Toward Long-Term Visual Field Appearance Forecasting Using Artificial Intelligence for Ophthalmic Education and...

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Glaucoma progression (GP) detection is crucial for timely treatment.
• Current methods for characterizing GP can be challenging for trainees.
• Visual field (VF) testing is a key diagnostic tool in glaucoma management.

Purpose of the Study:

• To develop and evaluate an AI tool for forecasting visual field (VF) appearance to aid in training ophthalmic residents.
• To assess the clinical utility and impact of AI-generated VF forecasts on resident decision-making.
• To provide state-of-the-art, post hoc-explainable AI forecasts for glaucoma progression detection.

Main Methods:

• Retrospective study utilizing the largest available VF dataset, including minority populations.
• Implementation of GenVF, a generative vision transformer, for forecasting VFs up to 10 years.
• User study involving 8 ophthalmology residents comparing AI-forecasted VFs with actual VFs.

Main Results:

• AI-forecasted VFs (AI-VF) showed uniform mean absolute error (MAE) across resident experience levels, unlike VF-only analysis.
• Residents using AI forecasts made more conservative decisions and reported higher confidence.
• Reaction times for glaucoma progression detection remained comparable with or without AI guidance.

Conclusions:

• The GenVF model provides state-of-the-art, post hoc-explainable AI for glaucoma progression detection.
• AI-assisted VF forecasting enhances resident confidence and decision-making, potentially expediting glaucoma treatment.
• Future research should address variability in AI effectiveness across expertise and patient-specific contexts.