An eyecare foundation model for clinical assistance: a randomized controlled trial

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• The increasing need for clinical assessment of artificial intelligence (AI) in healthcare.
• Development of multimodal vision-language AI models for clinical applications.

Purpose of the Study:

• To develop and evaluate EyeFM, a multimodal vision-language AI copilot for eyecare.
• To assess EyeFM's efficacy in improving ophthalmologists' diagnostic performance and patient outcomes.

Main Methods:

• Multifaceted evaluation including retrospective validations, multicountry efficacy validation, and a double-masked randomized controlled trial (RCT).
• EyeFM was pretrained on 14.5 million ocular images across five modalities and global clinical texts.
• RCT involved 668 participants and 16 ophthalmologists in China, comparing EyeFM copilot use against standard care.

Main Results:

• Ophthalmologists using EyeFM achieved significantly higher correct diagnostic rates (92.2% vs. 75.4%) and referral rates (92.2% vs. 80.5%).
• Improved standardization scores for clinical reports were observed with EyeFM (median 33 vs. 37).
• Intervention group showed higher compliance with self-management (70.1% vs. 49.1%) and referral suggestions (33.7% vs. 20.2%) at follow-up.

Conclusions:

• EyeFM demonstrates significant utility as a clinical copilot in ophthalmology.
• Implementation of EyeFM can enhance ophthalmologist performance and improve patient outcomes in retinal disease screening.
• Strong user acceptance and positive impact on clinical decision-making and patient adherence were noted.