs (CFPs).
Main Methods:
- G-PROG was trained and validated on data from multiple international centers.
- Utilized 161,827 fundus images from 13,913 patients across six glaucoma departments.
- Progression was defined by the G-RISK slope, validated against visual field (MD) and retinal nerve fibre layer (RNFL) thickness slopes.
Main Results:
- G-PROG achieved high internal validation AUCs up to 0.98.
- External validation across five cohorts showed maximum AUCs ranging from 0.74 to 0.86.
- The G-RISK slope demonstrated strong agreement with established progression markers (MD and RNFL slopes).
Conclusions:
- G-PROG is an externally validated deep learning model capable of predicting 2-5 year glaucoma progression from baseline CFPs.
- Further prospective evaluation is recommended to confirm its utility in risk stratification and resource allocation in glaucoma care.
