Developing an Artificial Intelligence Decision Support Tool for Orthopaedic Fracture Classification Using Plain Radiographs from Nigerian Clinical Datasets

Accurate fracture classification is essential for guiding surgical decision-making in orthopaedics, yet inter-observer variability in classification systems such as AO/OTA and Garden classifications poses a recognised challenge. Artificial intelligence (AI) tools based on convolutional neural networks (CNNs) show promise for automated fracture classification, but existing models are trained predominantly on non-African radiographic datasets, raising concerns about generalisability and equity. This study develops and validates an AI-powered decision support tool for orthopaedic fracture classification trained on a Nigerian radiographic dataset from six teaching hospitals spanning Lagos, Kano, Ibadan, Enugu, Port Harcourt, and Abuja. A retrospective dataset of 20,000 annotated plain radiographs of femoral, tibial, humeral, and wrist fractures will be assembled, with ground-truth labels established by consensus of three fellowship-trained orthopaedic surgeons. A multi-architecture CNN ensemble (ResNet-50, DenseNet-121, EfficientNet-B4) will be trained, validated, and externally tested. Model performance will be evaluated using area under the receiver operating characteristic curve (AUC-ROC), sensitivity, specificity, and inter-rater agreement (Cohen's kappa) against expert classification. Explainability analysis using gradient-weighted class activation mapping (Grad-CAM) will be applied to ensure clinical interpretability. The study hypothesises that the Nigerian-trained AI model demonstrates significantly superior performance on local radiographic datasets compared to models trained on non-African data. Outputs include an open-source model architecture and a clinical implementation framework for deployment at resource-limited Nigerian hospitals. Keywords: artificial intelligence, fracture classification, convolutional neural network, orthopaedic decision support, Nigerian radiographs.