June 2018


First artificial intelligence system approved
by the FDA to detect diabetic retinopathy

by Liz Hillman EyeWorld Senior Staff Writer

IDx-DR is the first FDA-approved diagnostic device that uses artificial intelligence technology to reach a diagnosis without physician input.
Source: IDx

“It takes about 30 seconds
for the AI system to make the
clinical diagnosis.”
—Michael Abramoff, MD, PhD

IDx-DR is the first autonomous AI device to receive FDA approval for reaching a diagnosis without physician input

Artificial intelligence (AI) technology that detects diabetic retinopathy (DR) without oversight from a specialist became the first of its kind to receive approval from the U.S. Food and Drug Administration (FDA) in April 2018.
Since then, IDx-DR (IDx, Coralville, Iowa) has been implemented at primary care clinics that are part of large hospital systems, bringing fully automated, easy-to-use technology that can detect more than mild DR or macular edema into areas that might prove more accessible and convenient to patients with diabetes.
“I think it’s huge. I’ve been working on this for 22 years. I started this as scientific research because I realized how terrible this problem is of people with diabetes going blind,” IDx founder and president Michael Abramoff, MD, PhD, said, adding that diabetic retinopathy can be managed to preserve vision, but the problem is that these patients are not coming to eye care providers for exams to identify disease in the first place. “I thought, this is an ideal way of using AI to improve this process, make it more efficient, better quality, and hopefully lower cost.”
In 2017, the American Diabetes Association recommended that patients with type 1 and type 2 diabetes, after an initial retinal exam within 5 years of diagnosis and upon diagnosis, respectively, be screened for diabetic retinopathy every 2 years.1 If diabetic retinopathy is detected, annual screening by an ophthalmologist or optometrist is recommended. The American Academy of Ophthalmology recommends annual dilated fundus exams for all patients with diabetes, and more often for those with advanced diabetic retinopathy.2 According to the Centers for Disease Control and Prevention (CDC), 30.3 million Americans have diabetes.3 The CDC’s National Health Interview Survey in 2008 found that 53.4% of adult patients with diabetes had received a dilated eye exam within the past year.4 Other estimates have placed non-adherence to retinal exam recommendations at as high as 60%.5
The IDx-DR system uses the Topcon NW400 robotic retinal camera (Topcon Medical Systems, Oakland, New Jersey). In the pivotal clinical trial, Dr. Abramoff and co-investigators showed that those who had never seen or taken a retinal image before could be trained within 4 hours to use the IDx-DR system. The AI system guides the operator to take high-quality retinal images with the NW400 camera, and the diagnostic AI analyzes the images.
“It says, this image was a little bit to the left of the disc, so you need to center it more, or retake this image because it was out of focus. It helps guide the operator to quickly get high-quality images,” Dr. Abramoff said, explaining that after four high-quality images are obtained, the diagnostic AI software determines a diagnosis. This, he said, happens automatically at the point of care with a few clicks, and the information is then sent to the patient’s electronic medical record and/or provided physically to a legally authorized provider who can discuss the results with the patient. It takes about 30 seconds for the AI system to make the clinical diagnosis, Dr. Abramoff said.
The pivotal trial for this technology, which led to FDA approval, included 900 patients with diabetes from 10 primary care sites across the country.6 In the study, the AI system was evaluated by the Wisconsin Fundus Photograph Reading Center (FPRC). Of the participants who could be evaluated by both the AI system and FPRC (819 patients), the AI system was shown to detect more than mild diabetic retinopathy at a sensitivity of 87.2% and specificity of 90.7%. Imageability was 95.6%. The study has been submitted for peer-review publication.
There are many groups working on assistive AI (AI that can help the physician decide on a certain diagnosis or treatment plan) that Dr. Abramoff said might make physicians better doctors, but it doesn’t make them more productive.
“I think an autonomous AI, which makes a decision by itself, is crucial,” he said, giving an example of one of the clinical trial sites in New Mexico. It was at a clinic where patients, prior to this technology, could have all other aspects of their diabetes exam performed, except their eye exams. For that, an ophthalmologist was 4 hours away. As a result, Dr. Abramoff said patients were foregoing retinal exams and losing vision. “We came in, trained clinic staff, and left, and then the staff was able to do high diagnostic- quality exams right there for anyone with diabetes in their practice. Now, instead of having to come to a retinal specialist, it’s being done by the AI system right there. I think it’s crucial.”
In addition to being the first AI medical diagnostic tool approved by the FDA, IDx-DR was reviewed through the FDA’s Breakthrough Devices Program, a program that seeks to expedite the FDA’s review process for technologies that are more effective in diagnosis or treatment of life-threatening or debilitating diseases.7 Malvina Eydelman, MD, director, Division of Ophthalmic and Ear, Nose and Throat Devices, Office of Device Evaluation, FDA’s Center for Devices and Radiological Health, said at the 2018 OIS@ASCRS meeting that the review process for IDx-DR took 85 days after acceptance into the program.
“The Breakthrough Device [designation] helped in making everything more efficient, allowing the FDA to come back to us sometimes within 24 hours,” Dr. Abramoff said, speculating, however, that some of the speed of the process was due to the strong results in the clinical trial. “It was an astounding experience.”
Dr. Abramoff said that he worked closely with the FDA after he founded the company 7 years ago, before even becoming part of the Breakthrough Devices Program.
“The FDA, understandably, was somewhat uncomfortable in the beginning with this idea, so we went back and forth for years, trying to see how we could bring this to patients safely and effectively, what the scientific evidence needed to be, how it fits into the clinic, etc. It was an enormous effort on the FDA’s part,” he said.
Moving forward, Dr. Abramoff said IDx has a pipeline of products. It is already developing an autonomous glaucoma detection product for primary care settings, which he said will enter clinical trials soon. Research has also been done on using autonomous AI with OCT images for detection of Alzheimer’s disease, detection of cardiovascular risk factors based on retinal vessels, and more.
“AI has been around in healthcare since the 1960s. The difference now is we have much faster computers, we are much better at algorithms, but more importantly, we have more objective input data from digital images and high-quality sensors. Because the images don’t require a doctor to interpret it and input it into a computer, you can get high-quality performance on very narrow tasks,” Dr. Abramoff said. “Everything you can image … is a prime candidate for using AI to improve diagnostics.”


