Hope in Sight

A new artificial intelligence tool helps identify diabetic retinopathy

During an appointment with her primary care physician this past July, Amy Morrow, an executive assistant and patient at OSF HealthCare, learned she had prediabetes.

Moments later, Morrow walked down the hall and placed her chin onto an artificial intelligence (AI) device called LumineticsCore. It’s the first AI tool with Food and Drug Administration approval to diagnose diabetic retinopathy — a leading cause of vision loss and blindness for people with diabetes.

In about five minutes, Morrow received her results: negative for diabetic retinopathy. “It was quick and simple,” she says.

The rise of diabetes

Some 3.6 million people in Illinois have prediabetes, and 1.3 million have diabetes (about 38% of the state’s population), according to the Illinois Department of Health. “It’s important for us to screen for diabetic retinopathy because of how prevalent diabetes is,” says Saaquib Bakhsh, MD, a vitreoretinal surgeon at Wheaton Eye Clinic.

According to the Centers for Disease Control and Prevention, almost 10 million people in the United States had diabetic retinopathy in 2021. Estimates suggest that number could increase by 50% over the next 25 years.

Over time, increased glucose levels circulating in the blood can damage the walls of arteries and veins within the body, especially organs with fragile blood vessels, such as the retina.

“If you think of the eye kind of like a camera, the retina is like the film in an old-style camera,” Bakhsh says. Damage to the blood vessels within the retina can lead to blurry vision and sometimes eventually to no vision at all, just like damage to film in a camera can lead to issues with image quality. If diabetic retinopathy is left untreated, it “almost universally will lead to blindness in its final stages,” he says.

When to screen

The American Diabetes Association recommends people with type 1 diabetes have a dilated eye exam with a specialist within five years of diagnosis, and people with type 2 diabetes have one at the time of diagnosis. Both groups should then be checked annually. Yet, many people don’t see an eye doctor annually — or ever. As Bakhsh says, “an eye exam is not exactly readily available to everyone in the population.”

Life gets busy, too. Some people might intend to go, and then 10 years pass without making an appointment, says Mark Meeker, an internal medicine physician and vice president for service line physicians at OSF HealthCare.

Having a quick AI screening readily available within a primary care office helps bridge this gap. For example, one study published in 2024 in Nature Communications found that participants — all young people with diabetes — were more likely to undergo AI screenings (rather than visit an ophthalmologist) and subsequently followup with a specialist if they had an abnormal result.

The rise of AI

The main goal of screening with the LumineticsCore device is to find signs of diabetic retinopathy early and save people’s vision, Meeker says. There are other benefits of screening as well. For example, when patients receive a positive result, they tend to become more engaged in their care.

“It really gets their attention and makes them pause and reconsider their approach to their disease management,” Meeker says.

As Morrow’s doctor explained to her, insurance should cover the screening, but there may be a fee. Screening was an easy decision for her, as both of her parents had type 2 diabetes, with worsening vision over the years after their diagnosis. “I’d rather just be safe than sorry,” Morrow says. It also helped that she didn’t have to take off work, travel to another office, or have her eyes dilated.

Is AI enough?

While the LumineticsCore device is the first of its kind to receive FDA approval, there are other devices that use AI in the works.

Still, Rama Jager, MD, managing partner at University Retina and Macula Associates, says AI can’t yet see everything. One challenge with AI is how good the image quality must be for a diagnosis. A high-quality image is sometimes difficult to obtain, Jager says.

For example, a 2024 study published in Investigative Ophthalmology & Visual Science found issues with image quality when another AI device, known as the EyeArt system, was operated by a medical assistant versus an ophthalmic photographer.

“AI just isn’t quite there yet, but I think it’s certainly going to happen, I would say, in the next few years,” Jager says. “It’s still a bit early.”

Meeker says in the first six months of implementing LumineticsCore devices in OSF offices, 28 out of 100 people with diabetes tested positive, with about a 2% false-positive rate. That’s 28% of diabetes patients who would have been at risk of losing their vision in the future without screening, he adds.

Bakhsh, a retinal specialist himself, is not familiar with the intricacies of any AI retinopathy screening devices. Still, he believes this is how screening will be done for the general population in the future. “AI is likely going to be a useful tool when applied appropriately,” he says.

One thing is certain: As with most medical issues, it’s better to act preventatively rather than retroactively, to reduce the risk of vision loss. After all, when we used to use cameras with film, think about how much easier it was to to prevent damage to the film rather than to try and save it, hoping a photo or two would come out, once it was ruined. Morrow now has a baseline vision test to compare with future tests, and she’s motivated to work with her care team to prevent diabetes and diabetic retinopathy.