Prediction of structural glaucoma progression from baseline fundus photographs using deep learning: a retrospective multicentre study.
G-PROG shows promise in predicting glaucoma progression from fundus photographs, but further prospective evaluation is needed to confirm its clinical utility.
Where it sits
this study against the rest of the aod-9604 corpusSummary and findings
This study aimed to develop and validate G-PROG, a deep learning model predicting glaucoma progression from baseline fundus photographs. The model was trained on data from a single center and validated externally across five international cohorts. Significant AUC values were reported for model performance, but no treatment outcomes were assessed.
Abstract
<h4>Background</h4>Identifying patients with glaucoma who are at risk of rapid disease progression is crucial to preventing vision loss. We aimed to develop and externally validate G-PROG, a deep learning model that predicts 2-5-year glaucoma progression from baseline colour fundus photographs (CFPs).<h4>Methods</h4>G-PROG was trained and validated on data from a single centre (UZ Leuven, Leuven, Belgium); the other datasets (Brussels, Belgium; Liège, Belgium; Tampere, Finland; Mainz, Germany; and Hangzhou, China) served as external test sets. Across six glaucoma departments, we analysed 161 827 fundus images from 127 962 visits (13 913 patients), totalling 128 021 eye-years of follow-up. Progression was defined by the G-RISK slope, calculated via within-eye linear regression on longitudinal G-RISK predictions over follow-up intervals of 2-5 years. G-RISK is a previously validated deep learning model that quantifies glaucomatous optic nerve damage from CFPs. We trained 20 G-PROG configurations with varying inclusion criteria applied to the number of visits, image quality, time between visits, and G-RISK at baseline. Performance was evaluated using the area under the receiver operating characteristic curve (AUC), the coefficient of determination (R<sup>2</sup>), and explained variance score (EVS). G-RISK slope as a progression biomarker was validated against the visual field mean deviation (MD) slope and average retinal nerve fibre layer thickness (RNFL) slope.<h4>Findings</h4>Significant AUC values were obtained in 18 out of 20 model configurations, with internal validation reaching a maximum AUC of 0·98 (95% CI 0·97-1·00) across follow-up intervals (2-5 years). In glaucomatous eyes with a baseline G-RISK exceeding 0·6, the maximum AUC was 0·92 (0·85-0·98). For external validation, the predictions from the eight top-performing configurations (selected based on positive R<sup>2</sup> and minimal discrepancy between R<sup>2</sup> and EVS in internal validation) were averaged. Maximum AUC values ranged from 0·74 to 0·86 across the five test datasets. G-RISK slope showed significant agreement with established progression markers, with maximum AUCs of 0·82 for MD slope and 1·00 for average RNFL slope.<h4>Interpretation</h4>Externally validated across five international cohorts, G-PROG predicts 2-5-year glaucoma progression from baseline CFPs. Prospective evaluation is warranted to assess whether G-PROG can improve risk stratification and resource allocation in glaucoma care.<h4>Funding</h4>This work was funded and supported by grants from the National Medical Research Council, National Research Foundation Singapore, National Health Innovation Centre Singapore, SingHealth and Duke-NUS, Duke-NUS, the Singapore Eye Research Institute and Nanyang Technological University and the Singapore Eye Research Institute, the Competitive Research Funding of the Pirkanmaa Wellbeing Services County, the LUX-Foundation for Glaucoma Research, state funding for university-level health research at Tampere University Hospital, Wellbeing Services County of Pirkanmaa, the Tampere University Hospital Support Foundation, and the Belgian Ophthalmology Cooperation in Clinical Sciences initiative hosted by the Funds for Research in Ophthalmology.
Background
This paper addresses the challenge of identifying patients with glaucoma at risk for rapid disease progression, a critical factor in preventing vision loss. Prior research has established the importance of early detection and monitoring of glaucoma progression. The development of a predictive model like G-PROG could enhance risk stratification and resource allocation in glaucoma care, making this study significant.
Methods
G-PROG was trained on data from a single center (UZ Leuven) and validated using external datasets from five additional centers. The study analyzed 161,827 fundus images from 127,962 visits, totaling 128,021 eye-years of follow-up. Progression was defined using the G-RISK slope calculated via linear regression over follow-up intervals of 2-5 years. The performance was evaluated using AUC, R², and explained variance score.
Results
The primary endpoint showed a maximum AUC of 0.98 (95% CI 0.97-1.00) in internal validation. Eighteen out of twenty model configurations achieved significant AUC values. For external validation, maximum AUC values ranged from 0.74 to 0.86 across five test datasets. G-RISK slope demonstrated significant agreement with established progression markers, with maximum AUCs of 0.82 for MD slope and 1.00 for average RNFL slope.
Interpretation
The findings suggest that G-PROG can effectively predict glaucoma progression, with high AUC values indicating strong predictive performance. However, while the statistical significance is notable, the clinical significance of these findings remains to be fully established. The reliance on external validation and the potential for variability in clinical settings limit the conclusions that can be drawn. Further prospective studies are needed to determine the practical implications of using G-PROG in clinical practice.
Key findings
- Maximum AUC of 0.98 (95% CI 0.97-1.00) in internal validation across follow-up intervals (2-5 years).
- In glaucomatous eyes with a baseline G-RISK exceeding 0.6, maximum AUC was 0.92 (0.85-0.98).
- Maximum AUC values for external validation ranged from 0.74 to 0.86 across five test datasets.
- G-RISK slope showed maximum AUCs of 0.82 for MD slope and 1.00 for average RNFL slope.
Limitations
- External validation across multiple centers may introduce variability.
- Further prospective evaluation is needed to assess clinical applicability.
- Not reported in abstract.