Bone Fracture Classification Using a YOLOv8–ANN Hybrid Model with SHAP and LIME-Based Interpretability

Bone fractures remain a critical diagnostic challenge in orthopedic medicine, requiring precise and timely interpretation of radiographic images in conjunction with clinical evaluation. This study proposes a multimodal artificial intelligence (AI) framework that integrates a YOLOv8n-based convolutional neural network (CNN) for image analysis with an artificial neural network (ANN) trained on structured clinical data to improve fracture detection and classification. The CNN, trained on annotated X-ray images spanning seven anatomical regions, achieved an overall accuracy of 97.1%, with strong localization and classification performance. Interpretability was enhanced using Gradient-weighted Class Activation Mapping (Grad-CAM) to highlight spatial regions of diagnostic relevance. In parallel, the ANN was trained on clinical profiles from 2,873 patients—including demographic, biochemical, and diagnostic parameters—and achieved 96.13% accuracy in binary fracture prediction. To further ensure transparency, SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) were employed to quantify the contribution of individual clinical features. Comprehensive evaluation through confusion matrices, per-class performance metrics, and training dynamics confirmed the robustness and generalizability of the proposed system. By combining radiological imaging with clinical data, this framework provides an accurate, interpretable, and scalable solution for AI-assisted fracture diagnosis in orthopedic practice.