Artificial Intelligence-Based Screening for Diabetic Retinopathy in Nigerian Primary Health Care Settings: Development and Validation

Artificial intelligence algorithms trained on retinal fundus images have demonstrated sensitivity and specificity comparable to ophthalmologists for diabetic retinopathy detection in high-income settings, but their performance in Nigerian primary health care contexts, characterised by distinct image quality profiles and patient demographics, has not been validated. This study develops and validates an artificial intelligence-based system for diabetic retinopathy screening adapted for Nigerian primary health care settings. A prospective, multi-phase study will be conducted. Phase one collects 5,000 fundus images from diabetic patients at four teaching hospitals across South West, South East, North West, and North Central Nigeria, with expert grading by a panel of consultant ophthalmologists as the reference standard. Phase two develops a deep learning algorithm using convolutional neural networks, trained and validated on the Nigerian dataset with attention to image quality variation inherent in low-resource imaging conditions. Phase three pilots the algorithm in five primary health centres in Ogun and Kano States, evaluating sensitivity, specificity, grading turnaround time, and health worker acceptability. The fourth phase conducts health economic modelling of the AI screening programme against the current standard of no systematic screening. Most available AI diabetic retinopathy algorithms are trained exclusively on high-resolution images from high-income countries and demonstrate performance degradation when applied to lower-quality images from resource-limited settings. Developing an algorithm optimised for Nigerian imaging conditions represents an original scientific contribution. Recommendations will address regulatory pathways for AI diagnostic devices in Nigeria, data governance frameworks, and integration with the national health information system. Keywords: artificial intelligence, diabetic retinopathy, deep learning, primary health care, Nigeria