Development and Validation of a Machine Learning Algorithm for Automated Detection of Diabetic Retinopathy from Retinal Images Collected in North West Nigeria

Diabetic retinopathy is a leading cause of preventable blindness in Nigeria, and the shortage of ophthalmologists capable of retinal image interpretation in the North West zone creates a significant diagnostic gap. This study develops and validates a convolutional neural network algorithm for automated detection and grading of diabetic retinopathy from fundus retinal images collected in the North West zone. A dataset of two thousand two hundred and forty retinal fundus images was collected from diabetic patients attending the Aminu Kano Teaching Hospital and Sir Yahaya Memorial Hospital in Kano State, using a non-mydriatic fundus camera. Images were graded by two independent ophthalmologists using the International Clinical Diabetic Retinopathy severity scale as ground truth. A ResNet-50 architecture pretrained on ImageNet was fine-tuned for five-class diabetic retinopathy severity classification using transfer learning. Data augmentation strategies including rotation, flipping, brightness adjustment, and Gaussian noise addition were applied to address class imbalance. Model performance was evaluated by five-fold cross-validation. The model achieved an area under the receiver operating characteristic curve of 0.94 for any diabetic retinopathy detection, with sensitivity of 91.3 percent and specificity of 89.6 percent. Performance for referable diabetic retinopathy detection, defined as moderate or worse severity, achieved an AUC of 0.97. Gradient-weighted class activation mapping confirmed that the model attended to clinically relevant retinal features including microaneurysms and neovascularization. The study demonstrates that a clinically viable automated diabetic retinopathy screening tool can be trained on Nigerian patient images, supporting deployment in low-resource ophthalmological screening programmes. Keywords: diabetic retinopathy, convolutional neural network, retinal imaging, automated screening, Kano State.