One in five adults in Korea has diabetes. About 30% of them also have diabetic retinopathy, in which the retina's microvessels are damaged, and about 10% of those progress to a severe stage with a high risk of blindness. But only three in 10 patients receive regular retinal exams at an ophthalmology clinic.
Woo Se-jun, a professor of ophthalmology at Seoul National University Bundang Hospital and head of the Medical Device R&D Center, saw the root of the problem not as "diagnostic technology" but as the care delivery system in which patients do not make it to ophthalmology.
In a recent written interview with ChosunBiz, Woo said, "Diabetic retinopathy is a disease that can be diagnosed without much difficulty by simply examining the eyes at an eye clinic," adding, "The problem is that patients do not come to ophthalmology."
He said, "Many patients keep up with internal medicine visits but put off eye exams," and added, "After seeing several times how young patients who should be in the thick of their working years came to the hospital only after losing vision in both eyes, I felt an urgent need to find a way to identify high‑risk patients before they come to ophthalmology."
This awareness of the problem led not to an "AI that looks at eyes," but to the idea of an "AI that finds patients before they come to ophthalmology."
He explained, "With the current approach of recommending eye exams to all diabetes patients, there is a shortage of medical personnel and low patient participation," and "If we can identify high‑risk groups using only blood tests performed in internal medicine or health checkups that patients visit frequently, patients who truly need care can receive ophthalmology treatment in time, and medical resources can be used far more efficiently."
◇ Blood tests in internal medicine, treatment in ophthalmology… a new care pathway linked by AI
Out of that came the diabetic retinopathy prediction software "iDMas‑DR." iDMas‑DR uses AI to analyze blood biomarkers and clinical information to predict the risk of diabetic retinopathy.
Woo said, "Our blood contains countless proteins and genes, so if AI analyzes them together with various clinical information, we can greatly improve predictive accuracy."
He added, "This would allow us to first pick out patients who truly need specialized ophthalmic testing," and "If AI identifies high‑risk patients so that at least those who need it receive retinal exams in time, we expect to significantly reduce cases of blindness."
iDMas‑DR received approval from the Ministery of Food and Drug Safety in 2024. In a pivotal clinical trial of about 700 people, it recorded an overall diagnostic accuracy of 85.8%, a positive case diagnostic accuracy (sensitivity) of 94.0%, and a negative predictive value (specificity) of 79.9%.
It is now in the process of being integrated into the Seoul National University Bundang Hospital health information system (HIS). Woo said, "It is highly meaningful in itself that commercialized AI software will support real-world care within a hospital's health information system."
He added, "Until now, AI use in clinical settings has been limited, and because of personal information protection, it has been difficult for HIS to bring in external programs," and "We hope this case will lead to an environment where various medical AIs support real clinical care within hospital HIS."
Once adoption is complete, the care pathway will also change. Previously, internal medicine would stop at recommending eye exams to all diabetes patients, but going forward, after blood draws at internal medicine or health checkup centers, AI will analyze risk and refer only high‑risk patients to ophthalmology, making a new care delivery system possible.
Woo emphasized that blood‑based AI does not replace fundus photography or optical coherence tomography (OCT)‑based AI, but complements it. He explained, "In ophthalmology today, we can determine whether diabetic retinopathy is present, but it is hard to know how much it will progress," and "If AI predicts risk, we can recommend more proactive treatment to high‑risk patients."
◇ "Good technology is not enough… prove it overseas first, beyond regulations"
The biggest hurdle in development was securing medical data. Woo said, "It takes a long time and massive research funding to secure good clinical data," adding, "The United States and others are investing in this at the national level. We also need to view medical data as a core future industry asset and support it over the long term to secure global competitiveness."
Once it reached commercialization, regulation became the stumbling block. Woo co‑founded the in vitro diagnostic medical AI company Retimark with Lee Cheol-ju, principal researcher at the Korea Institute of Science and Technology (KIST), and commercialized iDMas‑DR based on fundamental technology secured through a state‑funded research project.
Woo is subject to the conflict of interest prevention law that applies to employees of public institutions. To introduce the technology of the company he co‑founded to his hospital, he had to resign from Retimark's outside directorship and sever legal ties with the company.
He said, "It took more than two years to introduce the technology," adding, "It is regrettable that, without legal hurdles, it could have been introduced faster and more easily. Fundamental measures are needed to improve the reality where many medical technologies developed at national university hospitals are shelved because of such regulations."
Challenges remain even after entering the hospital. Woo pointed out, "There is currently no appropriate reimbursement for AI medical software," and "In a structure that keeps physicians performing tasks that AI can do quickly and accurately, corporations will struggle to survive and, ultimately, no one will try to develop medical AI."
Retimark is now proving its business viability overseas. In health screenings at the Changshin factory in Vietnam, it secured a case in which asymptomatic diabetic retinopathy patients were found early and connected to treatment, and it recently completed regulatory approvals in Vietnam and Indonesia and is pursuing entry into Singapore.
Woo said, "With only the domestic market, innovative medical technologies find it hard to generate sufficient revenue," adding, "Institutional support is needed so that technologies developed in Korea can move quickly into overseas medical markets."
He went on, "Introducing AI into medicine is not a choice but a necessity," emphasizing, "Laws and systems, national health insurance reimbursement, and data utilization frameworks must change together so that technologies developed in laboratories by physicians can become medical technologies actually used to treat patients."