From left: Prof Clement Tham, Prof Carol Cheung, Dr Shu-Yi Zhang, Dr Simon SzetoAn AI-based optical coherence tomography (AI-OCT) system for add-on secondary screening of diabetic macula oedema (DME) reduces false-positive DME referrals by 45 percent vs standard screening by fundus photography alone, a randomized controlled trial (RCT) in Hong Kong has shown.
Developed by researchers from the Department of Ophthalmology and Visual Sciences, Chinese University of Hong Kong, the AI-OCT system automatically identifies and filters ungradable OCT scans, detects DME, and flags uncertain DME cases for real-time clinical review. Its image quality assessment model addresses common barriers to AI adoption in clinical practice by automating image gradability assessment prior to disease detection. [JAMA 2026;336:215-223]
To evaluate the diagnostic and referral performance of the AI-OCT system as a secondary tool for DME screening within diabetic retinopathy screening pathways, the researchers conducted a stepwise clinical study involving a prospective silent-mode validation (February 2021 to August 2023) followed by a multicentre noninferiority RCT (September 2023 to April 2025; follow-up completed in May 2025).
High sensitivity & specificity
The AI-OCT system demonstrated a sensitivity of 98.8 percent and a specificity of 90.7 percent for DME detection in the prospective silent-mode validation involving 1,200 eyes from 603 patients (mean age, 64.4 years; male, 56.2 percent) with diabetes, who were recruited at the triage unit of a tertiary eye hospital.
The image quality assessment model identified 7.2 percent of OCT images as ungradable, demonstrating a sensitivity of 98.1 percent and a specificity of 91.4 percent.
RCT supports AI-OCT as add-on secondary screening tool
The noninferiority RCT included 276 patients (mean age, 63.9 years; male, 54.7 percent) with suspected DME recruited from a territory-wide diabetic retinopathy screening programme spanning primary, secondary and tertiary care.
The participants were randomized to an intervention group where referral for DME evaluation was based on both AI-OCT reports and fundus photography screening reports (n=137), or a control group where referral was based solely on fundus photography screening reports (n=139).
Results showed a 45 percent reduction in false-positive DME referral rate in the intervention vs control group (24.1 vs 69.1 percent; pnoninferiority<0.001), with the upper bound of the 95 percent confidence interval below the trial’s prespecified noninferiority margin of 20 percent.
An exploratory superiority analysis showed a significant difference in false-positive DME referral rate between the two groups (p<0.001).
Overall, the DME referral rate was 39.4 percent in the intervention group vs 100 percent in the control group, representing a 60.6 percent reduction in referrals with add-on AI-OCT screening. The prevalence of DME was similar between the two groups (29.9 vs 30.9 percent).
“Sensitivity for DME referral was 100 percent in both groups, while specificity was 86.5 percent in the intervention group and 0 percent in the control group. No cases of DME occurred among nonreferred participants in the intervention group,” the researchers reported.
“Compared with standard practice, incorporation of the AI-OCT system as a secondary screening tool was noninferior with respect to false-positive referral rates and was associated with a substantial reduction in potentially unnecessary DME referrals without compromising sensitivity,” the researchers concluded.
“These findings support the use of the AI-OCT system as an add-on secondary tool to reduce false-positive DME referrals and alleviate workload in tertiary care settings,” they suggested.