Artificial Intelligence-Guided Endpoint Selection for Neuroprotection Trials in Glaucoma.

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Standard Automated Perimetry (SAP) is crucial for glaucoma monitoring and clinical trials.
• Identifying vulnerable visual field locations for early glaucoma progression detection remains a challenge.

Purpose of the Study:

• To develop an AI model predicting glaucoma progression hotspots using baseline SAP data.
• To enable targeted endpoint selection for neuroprotection trials and personalized glaucoma monitoring.

Main Methods:

• A graph attention network (GAT) was trained on baseline SAP data to predict progression risk at each test location.
• The model identified the five highest-risk points (High-5) for each eye.
• Validation was performed on independent datasets using metrics like AUC and rate of change.

Main Results:

• High-5 locations showed significantly steeper progression rates (4-5 times) compared to Low-5 and Mean Deviation (MD).
• The AI model demonstrated high discrimination performance (AUC 0.883-0.937) for identifying progressors.
• Nearly all progressing eyes reached the High-5 ≥7 dB threshold during follow-up.

Conclusions:

• The AI model effectively identifies the most vulnerable visual field locations from baseline data.
• The High-5 metric offers a sensitive and efficient framework for glaucoma clinical trials and patient monitoring.