If you have diabetes, you’re at risk for a condition called diabetic retinopathy. If not treated properly, severe diabetic retinopathy can cause blindness.
Eye doctors recommend that anyone with diabetes get a dilated eye exam at least once a year to check for diabetic retinopathy, even if you don’t have any vision-related symptoms. “You can have severe disease but not be aware of any changes to the eyes,” says Dr. Lloyd P. Aiello, director of the Beetham Eye Institute at Joslin Diabetes Center in Boston.
A dilated eye exam allows the doctor to see the back of your eye, or the retina. That’s where signs of diabetic retinopathy are found. Eye doctors will use special cameras to take images of the eye and will analyze the images. Those images are given a grade based on the extent of diabetic eye disease found, if any.
With the growing number of patients in the world who have diabetes, artificial intelligence is expected to play an increasing role in detecting who has diabetic retinopathy.
Artificial intelligence is a kind of learning that can be taught to a machine, such as a computer. Nowadays, artificial intelligence, or AI, is used for facial recognition, handwriting recognition and speech recognition, among many other uses. One kind of AI used in health care is called deep learning technology. “This deep learning technology is the same artificial intelligence that name-brand tech companies, such as Google and Facebook use in their own software,” says Dr. Alan Mendelsohn of Eye Surgeons & Consultants in Hollywood, Florida.
Health care leaders are also finding more ways to use AI to make patient diagnosis and treatment more efficient. Deep learning research in ophthalmology is underway to better identify lesions and other signs of diabetic retinopathy. This is done by first helping the technology identify what images of an eye with diabetic retinopathy lesions or hemorrhages look like.
By using a computer to identify signs of diabetic retinopathy, it makes patient care more efficient, says Dr. Julia A. Haller, ophthalmologist-in-chief at Wills Eye Hospital in Philadelphia. “If you have a machine or computer that says you don’t need to do anything or you need to do something, that could save a lot of time,” she says. Otherwise, it can be all-consuming for eye doctors or trained readers of eye images to review so many photos, considering the vast number of people with diabetes – and the recommendation that those patients get an eye exam at least once a year. This is an important point as countries around the globe try to screen a larger number of people with diabetes, Aiello says.
That’s why the use of AI to analyze images of the retina seems so appealing and fits in with the already growing trend of telemedicine. “In locations with reduced access to care, such as rural communities and developing countries, primary care physicians have used telemedicine to photograph their patients’ retinas and send the images to ophthalmologists in other locales for backup diagnostics,” Mendelsohn says. But again, that takes up the time of doctors who are already pressed for time.
The initial results from studies of AI to identify diabetic retinopathy are promising, and the results will likely only get better with time, Aiello says. “This type of approach will be able to look at an image and grade it as well as our current graders and do it faster and possibly better,” Aiello says. “This will have big ramifications for diabetes care.”
In fact, at larger institutions, using AI to read photos of patients with diabetic retinopathy will likely become more predominant in the next five to 10 years, Aiello predicts. It may take longer to use this kind of technology elsewhere because approaches to health care vary so widely from state to state in the U.S., he adds.
“Naturally the medical community is especially excited about this technology because it will enable patients worldwide to receive ophthalmologic-quality diagnostics without needing to travel to a specialist,” Mendelsohn says.
However, one limitation of using AI for diabetic retinopathy screening could be image quality – something that’s already a concern for human analysis of images. “We find that for about 10 percent of our photos when we do screenings, they aren’t readable,” Haller says. It could be that that patient has a cataract or is on certain drugs that block a good view of the retina. “They may not have something wrong with them, but if we can’t read the image, they need to come in for an exam. That’s a bit of a drawback,” Haller says.
Patients also need to know that having a photo of your retina analyzed by AI doesn’t replace the need for a thorough eye exam, Haller cautions. In a face-to-face exam, your eye doctor can perform tests that help evaluate other important parts of eye health. “It’s like when you go to your pharmacy and do a blood pressure check, it doesn’t mean you can skip seeing your internist this year. There are limitations, and we still need physicians,” Haller says.
Still, the growing use of AI to analyze retinal images can help separate which patients need special attention for diabetic retinopathy and which ones don’t. The technology is also ideal to make this type of eye screening available at places other than an eye doctor’s office. In fact, Haller has a prediction. “With better technology, you could do a selfie on your cellphone” for a retinal eye screening, Haller says. “We’re not there yet, but we can all see it happening.”
Although artificial intelligence may one day help track your eye health if you have diabetes, the recommendations to follow now for better eye health remain the same. Don’t smoke, aim to keep your blood sugar under control, eat healthy and exercise, Mendelsohn advises.
Of course, make sure you get a dilated eye exam once or twice a year. Only about 60 percent of patients with diabetes in the U.S. get this exam regularly – and the percentage is even lower in underserved communities. However, there are medicines nowadays that can prevent blindness in 95 percent of patients with diabetic retinopathy, Aiello says.