The Eyer portable smartphone-based fundus camera was operationally feasible for ocular screening by trained non-ophthalmologists in elderly shelter residents, but the high prevalence of media opacities and non-diabetic retinopathy conditions compromised the AI system's performance, highlighting the need for broader screening strategies in elderly and socially vulnerable populations.
The study enrolled 38 elderly shelter residents (76 eyes), most of them male, and 39.5% had confirmed diabetes mellitus.
Trained non-ophthalmologist professionals were able to acquire retinal images with the Eyer portable fundus camera, demonstrating operational feasibility for ocular screening.
The retinal specialist found ocular abnormalities in the large majority of eyes but identified no diabetic retinopathy.
Against the specialist's assessment of any ocular abnormality as the reference standard, the EyerMaps AI system had low sensitivity, specificity, NPV, and accuracy, with a high PPV.
The high prevalence of media opacities and non-diabetic retinopathy conditions compromised AI performance, suggesting a need for broader screening strategies in elderly and socially vulnerable populations.
This small pilot study tested a portable, smartphone-based retinal camera for eye screening among 38 older residents (76 eyes) of a specialized shelter in São Paulo, Brazil. Trained staff who were not eye doctors took the photos, and the images were reviewed remotely by a retinal specialist and also analyzed by an artificial intelligence (AI) program designed to detect diabetic eye disease. Fifteen of the participants had diabetes. The specialist found some eye abnormality in 92% of eyes, most often suspected glaucoma, cataract-like clouding, changes in the retina and choroid, and signs of age-related macular degeneration. No diabetic retinopathy was found. When the AI's judgments were compared with the specialist's assessment of any abnormality, it performed poorly: it correctly flagged only about 26% of abnormal eyes, and its overall accuracy was about 26%. The authors attribute this to the many eyes with cloudy media and other non-diabetic conditions. This research suggests that portable cameras can be used by trained non-specialists to reach people who might not otherwise get eye exams. However, an AI tool built for diabetic retinopathy may not be enough in older, socially vulnerable groups with many other eye problems, and broader screening approaches may be needed. The study was small and a pilot, so the findings should be treated cautiously.
Dias J, Andrade G, Macedo J, Graziano A. (2026). Fundus camera and artificial intelligence for diabetic retinopathy screening in a São Paulo shelter.. Revista da Associacao Medica Brasileira (1992). https://doi.org/10.1590/1806-9282.20260862