Artificial Intelligence-Driven Multimodal Prediction of 10-Year Incident Glaucoma Integrating Genetic and Deep...

Area of Science:

• Ophthalmology and computational medicine.
• Artificial intelligence in healthcare diagnostics.

Background:

• Glaucoma is a leading cause of irreversible blindness globally, often asymptomatic until advanced stages.
• Early prediction of glaucoma is critical for timely intervention and vision preservation.

Purpose of the Study:

• To develop and validate an interpretable, multimodal machine learning (ML) framework for predicting 10-year incident primary open-angle glaucoma (POAG).
• To integrate diverse data sources including imaging, genetics, and clinical factors for enhanced predictive accuracy.

Main Methods:

• A population-based prospective cohort study using UK Biobank data.
• Development of a multimodal ML framework integrating deep learning (DL)-derived color fundus photograph (CFP) features, ocular measurements, polygenic risk scores (PRS), lifestyle factors, and electronic health records.
• Benchmarking of ten ML algorithms and application of SHapley Additive exPlanations (SHAP) for interpretability.

Main Results:

• The XGBoost model achieved high accuracy (AUC 0.927) for predicting incident POAG.
• Deep learning imaging features and PRS significantly contributed to predictive value beyond traditional factors.
• SHAP analysis identified CFP DL score, age, PRS, and intraocular pressure as key predictors.

Conclusions:

• A multimodal ML framework integrating genetic and DL-derived imaging features enables accurate and interpretable prediction of incident POAG.
• This approach holds promise for early detection and intervention strategies to prevent glaucoma-induced vision loss.