AI-Powered OCT Improves Diabetic Eye Screening Efficiency: JAMA

A new study published in The Journal of American Medical Association showed that during diabetic retinopathy screening, referral judgments for diabetic macular edema (DME) were enhanced using an AI-based optical coherence tomography (OCT) system.

The worldwide standard of treatment is to screen for diabetic retinopathy using fundus photos, however this method has a high rate of false-positive referrals to assess DME, which puts a significant strain on specialty eye clinics. Potentially needless referrals might be decreased by including an AI-based OCT (AI-OCT) technology into screening procedures. In order to assess an AI-OCT system's diagnostic and referral performance for DME detection within a diabetic retinopathy screening pathway in clinical settings, this study was carried out.

A multicenter noninferiority RCT from September 2023 to April 2025 recruited 276 patients with suspected DME referred from a territory-wide diabetic retinopathy screening program. A prospective silent-mode validation (February 2020 to July 2023) recruited 603 patients with diabetes at a tertiary hospital triage unit. Participants in the RCT were randomly assigned to either the control group or the intervention (referral for DME assessment based on both fundus photograph-based screening reports and AI-OCT reports [n = 137]). The AI-OCT system included uncertainty flagging, DME identification, and picture quality evaluation. 

The system showed good diagnostic performance for diabetic macular edema (DME) detection in prospective silent-mode validation, with 98.8% sensitivity and 90.7% specificity, 7.2% ungradable scans, and 4.4% doubtful scans. The intervention group significantly decreased the false-positive referral rate to 24.1% from 69.1% in the control group in the subsequent randomized controlled trial, while maintaining perfect sensitivity (100%) for DME referrals (absolute difference, −45%; $P <.001$ for noninferiority).

Also, there were no missed DME instances among patients who were not referred, and the specificity for DME referral increased to 86.5% in the intervention group compared to 0.0% in the control arm. Overall, the add-on AI-based OCT system was linked to a significant decrease in possibly needless referrals and satisfied the predetermined criterion for noninferiority when compared to regular treatment. This discovery offers a useful framework for the actual application of AI-enabled solutions in clinical specialties such as ophthalmology.

Source:

Zhang, S., Ran, A., Zhou, J., Ling, A., Sham, K., Zhang, Y., Tang, Z., Nguyen, T. X., Yang, D., Lam, N., Yuen, H. K. L., Chan, V. T. T., Ho, M., Chan, J. Y. Y., Lam, T. C. H., Yim, C. C. L., Chow, C. W. Y., Cheung, S. S. L., Lam, M. C. W., … Cheung, C. Y. (2026). An AI-based OCT system to detect diabetic macular edema: A prospective validation and noninferiority randomized clinical trial. The Journal of the American Medical Association. https://doi.org/10.1001/jama.2026.7025