Clinical Report: AI-Based System for Assessing Multiple Disease Risks
Overview
This report evaluates an AI framework designed for simultaneous risk assessment of 15 diseases using retinal fundus images. Preliminary results include internal metrics such as ROC AUC and F1 scores, but challenges in real-world application are noted.
Background
Non-communicable diseases contribute significantly to global mortality, with many cases of vision impairment being preventable through early detection. Retinal imaging serves as a valuable tool for identifying both ocular and systemic diseases.
Data Highlights
Metric
Range
ROC AUC
0.9524–0.9971
F1 Score
0.8968–0.9649
Overall Accuracy (exploratory evaluation)
64.7% (95% CI: 52.9–76.5)
Key Findings
The AI system achieved ROC AUC of 0.9524–0.9971 across 15 conditions.
F1 scores ranged from 0.8968 to 0.9649.
Rare conditions with fewer than 100 training examples showed robust risk stratification.
In a single-site evaluation, the system's overall accuracy was 64.7%.
Challenges in real-world translation were noted due to limited sight-threatening cases.
Clinical Implications
Clinicians should be aware of the current limitations in sensitivity estimates for certain conditions.
Conclusion
This pilot study reports on the feasibility of AI in multi-pathology risk stratification from retinal images.
At the Outpatient Ophthalmic Surgery Society’s “OOSS Perspective 2026” symposium in Washington, DC, the organization's Washington counsel, Michael Romansky, JD, delivered an update on reimbursement, regulatory developments, and advocacy priorities affecting ophthalmic ambulatory surgery centers (ASCs).