1. American College of Physicians. ADA position statement updates recommendations for diabetic retinopathy. ACP Diabetes Monthly. March 2017.
2. American Academy of Ophthalmology Retina/Vitreous Panel. Diabetic Retinopathy Preferred Practice Pattern Guidelines.
3. Centers for Disease Control and Prevention. National Diabetes Statistics Report, 2017. Estimates of Diabetes and Its Burden in the United States.
4. Office of Disease Prevention and Health Promotion. 2020 Topics & Objectives – Diabetes. www.healthypeople.gov/2020/topics-objectives/topic/diabetes/objectives. Accessed April 25, 2018.
5. Murchison AP, et al. Non-adherence to eye care in people with diabetes. BMJ Open Diabetes Res Care. 2017;5:e000333
6. Abramoff M. Artificial intelligence for automated detection of diabetic retinopathy in primary care. 2018 Macula Society Annual Meeting.
7. Eydelman M. U.S. Food and Drug Administration. Brain implant for some blind people shows benefits of FDA’s Breakthrough Device Program. FDA Voice. February 27, 2018.

Editors’ note: Dr. Abramoff has financial interests with IDx.

Contact information

Abramoff: abramoff@eyediagnosis.net

First artificial intelligence system approved by the FDA to detect diabetic retinopathy First artificial intelligence system approved by the FDA to detect diabetic retinopathy
Ophthalmology News - EyeWorld Magazine
283 110
220 101
True, 6