Abstract
Purpose :
With over 100 million diabetics, India faces an increasing prevalence of diabetic retinopathy (DR) coupled with a shortage of trained eye care providers. Early detection and treatment can prevent visual loss from diabetic retinopathy. Thus, routine retinal screening of diabetics is recommended, but is challenging in low and middle income countries where access to qualified providers is lacking. Here, we evaluate the use of artificial intelligence (AI) as a tool to allow screening of diabetics for diabetic retinopathy in an Indian population.
Methods :
Between November 2020 and June 2023, a total of 142,565 adults were screened including 15,552 self-reporting diabetes. Retinal examination was conducted at 55 sites across 10 Indian states as part of corporate screening programs (53) or free eye camps (2). Technicians captured retinal images in both eyes using non-mydriatic fundus imaging (3nethra, Forus Health, Bangalore, India). The mean (±SD) age was 39.0 ± 13.7 and the majority of patients were males (58.3%). Images were securely transmitted to a server and asynchronously graded for the presence of diabetic retinopathy by expert readers (certified ophthalmologists) and a novel AI algorithm.
Results :
246,667 eyes from 142,565 adults were included in the study. Diabetic retinopathy was present in 0.44% (0.42%, 0.47%) of all eyes. Of 1,097 eyes diagnosed with diabetic retinopathy, 768 were correctly identified by AI. The novel AI algorithm demonstrated a sensitivity of 70.0% and a specificity of 92.3%.
Conclusions :
The use of a proprietary AI algorithm successfully identified diabetic patients with diabetic retinopathy across multiple Indian states. AI may represent a viable option to screen for diabetic retinopathy in low and middle income countries with a shortage of qualified eyecare providers.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.