Addressing the Overlooked: Limitations of AI in diabetic retinopathy screening and the need for preventative education in urban populations

Andra Sincari,Saieesha Yogesan,Chandrashan Perera,Jonathan Hiew,Emma Hamilton,Gerry Fegan,Fiona Wood,Yogesan Kanagasingam

DOI: https://doi.org/10.1038/s41433-024-03538-8

IF: 4.4563

2024-12-10

Eye

Abstract:This case is particularly concerning because it highlights the reality that many young people, even those living in major cities, are not receiving regular eye screenings, despite being at high risk for conditions like DR. Without the opportunistic screening, the patient's condition could have gone undetected until more severe complications arose. This points to a larger issue of under-screening in urban populations, where healthcare resources may be available, but individuals are not consistently educated or engaged in regular preventative care. While AI-based DR screening has the potential to revolutionise early detection, its success is limited by the quality of retinal images obtained [2, 3]. In our study, over 50% of the images captured using a non-mydriatic fundus camera were non-gradable due to poor image quality (Fig. 1b). Non-mydriatic cameras, though less invasive, often struggle to produce clear images, particularly in patients with smaller pupils or darker irises. This severely restricts the ability of AI systems to provide accurate and reliable diagnoses, ultimately delaying essential care for at-risk individuals.

ophthalmology