Abstract

Study design: Prospective
Purpose: Evaluating the performance of a novel, on-the-edge AI system deployed on a portable fundus camera for screening referable AMD.
Methods: Macula-centered images were captured using a standard camera (Clarus 700) and study device (smartphone-based fundus camera) along with SD-OCT (macula line scan). 3 Retina specialists provided a masked diagnosis of AMD (pre-defined criteria) using both OCT and fundus images (Reference Standard). The Referable AMD AI (intermediate/advanced) was compared against this diagnosis.
Results: 485 patients (52% referable AMD) were analyzed. AI had a sensitivity of 90.91% (95% CI 86.67-94.15) and specificity of 88.21% (95% CI 83.09-92.22) against an adjudicated reference standard. False positives- 64% (16/25) early AMD ; false negatives -16 intermediate & 7 advanced AMD cases.
Conclusion: The novel, offline AMD AI deployed on a portable fundus camera has potential to make AMD screening accessible, affordable, and effective.