AI-Augmented Fundus Disease Screening by Non-Ophthalmologist Physicians: A Paired Before-After Study

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence in Healthcare

Background:

• Retinal disease screening is expanding in primary care, often performed by non-specialists with variable accuracy.
• Limited access to ophthalmology services necessitates improved diagnostic capabilities in primary care settings.

Purpose of the Study:

• To evaluate an AI-based decision support tool for enhancing non-ophthalmologist diagnostic accuracy in retinal disease screening.
• To assess the impact of AI assistance on referral decisions made by general practitioners.

Main Methods:

• A paired before-after study design was employed.
• Four non-ophthalmologists reviewed 500 color fundus photographs, first without AI assistance and then with AI support after a washout period.
• Diagnostic accuracy and referral decisions were compared between unaided and AI-assisted reviews.

Main Results:

• AI assistance significantly improved diagnostic accuracy from 82.8% to 91.1% (p < 0.0001).
• The most substantial improvements were noted in identifying glaucoma-suspect cases and patients with multiple pathologies.
• A favorable safety profile was observed, with clinicians retaining final diagnostic authority.

Conclusions:

• AI-assisted diagnostic tools can meaningfully augment the screening and referral capabilities of non-ophthalmologists in primary care.
• The findings support the integration of AI decision support to improve real-world primary care for retinal diseases.
• Further validation and implementation studies are recommended to broaden the application of this technology.