In the Journals

AI system rivals human screeners for identifying diabetic retinopathy

An artificial intelligence screening device accurately identified potential cases of diabetic retinopathy, according to findings published in Diabetes Technology & Therapeutics.

Malavika Bhaskaranand

“With the growing diabetes population, AI eye screening systems are necessary to make diabetic eye screening more accessible, efficient and cost-effective and thereby reduce incidence of preventable vision loss due to diabetes,” Malavika Bhaskaranand, PhD, director of product development at Eyenuk Inc. told Endocrine Today. “Large-scale, real-world validation of these AI systems is necessary to ensure their applicability to clinical use.”

Bhaskaranand and colleagues used the EyeArt system from Eyenuk to assess 850,908 fundus images collected from January 2014 to September 2015. Researchers used 404 primary care clinics as the sources of the images, which were consecutively gathered from 107,001 encounters with patients with diabetes. Screening results from optometrists or ophthalmologists were compared with those of the EyeArt system retrospectively.

The system identified 15,177 cases of moderate nonproliferative diabetic retinopathy, 2,625 cases of severe diabetic retinopathy and 2,819 cases of proliferative diabetic retinopathy, for a total of 20,621 referable cases. The researchers stated that 5,446 of these cases were potentially treatable, and that 98.5% of the potentially treatable cases were positively referred by the system.

Hospital corridor 
An artificial intelligence screening device accurately identified potential cases of diabetic retinopathy.
Source: Adobe Stock

Meanwhile, 72,189 cases exhibited no diabetic retinopathy and 8,816 revealed mild nonproliferative diabetic retinopathy based on the system’s analysis. An additional 5,373 cases had unknown severity.

The researchers stated that the screening system showed a sensitivity of 91.3% (95% CI, 90.9-91.7) and specificity of 91.1% (95% CI, 90.9-91.3) compared with screenings conducted by medical providers.

Conversely, fewer than 0.02% of potentially treatable instances of diabetic retinopathy (n = 16) were not identified by the EyeArt system. In addition, a safeguard put in place to identify images that were unable to be screened caused the EyeArt system to provide a positive diabetic retinopathy reading in these instances, although this occurred in 0.9% of cases.

“These results indicate that patients can be screened for diabetic retinopathy using the EyeArt system in a quick, safe and effective manner at the point of care without requiring dilation,” Bhaskaranand said. “Fast and easy-to-use point-of-care screening can help reduce incidence of vision loss due to diabetic retinopathy by making screening more accessible and improving access to eye care specialists by increasing the necessary referrals (patients requiring eye care treatment and management for diabetic retinopathy) while reducing their diabetic retinopathy screening workload.” – by Phil Neuffer

Disclosures: Bhaskaranand reports that she is an employee of Eyenuk Inc. and a co-inventor of the EyeArt system.

An artificial intelligence screening device accurately identified potential cases of diabetic retinopathy, according to findings published in Diabetes Technology & Therapeutics.

Malavika Bhaskaranand

“With the growing diabetes population, AI eye screening systems are necessary to make diabetic eye screening more accessible, efficient and cost-effective and thereby reduce incidence of preventable vision loss due to diabetes,” Malavika Bhaskaranand, PhD, director of product development at Eyenuk Inc. told Endocrine Today. “Large-scale, real-world validation of these AI systems is necessary to ensure their applicability to clinical use.”

Bhaskaranand and colleagues used the EyeArt system from Eyenuk to assess 850,908 fundus images collected from January 2014 to September 2015. Researchers used 404 primary care clinics as the sources of the images, which were consecutively gathered from 107,001 encounters with patients with diabetes. Screening results from optometrists or ophthalmologists were compared with those of the EyeArt system retrospectively.

The system identified 15,177 cases of moderate nonproliferative diabetic retinopathy, 2,625 cases of severe diabetic retinopathy and 2,819 cases of proliferative diabetic retinopathy, for a total of 20,621 referable cases. The researchers stated that 5,446 of these cases were potentially treatable, and that 98.5% of the potentially treatable cases were positively referred by the system.

Hospital corridor 
An artificial intelligence screening device accurately identified potential cases of diabetic retinopathy.
Source: Adobe Stock

Meanwhile, 72,189 cases exhibited no diabetic retinopathy and 8,816 revealed mild nonproliferative diabetic retinopathy based on the system’s analysis. An additional 5,373 cases had unknown severity.

The researchers stated that the screening system showed a sensitivity of 91.3% (95% CI, 90.9-91.7) and specificity of 91.1% (95% CI, 90.9-91.3) compared with screenings conducted by medical providers.

Conversely, fewer than 0.02% of potentially treatable instances of diabetic retinopathy (n = 16) were not identified by the EyeArt system. In addition, a safeguard put in place to identify images that were unable to be screened caused the EyeArt system to provide a positive diabetic retinopathy reading in these instances, although this occurred in 0.9% of cases.

“These results indicate that patients can be screened for diabetic retinopathy using the EyeArt system in a quick, safe and effective manner at the point of care without requiring dilation,” Bhaskaranand said. “Fast and easy-to-use point-of-care screening can help reduce incidence of vision loss due to diabetic retinopathy by making screening more accessible and improving access to eye care specialists by increasing the necessary referrals (patients requiring eye care treatment and management for diabetic retinopathy) while reducing their diabetic retinopathy screening workload.” – by Phil Neuffer

Disclosures: Bhaskaranand reports that she is an employee of Eyenuk Inc. and a co-inventor of the EyeArt system.