CU Medicine’s Department of Ophthalmology and Visual Sciences has developed an AI-OCT system as an improved alternative to the current screening standard for diabetic macular edema (DME). The system utilises AI to re-examine patients suspected of having DME, successfully screening out 60% of false-positive referrals to specialist clinics. The team hopes the system will minimise unnecessary referrals, reduce the burden on the healthcare system, and shorten waiting times for patients requiring urgent care.

(From left) Research team from the Department of Ophthalmology and Visual Sciences at CU Medicine: Professor Clement Tham Chee-yung, S.H. Ho Professor of Ophthalmology and Visual Sciences and Department Chairman; Professor Carol Cheung Yim-lui; Dr Zhang Shuyi, Postdoctoral Fellow; and Dr Simon Szeto Ka-ho, Assistant Professor.

A research team from The Chinese University of Hong Kong (CUHK)’s Faculty of Medicine (CU Medicine) has developed an AI-assisted optical coherence tomography (AI-OCT) system as an improved alternative to current screening standard for diabetic macular edema (DME), a a major cause of blindness. The system utilises AI to re-examine patients suspected of having DME, successfully screening out 60% of false-positive referrals to specialist clinics. The team hopes the system will minimise unnecessary referrals, reduce the burden on the healthcare system and shorten waiting times for patients requiring urgent care, achieving the goal of early detection and timely treatment. The findings have been published in the Journal of the American Medical Association (JAMA), one of the world’s leading scientific journals.

Diabetic retinopathy, including DME, is a major cause of blindness among working-age individuals with diabetes. When high blood sugar persists, DME damages the small blood vessels in the retina, reducing blood supply and triggering retinal hypoxia. In response, the retina signals for the growth of new blood vessels to increase nutrient supply. However, these new vessels are abnormally fragile and prone to leakage and bleeding, leading to fluid accumulation and swelling. If this swelling occurs at the macula, the area with the highest concentration of photoreceptor cells, diabetic macular edema develops.

Routine diabetic retinopathy screening using fundus photographs, currently the standard worldwide, has limitations in evaluating DME, resulting in a high rate of false positives.

Silent early symptoms of DME may lead to blindness if untreated

Professor Clement Tham notes that while relying solely on fundus photographs can lead to high false-positive referral rates, the AI-OCT system significantly enhances screening accuracy and referral precision for DME. However, he cautions that while AI is designed to streamline clear-cut cases, a doctor’s final assessment remains indispensable for ambiguous ones. He emphasises that the team's core screening principle is to never miss a genuine case of DME.

Professor Clement Tham Chee-yung, S.H. Ho Professor of Ophthalmology and Visual Sciences and Chairman of the Department of Ophthalmology and Visual Sciences at CU Medicine, said: “DME is one of the leading causes of vision impairment in patients with diabetes, affecting approximately 7% of this population. In its early stages, DME is often silent and asymptomatic. However, if left untreated, by the time a patient notices blurred vision, it may already be too late and irreversible. In severe cases, this can lead to significant vision loss or even blindness.”

 

Screening for diabetic retinopathy among diabetic patients is an important strategy to reduce vision loss caused by diabetic eye disease. Currently, regular fundus photography is widely used in clinical practice to detect and assess vascular leakage or swelling in the macular area. However, as it only provides a 2D view of the retina, relying solely on them as proxy to refer suspected DME has resulting in a high rate of false positives. This limitation leads to unnecessary consultations, creates unnecessary anxiety among patients, increases the workload for ophthalmologists, and strains valuable medical resources.

AI-OCT significantly reduces false-positive in screening

The AI-OCT system, developed by a CU Medicine’s Department of Ophthalmology and Visual Sciences team, combines new AI techniques with OCT to provide 3D imaging of the retina, aiming to improve the accuracy of screening and referrals.  A stepwise clinical evaluation of the AI-OCT system was conducted. In Phase 1, between 2020 and 2023, 603 diabetic patients were identified through a triage clinic. The findings revealed that the AI-OCT system exhibited a sensitivity of 98% and a specificity of 91% in detecting DME among the diabetic population.

Based on the findings from Phase 1, a subsequent Phase 2 study was conducted. Between 2023 and 2025, the team identified 276 patients initially suspected of having DME via territory-wide diabetic retinopathy screening using fundus photography. These patients were randomly assigned to either an intervention group, where follow-up examinations were assisted by the AI-OCT system, or a control group, where patients were directly referred to specialist clinics for assessment.

In the intervention group, the AI-OCT system identified 39 patients as having DME and 17 cases as uncertain. After ophthalmologists determined the need for referral, all patients were referred for DME evaluation. Ultimately, 41 cases were confirmed as DME. In comparison, the control group saw 43 cases ultimately confirmed as DME. In other words, with follow-up examinations assisted by the AI-OCT system, the proportion of patients requiring referral to specialist clinics dropped by approximately 60% to just 39.4%. (See table below)

Aiming for broad healthcare adoption

Dr Simon Szeto Ka-ho, Assistant Professor of the Department of Ophthalmology and Visual Sciences at CU Medicine, commented: “The progression of diabetic retinopathy varies among patients. Currently, after initial screening, patients often need multiple referrals to specialists and undergo OCT examinations, making the diagnostic process lengthy and complex. During this period, some patients may already suffer irreversible vision loss. If false positive cases can be screened out early using the AI-OCT system, unnecessary referrals for these patients can be avoided, which would significantly reduce healthcare costs and prevent delays in treatment for those with urgent and confirmed diagnoses.”

Professor Carol Cheung Yim-lui from the Department of Ophthalmology and Visual Sciences at CU Medicine added: “This achievement provides a practical framework for the real-world implementation of AI-enabled tools in ophthalmology and other clinical specialties. The team hopes to see the AI-OCT system implemented across all levels of healthcare services, enabling diabetic retinopathy screening at the primary and secondary care levels and assisting in triaging suspected cases of DME at the tertiary care level. This would ensure urgent cases of DME receive earlier treatment, ultimately reducing vision loss.”

Dr Simon Szeto notes that DME is the leading cause of vision loss among people with diabetes. Using the AI-OCT system to screen DME can reduce unnecessary referrals, enabling medical resources to be more focused on patients in urgent need. It is hoped that this approach can shorten waiting times by six months to a year.

Professor Carol Cheung remarks that the team hopes to see the AI-OCT system implemented across all levels of healthcare services, enabling diabetic retinopathy screening at the primary and secondary care levels and assisting in the triage of suspected DME cases at the tertiary care level. This would help achieve the goal of early detection and timely treatment.