Enhancing Radiographic Diagnosis: A Novel AI-based Bone Fracture Detection System

This paper presents a novel AI-based system for automated bone fracture detection in radiographic images, designed to meet the accuracy, speed, and reliability demands of modern clinical practice. The proposed method integrates EfficientNet-B4 with an attention mechanism to enhance feature sensitivity, particularly in identifying subtle and complex fractures. Unlike prior approaches, this system is optimized for real-time inference and has been rigorously validated on a diverse, multi-institutional dataset comprising over 12,000 annotated radiographs. The model achieves high diagnostic performance, with a sensitivity of 96.2%, specificity of 94.8%, and an AUC of 0.983, significantly outperforming baseline CNN and hybrid models. It also delivers sub-50 millisecond inference time, making it suitable for deployment in emergency and high-throughput healthcare settings. Furthermore, the integration of Grad-CAM visualizations supports interpretability, enhancing clinical trust. These advancements address persistent challenges in fracture diagnosis, including generalizability, speed, and explainability, marking a substantial step toward AI-assisted radiology in real-world environments.