Why does diabetic eye screening matter so much?
Over time, diabetes can damage the tiny blood vessels of the retina, the layer at the back of the eye that senses light. This is diabetic retinopathy. Early on, it does not hurt or blur your vision. Yet according to the authors, it is a leading cause of preventable blindness, especially in low- and middle-income countries.
To catch it in time, the back of the eye has to be photographed regularly and the images read by a specialist. Many countries do not have enough specialists. AI systems can already read these photos, but studies testing them with real patients in everyday care were still rare.
How was the AI tested in real clinics?
From December 2018 to March 2020, the team added the AI to Thailand's national screening program at nine primary care sites. Adults with diabetes on the national registry who were due for their eye check could take part. Of 7,940 patients screened, 7,651 were included in the analysis.
During the visit, the AI read the photos and said whether the patient should be referred to an eye doctor. As a safety step, regional retina specialists also reviewed every image. To decide who was right, expert retina specialists later set a reference answer.
Results close to those of retina specialists
For sight-threatening retinopathy, the AI gave the right answer 94.7% of the time. It caught 91.4% of patients who truly had it, a measure called sensitivity, compared with 84.8% for the retina specialists reviewing the images. Both did about equally well at avoiding false alarms: 95.4% for the AI and 95.5% for the specialists.
In total, 2,412 patients, or 31.5%, were referred to a specialist for retinopathy, diabetic macular edema (swelling in the center of the retina), an image that could not be read, or poor vision.
Key figures
- 7,651people with diabetes included in the analysis, at nine sites in Thailand
- 94.7%accuracy of the AI for sight-threatening retinopathy
- 91.4%of serious cases caught by the AI (sensitivity)
- 84.8%of serious cases caught by the retina specialists reviewing the images
Staying on top of your eye exams with Mima
In some countries, eye photos of people with diabetes can already be read by an AI, with a specialist behind it. Where you live, the exam may not work this way yet, but regular checks are still what matters.
Mima helps you keep track of your appointments and test results and prepare your questions, such as: "When was my last diabetic eye exam, and when is the next one due?" She also explains the words in your report.
What this means for people with diabetes
The study shows that an AI can give a result on the spot, during the visit, about as reliably as a specialist. In a country with few retina specialists, this could help screen more people and get those who need treatment to an eye doctor faster.
For anyone with diabetes, the takeaway is the same everywhere: regular diabetic eye exams matter, even when your vision seems fine.
What are the limits of this study?
The study was funded by Google and by a hospital in Bangkok. It measures how well images were read, not the long-term effect on patients' eyesight. Patients already known to have serious eye disease were left out.
Above all, the authors say that local conditions and clinic workflows have to be taken into account when rolling out such an AI on a large scale. A strong algorithm is not enough if the path to follow-up care does not keep up.
Frequently asked questions
What is diabetic retinopathy?
It is damage to the small blood vessels of the retina caused by diabetes. It often has no symptoms at first, which is why eye screening with retinal photos matters.
Is AI as accurate as an eye doctor for diabetic retinopathy?
In this study, the AI did as well as the retina specialists reviewing the images and caught more serious cases. A specialist is still needed to confirm and treat.
Where was this AI screening tested?
At nine community clinics in Thailand's national diabetic eye screening program, between 2018 and 2020, with 7,651 patients.
How often should people with diabetes get an eye exam?
How often depends on your situation and is decided with your doctor. The key is not to skip it, even if your vision seems fine.
The original study
Ruamviboonsuk P, Tiwari R, Sayres R, et al. “Real-time diabetic retinopathy screening by deep learning in a multisite national screening programme: a prospective interventional cohort study” The Lancet Digital Health, March 7, 2022. DOI : 10.1016/S2589-7500(22)00017-6
Read the full study in The Lancet Digital HealthThis article summarizes a scientific study and is not medical advice.