BMJ Open Ophthalmology · 9 authors, 4 centres
This summary was generated by AI from a single paper. It has not been reviewed by a clinician and is not clinical advice. Verify against the source before acting on it.
This study evaluated the real-world performance of a smartphone-based AI system for diabetic retinopathy screening in a primary care setting in Dominica. The AI system demonstrated reasonable accuracy, with a sensitivity of 77.5% and specificity of 91.5% for detecting referable diabetic retinopathy compared to a human field grader.
This prospective, cross-sectional study validated a smartphone-based AI system for diabetic retinopathy screening in a real-world primary care programme in Dominica. The study enrolled 587 participants with diabetes. Retinal images were graded by a human field grader at the point of care, and then by the AI system. The primary analysis included all participants, even those with ungradable images. The AI system's performance for detecting referable diabetic retinopathy (RDR), defined as moderate non-proliferative diabetic retinopathy or worse or diabetic macular oedema, showed a sensitivity of 77.5% and specificity of 91.5% when compared to the field grader. The findings suggest that AI can be deployed to assist diabetic screening programmes in middle-income settings with reasonable accuracy, though performance was lower than some previously reported values.
