AI-Driven OCT Matches Human Precision in Retinal Monitoring
Home-based optical coherence tomography (OCT) paired with artificial intelligence offers a reliable path for tracking retinal fluid in patients with neovascular age-related macular degeneration (nAMD). Findings presented at the American Society of Retina Specialists annual meeting reveal that AI-driven monitoring achieves diagnostic agreement levels mirroring those of human experts across more than 21,000 scans.
Evaluating Data Across 21,421 Longitudinal Scans
Researchers scrutinized 381 eyes gathered from four longitudinal studies to test the clinical utility of home OCT. The dataset, spanning 906 monitoring months, included 21,421 individual scans. Anat Loewenstein, MD, MHA, of Tel Aviv Medical Center, presented the effort to determine if AI-generated data trajectories match the judgment of expert human graders when identifying significant changes in retinal fluid.
To pinpoint these shifts, the team employed a reference change value (RCV) method typically reserved for laboratory medicine. By calculating the signal-to-noise ratio—the maximum fit minus the minimum fit, divided by the estimated variation—the researchers moved beyond generic metrics to establish personalized thresholds for detecting fluid activity.
Personalized Thresholds Outperform Population Models
The study pitted population-based approaches against personalized thresholding, with the latter proving superior. The personalized model yielded 99% sensitivity in detecting fluid changes, besting the 94% recorded by the population-based model. When reviewing about 300 clinical trajectories, expert graders classified one-third as “weak” and two-thirds as “stable.” These results confirm that AI assessments of hyperreflective spaces align closely with the clinical judgment of experienced retina specialists.
Bridging the Gap Between Clinic Visits
Previous prospective trials already confirmed that home OCT provides visualization comparable to in-office equipment. By proving that AI-driven longitudinal analysis is also comparable to human oversight, this technology supports a shift toward proactive management of nAMD, enabling the early detection of disease reactivation.
Summary of Findings
- Study Scale: Researchers analyzed 21,421 home OCT scans from 381 eyes.
- Diagnostic Sensitivity: A personalized threshold approach achieved 99% sensitivity in detecting fluid changes, outperforming population-based models.
- Expert Agreement: AI-based analysis demonstrated clinical agreement with expert human graders regarding disease stability and fluid presence.
- Clinical Utility: Home OCT facilitates the early detection of disease reactivation, providing a viable alternative for monitoring patients with nAMD between clinic appointments.
Disclosures: Anat Loewenstein reports consulting for Notal Vision.
Keep reading