Diagnosing DR with AI

Every person with diabetes is at risk of developing diabetic retinopathy (DR) – one of the leading causes of preventable blindness among working-age adults. And because the disease progresses without symptoms, early detection is difficult. Moreover, estimates suggest that 60 percent of US citizens with diabetes do not receive an annual eye exam.

We spoke with Kaushal Solanki, Founder of Eyenuk, who said AI can help alleviate the diagnostic problems around DR and beyond.

How did you get into AI eye screening? 
 

During a routine eye exam, I was told that my retinal images showed I might have signs of early glaucoma and that I needed to see an ophthalmologist fast – but I would have to wait four months to see one. I sent my images to a friend in medical school and she shared them with an ophthalmologist who was able to confirm that I may not have glaucoma but still needed to be seen by an ophthalmologist.

The experience stuck with me. At the time, I was a research scientist at Mayachitra where I led DARPA and Department of Defense projects using AI to analyze images from military drones. Using these skills, I decided to try and change the way eye disease is diagnosed and founded Eyenuk in 2010. It was actually one of the first attempts to bring AI to healthcare. 

Our EyeArt AI Eye Screening System received FDA clearance in 2020 for autonomous detection of DR and in 2023 received European Commission CE marking for detecting signs of DR and two other leading causes of blindness – age-related macular degeneration (AMD) and glaucoma. As of this year, it has been used to screen nearly 300,000 patients (one-third of them in 2023 alone) in the US and 26 other countries. EyeArt is a cloud-based system paired with a retinal fundus camera to analyze retinal images, detect signs of disease, and produce an actionable report in under 30 seconds. 

What benefits can AI offer healthcare providers and patients?
 

Unfortunately, there are multiple patient barriers to eye screening, including poor access to care, high out-of-pocket expenses, insufficient patient knowledge and awareness, and a lack of care coordination – especially among low-income populations and minorities. We believe we can help solve some of these issues by making disease screening exponentially more accessible. The system can be operated by anyone with minimal training in primary care offices, and patients with diabetes can have their eyes screened during their regular doctor’s visit. The AI tool’s ability to detect both more-than-mild and vision-threatening DR also allows providers to help determine the most optimal referral pathway for patients based on the severity of the condition. 

Offering direct screenings for DR in primary care settings greatly benefits both patients and medical professionals. In its first full year using EyeArt, for example, Mary Lanning Healthcare in rural Nebraska saw a 39 percent rise in adherence to the recommended annual diabetic eye exam. More than 20 percent of Mary Lanning patients were diagnosed with either more-than-mild DR or the more severe vision-threatening form of the disease. Those with the former were referred to an ophthalmologist, while patients with vision-threatening DR were directed to a retina specialist. The process also reduced care bottlenecks for regional ophthalmologists who could now spend less time screening patients and more time managing and treating patients who had already been screened. 

What impact can AI have in DR diagnostics?
 

We know that early detection and treatment of DR can reduce the risk of blindness by 95 percent, yet fewer than half of people with diabetes actually get the recommended annual eye exam that can detect early cases. One reason for this is that traditional disease screening requires an eye care specialist to dilate a patient’s eyes and examine the back of the eyes, which is an expensive process that often requires long wait times. The annual exams also force ophthalmologists and retina specialists to devote precious time to healthy patients, reducing their available time for focusing on those who need treatment.

Given these concerns, AI has proven to be a useful adjunct in the detection of vision-threatening disease and has been shown to be safer than clinical ophthalmoscopy for routine DR screening. The technology detects eye disease quickly, accurately and consistently, with higher sensitivity than dilated eye exams by eyecare specialists; requires minimal training; and can be easily used by anyone with a high school diploma. This means disease screening can now occur outside of eye care offices, greatly reducing the cost of, and expanding access to, screenings for potential vision loss from the disease. 

How good is AI at detecting DR? 
 

A recent study evaluated general ophthalmologists, retina specialists, and our AI system for detecting referral-warranting DR. It found that the AI system detected disease with higher sensitivity than dilated eye exams by an eye care specialist. In fact, it didn’t miss any cases of vision-threatening diabetic retinopathy.

How might AI impact the patient–doctor relationship?
 

Just as ChatGPT has triggered widespread alarm about its supposed ability to replace workers in a variety of industries, the use of AI in healthcare has triggered fears among some medical specialists who envision being replaced by robots. But the idea that AI could replace ophthalmologists or retinal surgeons is absurd. Instead, the technology’s ability to detect diseases of the eye is expanding access to eye care by freeing specialists to focus on treating disease rather than spending time on screening (early detection).

What is the future of AI in ophthalmology?
 

When it comes to detecting and preventing vision-threatening eye diseases, AI is no longer a promising but unproven technology. I predict that 2024 will see the technology more widely adopted and deployed around the world to rapidly and accurately diagnose diseases and prevent vision loss.

We are also hoping to extend our technology for the detection and prediction of early signs of neurological diseases, such as Alzheimer’s disease (AD), as well as cardiovascular diseases. Biomarkers that reflect structural changes in the retinal microvasculature have been shown to be indicative of the most commonly seen age-related cognitive disorders, including mild cognitive impairment, AD and vascular dementia. 

I’m convinced that AI will one day enable cost-effective and early diagnosis of AD and other neurodegenerative diseases, similar to the way it can detect DR at high sensitivity based on noninvasively captured fundus photographs. 

Kaushal Solanki is Founder and CEO of Eyenuk.

^{Headshot supplied by Kaushal Solanki}