The Use of AI Boosts Diabetic Retinopathy Screening and Follow-up
The Use of AI Boosts Diabetic Retinopathy Screening and Follow-up : Diabetic retinopathy (DR) is a condition affecting the retina, the thin membrane crucial for vision that covers the back of the eye. It can lead to vision impairment and even blindness. Early diagnosis is crucial, and endocrinologists recommend ophthalmological monitoring for diabetic patients. A study conducted at the Johns Hopkins Children's Center reveals that AI-based eye exams enhance the screening and follow-up of DR.
Studies have shown that only 35 to 72% of young diabetic individuals undergo recommended screenings, with even higher care gap rates among minorities and disadvantaged communities.
According to WHO and the International Diabetes Federation (IDF) predictions, diabetes will affect 240 million people worldwide by 2025. Currently, it affects 34 million in the United States, the majority of whom do not seek eye examinations.
While the prevalence of DR is lower in young diabetic individuals, it affects approximately 4 to 9% of those with type 1 diabetes and 4 to 15% of those with type 2 diabetes, according to researchers.
Led by Dr. Risa M. Wolf from the Johns Hopkins School of Medicine in Baltimore, the researchers conducted a randomized trial on young patients from the Johns Hopkins Pediatric Diabetes Center with type 1 and type 2 diabetes, named ACCESS (AI for Children's Diabetic Eye Exams Study). With parental consent, they recruited 164 participants aged 8 to 21 years (average 15.5 years, 58% girls). The racially diverse population included 35% Black participants and 6% Hispanic participants. Approximately 47% of participants were covered by Medicaid, indicating a significant representation of lower socio-economic backgrounds. The majority (73%) had type 1 diabetes.
Participants were randomly assigned to two groups:
- A control group of 83 patients referred to an optometrist or ophthalmologist for an eye exam, with instructions on scheduling the appointment.
- For the remaining 81 patients, the eye exam was conducted during the endocrinologist visit using an autonomous AI tool, IDx-DR, FDA-approved in 2018 for detecting diabetic retinopathy in those over 21.
The system guides the operator in capturing two color fundus images, one centered on the fovea and the other on the optic nerve, using an image quality algorithm, prompting retakes when the images are of insufficient quality. The process takes about 10 minutes, after which the AI provides one of the following results within 60 seconds: "DR present, refer to a specialist," "DR not present, retest in 12 months," or "insufficient image quality."
Results from 25 out of the 81 participants in the latter group, 31%, indicated the presence of retinopathy. Sixteen of them, 64%, subsequently scheduled an eye professional appointment.
In the first group advised to see an eye specialist, only 22% had done so six months later.
Limitations: The authors note that the autonomous AI used in their study is not FDA-approved for those under 21. One of the study's co-authors, Michael D. Abramoff, is the founder of Digital Diagnostics, the company marketing the medical device IDx-DR.
Results: The study reveals that young patients with a positive DR screening using an autonomous AI tool were more likely to consult an ophthalmologist. AI can not only optimize diabetic retinopathy diagnosis and prevent progression but also enhance health equity.