Meeting News

AI could assist in early detection of diabetic retinopathy

SriniVas Sadda

SAN FRANCISCO — The EyeArt artificial intelligence eye screening system showed high sensitivity and high specificity for diabetic retinopathy in a pivotal multicenter clinical trial, according to a speaker here.

“I don’t need to introduce to this audience the impact of diabetes and diabetic blindness around the world. We recognize if we could detect it early and intervene in a timely fashion, it could largely be prevented, but there are many barriers out there to screening,” SriniVas Sadda, MD, said at the American Academy of Ophthalmology annual meeting.

While some telescreening solutions have been developed using fundus cameras, they generally require reading by humans, adding expense and time to the process. Artificial intelligence systems such as EyeArt (Eyenuk) remove that barrier, sending images to the cloud and having them evaluated by AI to report if the patient needs to be referred for follow-up.

In the clinical trial, two-field non-mydriatic fundus photographs were imported into EyeArt to detect referable diabetic retinopathy and vision-threatening diabetic retinopathy.

These patients also received the clinical standard adjudicated ETDRS grading of four-wide field stereoscopic mydriatic fundus photographs, equivalent to seven-field ETDRS photographs, which were graded by the Wisconsin Fundus Photograph Reading Center.

A total of 1,674 patients were included and able to be graded by the EyeArt system, which determined 480 were DR positive.

“Overall, the sensitivity and specificity were 95% and 86.5%,” Sadda said. “The most important thing with a screening system is you don’t want to miss patients who have significant disease, and in fact, when the system thought the patient did not need to be referred, only 14 of these cases of the almost 1,200 were deemed by the referral center to be referral warranted.”

In the cases that the system determined a referral was necessary but the reading center determined it was not, the majority had at least some DR or other non-DR eye disease.

“I think it’s very clear that automated AI screening would be a value, and really it’s maybe quite the right time for that given that we are really in the AI era,” Sadda said. “The hope is these kinds of systems will actually be useful tools for us in helping us with our patients, particularly in these telescreening programs.” – by Rebecca L. Forand

 

Reference:

Sadda S, et al. AI eye screening for diabetic retinopathy: results from a pivotal multicenter clinical trial. Presented at: American Academy of Ophthalmology annual meeting; Oct. 11-15, 2019; San Francisco.

Disclosure : Sadda reports he has received NEI-funded SBIR and STTR grants in collaboration with Eyenuk.

SriniVas Sadda

SAN FRANCISCO — The EyeArt artificial intelligence eye screening system showed high sensitivity and high specificity for diabetic retinopathy in a pivotal multicenter clinical trial, according to a speaker here.

“I don’t need to introduce to this audience the impact of diabetes and diabetic blindness around the world. We recognize if we could detect it early and intervene in a timely fashion, it could largely be prevented, but there are many barriers out there to screening,” SriniVas Sadda, MD, said at the American Academy of Ophthalmology annual meeting.

While some telescreening solutions have been developed using fundus cameras, they generally require reading by humans, adding expense and time to the process. Artificial intelligence systems such as EyeArt (Eyenuk) remove that barrier, sending images to the cloud and having them evaluated by AI to report if the patient needs to be referred for follow-up.

In the clinical trial, two-field non-mydriatic fundus photographs were imported into EyeArt to detect referable diabetic retinopathy and vision-threatening diabetic retinopathy.

These patients also received the clinical standard adjudicated ETDRS grading of four-wide field stereoscopic mydriatic fundus photographs, equivalent to seven-field ETDRS photographs, which were graded by the Wisconsin Fundus Photograph Reading Center.

A total of 1,674 patients were included and able to be graded by the EyeArt system, which determined 480 were DR positive.

“Overall, the sensitivity and specificity were 95% and 86.5%,” Sadda said. “The most important thing with a screening system is you don’t want to miss patients who have significant disease, and in fact, when the system thought the patient did not need to be referred, only 14 of these cases of the almost 1,200 were deemed by the referral center to be referral warranted.”

In the cases that the system determined a referral was necessary but the reading center determined it was not, the majority had at least some DR or other non-DR eye disease.

“I think it’s very clear that automated AI screening would be a value, and really it’s maybe quite the right time for that given that we are really in the AI era,” Sadda said. “The hope is these kinds of systems will actually be useful tools for us in helping us with our patients, particularly in these telescreening programs.” – by Rebecca L. Forand

 

Reference:

Sadda S, et al. AI eye screening for diabetic retinopathy: results from a pivotal multicenter clinical trial. Presented at: American Academy of Ophthalmology annual meeting; Oct. 11-15, 2019; San Francisco.

Disclosure : Sadda reports he has received NEI-funded SBIR and STTR grants in collaboration with Eyenuk.

    See more from American Academy of Ophthalmology Meeting