Diabetic Retinopathy: Can Artificial Intelligence Provide a Better Way to Detect Disease?

Linda Brookes, MSc


June 07, 2018

Not a Fail-Safe

Approval of the IDx-DR device by the FDA was based on data from a study of retinal images from 900 patients with diabetes at 10 primary care sites.[5,6] The IDx-DR was able to detect mtmDR at a sensitivity of 87.4% and a specificity of 89.5%.[1] All participants with moderate or more severe DR were identified.[5]

Wykoff points out that because the sensitivity and specificity were both below 90%, it means that for sensitivity, 1 in 10 patients may in theory have a false-positive result, and for specificity, 1 in 10 patients may in theory have a false-negative result. "So it is not a fail-safe," he stresses. "It is very important that we educate patients and doctors who are going to be using this device that it is not a 100% reliable system. It is fallible and imperfect, and it does not replace the need for a comprehensive dilated eye examination."

Shubrook agrees, stating, "My biggest concern is, is it like the eye exams that I do in my office? The positive part of the finding is great, but I am concerned when a patient has a negative screen and then does not need to go forward to an ophthalmologist, because I feel we are missing an opportunity to give a dilated eye examination. I think I have to know the false-negative rate to know whether this system could be widely applicable."

Limitations of Use

A "huge caveat," according to Wykoff, is that the IDx-DR is indicated for detection of only mtmDR in patients with diabetes who have not been previously diagnosed with eye disease.[7] "This is not a screening for other potential eye diseases. It does not replace a comprehensive eye examination," he reiterates.

Another limitation is that it is indicated for use only with a Topcon TRC-NW400 camera. "In an ideal world, we would have a software program that could interpret images from any camera," he notes. "Right now, we are tied to one particular imaging system and one particular software program that can interpret images from it, so it is certainly limited in its scope. But it's a first step, and I think it will drive more innovation and more progress," he predicts.

The company stresses that only minimal training is required to operate the IDx-DR, which consists of a one-time standardized 4-hour program; as part of the validation study, operators at each primary care site, who were without previous experience in ocular imaging, were identified to complete this same program.

The company also emphasizes the speed of operation, claiming that the results will be available "within 1 minute."[3] Shubrook is somewhat skeptical, though. He says that he hears these kinds of claims on a regular basis.

"In primary care, there is no wiggle room," he explains. "The 15-minute appointment is the standard, and that is already insufficient. One medical assistant, nurse, or provider could be trained to use this system, but I have not seen anyone around our offices who does not have enough to do already. It is an addition to the workflow."

Legal liability in cases of misdiagnosis with AI is another issue that is yet to be resolved.[8] Using AI systems raises questions about whether the physician is responsible for the system's diagnosis or whether any blame will transfer to the system itself, the designers of the system, or the company that owns the system. Legislation in this area is still evolving, both in the United States and overseas.

Financial Considerations

According to the company, the IDx-DR will produce "tremendous cost savings" and "a new source of revenue for the practitioners who use it."[9]

Shubrook says, "In a perfect world, we would use all the tools, because cost should not be the most important issue, but unfortunately it is a real issue in primary care."

Wykoff points out that "having to own a Topcon TRC-NW400 camera is certainly going to limit accessibility in places that are strapped for resources." Although Topcon does not publish prices online, US ophthalmic equipment suppliers quote prices up to $20,000 for the camera. "It would take a while to earn any of that back," Shubrook comments.

Shubrook also notes that "reimbursement is an issue. If primary care is expected to do it, but we do not get reimbursed, we are not going to do it." According to the company, it is in discussions with the Centers for Medicare & Medicaid Services and is "working to finalize the appropriate pathways to reimbursement as soon as possible."


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.
Post as: