Comparison of AI and dilated eye exams for the detection of diabetic retinopathy

What is it about?

The study evaluated the sensitivity and specificity of the EyeArt system and dilated eye exams performed by general ophthalmologists and retina specialists against the rigorous Early Treatment Diabetic Retinopathy Study (ETDRS) clinical reference standard on the same cohort of 521 study participants. The ETDRS reference standard was established by experts at the University of Wisconsin Reading Center using 10 fundus images per eye captured after dilation by certified photographers, whereas the EyeArt system only analyzed 2 images per eye, typically without dilation. Sensitivity, a measure of safety (percentage of patients with disease identified correctly), was 96.4% for the EyeArt system in identifying more than mild DR (mtmDR), while that of ophthalmologists’ dilated exams was 27.7% on the identical cohort of study participants. Specificity, a measure of effectiveness (percentage of patients without disease identified correctly), was 99.6% by ophthalmologists’ dilated exams compared to 88.4% with the EyeArt system. This result demonstrates that dilated exams by ophthalmologists are better at ruling out disease as evidenced by their high specificity. However, the EyeArt system, with its excellent sensitivity, is much better at identifying patients with disease (at the frontlines of care), a critical factor for a screening scenario in which patients are being identified for referral and further evaluation. The study also reported that the EyeArt system generated actionable results for more than 97% of eyes with most (85.3%) not requiring dilation. In contrast, dilated exams provided actionable results for 99.9% of eyes but required all patients to be dilated.

Featured Image

Why is it important?

Lim and co-authors concluded the paper by stating that, “Given the current low rate of compliance with the recommendation for an annual diabetic retina examination, this (EyeArt AI) system can be a useful adjunct in the detection of mtmDR (more than mild diabetic retinopathy) and appears to be more accurate than clinical ophthalmoscopy for routine retinal screening.”