Imagine a primary care doctor in a remote clinic, diagnosing retinal diseases with confidence thanks to an AI-powered assistant. That vision is closer to reality after Chinese mainland scientists from Tsinghua University and Shanghai Jiao Tong University led an international effort to develop EyeFM, a foundation vision model trained on 14.5 million ocular images and paired clinical texts from multi-ethnic, global datasets.
Published in Nature Medicine, the study brings together 44 ophthalmologists from the Chinese mainland, India, Malaysia, Denmark, Equatorial Guinea and the United States, validating EyeFM’s performance as a clinical co-pilot in settings ranging from low-resource primary care clinics to high-complexity specialty hospitals.
In a single-center, double-masked trial with 668 high-risk patients in the Chinese mainland, 16 ophthalmologists were randomized to use either EyeFM or standard care. The result? Diagnostic accuracy jumped to 92.2% with EyeFM support, compared with 75.4% using traditional methods.
EyeFM’s key innovation is a fast doctor feedback loop, allowing practitioners to refine the AI’s recommendations in real time. This feature makes it adaptable for a spectrum of clinical environments—from small rural clinics to urban hospitals—and establishes a playbook for integrating AI into everyday practice.
With high-level evidence now in hand, EyeFM exemplifies how large medical AI models can strengthen both primary and specialty eye care worldwide. For young global citizens, entrepreneurs and changemakers, this breakthrough signals a new era where AI and humans collaborate to deliver better health outcomes.
Reference(s):
cgtn.com
