Predicting Retinal Nerve Fiber Layer Thickness From Ocular Hypertension Treatment Study Optic Disc Photographs
James C Liu1, Alessandro A Jammal2, Rafael Scherer2
1Department of Ophthalmology, Washington University, St Louis, Missouri.
JAMA Ophthalmology
|June 26, 2025
Summary
Deep learning models can predict retinal nerve fiber layer (RNFL) thickness from optic disc photos, identifying ocular hypertension patients at risk for primary open-angle glaucoma (POAG). Faster RNFL thinning also predicts POAG development.
Area of Science:
- Ophthalmology and Artificial Intelligence
- Glaucoma Diagnostics and Risk Prediction
Background:
- Ocular hypertension is a significant risk factor for primary open-angle glaucoma (POAG).
- Early detection and risk stratification are crucial for managing ocular hypertension and preventing POAG progression.
- Retinal nerve fiber layer (RNFL) thickness is a key indicator of glaucomatous damage.
Purpose of the Study:
- To predict mean RNFL thickness using deep learning on optic disc photographs from the Ocular Hypertension Treatment Study (OHTS).
- To evaluate the predictive utility of RNFL thickness for the development of POAG in patients with ocular hypertension.
Main Methods:
- Utilized a diagnostic study design involving 3272 eyes from 1636 participants with ocular hypertension in the OHTS 1 and 2 trials.
- Employed an OCT-trained deep learning (machine-to-machine [M2M]) model to generate predicted RNFL thicknesses from 66,714 optic disc photographs.
- Analyzed factors correlating with POAG conversion using proportional hazards models, including predicted RNFL thickness and its longitudinal change.
Main Results:
- Lower baseline predicted RNFL thickness was significantly associated with conversion to POAG (HR, 1.83; P < .001 per 10-μm thinner).
- Faster longitudinal change in predicted RNFL thickness was a strong predictor of POAG development (HR, 6.01; P < .001 per 1-μm/year faster loss).
- Traditional risk factors such as age, intraocular pressure, and cup-disc ratio also remained significant predictors.
Conclusions:
- Machine-to-machine (M2M)-predicted RNFL thickness from optic disc photographs serves as a putative risk factor for glaucoma development in ocular hypertension.
- Baseline M2M-predicted RNFL thickness and its rate of change are valuable for assessing glaucoma risk and monitoring disease progression.
- These findings highlight the potential of deep learning in enhancing glaucoma risk assessment and patient management.
