TOPLINE:

Adding an AI-based eye scan to fundus photograph-based screening pathways for diabetic retinopathy showed strong performance in detecting diabetic macular edema and did not perform worse than standard screening in referring people who needed specialist care. It also reduced unnecessary referrals for diabetic macular edema from 69.1% to 24.1%.

METHODOLOGY:

• Researchers conducted a stepwise evaluation in the Hong Kong Special Administrative Region to assess the diagnostic and referral performance of an AI-based optical coherence tomography (AI-OCT) system for detecting diabetic macular edema within a diabetic retinopathy screening pathway in clinical settings.
• The system analyzed eye scans and produced a report showing scan quality; status of diabetic macular edema (yes, uncertain, or no); a probability score indicating the likelihood of macular edema; and advice on whether to refer, observe, or review further. For the reference standard, two trained graders checked image quality, and a panel of masked ophthalmologists determined whether diabetic macular edema was present.
• The AI-based system was first prospectively tested from February 2021 to August 2023 in 603 participants with diabetes (mean age, 64.7 years; 56.2% male) at a tertiary hospital’s triage unit to assess how well it detected diabetic macular edema and judged image quality.
• This analysis was followed by a multicenter, noninferiority randomized clinical trial from September 2023 to April 2025 in 276 participants with suspected diabetic macular edema referred from a territory-wide diabetic retinopathy screening program; patients were allocated to the intervention group on the basis of both fundus photograph-based screening reports and AI reports (n = 137; mean age, 64 years; 44.5% female) or to the control group based on fundus photograph-based screening reports alone (n = 139; mean age, 63.7 years; 46.0% female).
• The main outcome in the noninferiority trial was false-positive referral (ie, the proportion of referred participants who did not have diabetic macular edema).

TAKEAWAY:

• In the prospective validation cohort involving 603 participants, the AI-based system achieved 98.8% sensitivity (95% CI, 94.5%-100%) and 90.7% specificity (95% CI, 88.7%-92.4%) for detecting diabetic macular edema, with 7.2% of scans identified as ungradable and 4.4% flagged as uncertain.
• The false-positive referral rate for diabetic macular edema was 24.1% (95% CI, 14.6%-37.0%) in the intervention group compared with 69.1% (95% CI, 61%-76.1%) in the control group, representing an absolute difference of -45% (95% CI, -58.2% to -31.9%; P < .001 for noninferiority).
• Sensitivity for diabetic macular edema referral was 100.0% (95% CI, 100.0%-100.0%) in both groups, whereas specificity was 86.5% (95% CI, 79.3%-92.9%) in the intervention group and 0.0% (95% CI, 0.0%-0.0%) in the control group; no cases of diabetic macular edema were found among nonreferred participants in the intervention group.
• The referral rate for diabetic macular edema was 39.4% (95% CI, 30.7%-47.5%) in the intervention group and 100.0% (95% CI, 100%-100%) in the control group, representing a 60.6% reduction in the referral rate, with an exploratory superiority analysis showing statistically significant between-group differences (P < .001).

IN PRACTICE:

“Incorporating the AI-OCT system was associated with a substantial reduction in potentially unnecessary referrals. Together, these findings support the integration of the AI-OCT system as an add-on secondary screening tool, with the potential to enhance the referral performance of diabetic retinopathy screening programs,” the researchers of the study reported.

SOURCE:

The study was led by Shuyi Zhang, PhD, Department of Ophthalmology and Visual Sciences, The Chinese University of Hong Kong, Hong Kong, China. It was published online on June 15 in JAMA.

LIMITATIONS:

The study was first tested in people seen at a hospital eye clinic, not directly in a regular diabetic retinopathy screening program. It focused only on people with suspected diabetic macular edema, so it may not apply to other eye problems. The researchers also did not test how the system would work in routine real-life use because the referral decisions were only hypothetical.

DISCLOSURES:

The study received funding from the Innovation and Technology Fund and the General Research Fund, Hong Kong Special Administrative Region, China. Some authors reported receiving grants, research support, consulting fees, and personal fees, and/or holding patents or equity interests, including founder or co-founder roles, in companies developing related technologies.

This article was created using several editorial tools, including AI, as part of the process. Human editors reviewed this content before publication.