Deep Learning Approaches for Automated Bone Fracture Detection

Abstract

Bone fractures necessitate prompt diagnosis and treatment to prevent complications and long-term sequelae. Traditional diagnosis involves the analysis of computed tomography (CT) images, a process often impeded by time constraints and a shortage of specialized personnel. To address these challenges, existing systems typically rely on conventional YOLO (You Only Look Once) algorithm models for fracture prediction. However, concerns regarding prediction accuracy, precise localization, and classification performance persist. In this proposed work, we introduce an advanced hybrid deep learning approach by integrating YOLOv12 for fracture segmentation and U-Net for fracture classification. YOLOv12 is employed to accurately detect and segment fractured regions in CT images, enabling precise localization of affected bone structures. The segmented fracture regions are then provided to the U-Net model, which performs detailed classification of fracture type and severity. This combined framework leverages the real-time efficiency of YOLOv12 and the strong feature extraction capability of U-Net, resulting in improved diagnostic accuracy and faster analysis. The proposed system assists healthcare professionals in making timely and reliable decisions for bone fracture management. Overall, this integration offers a comprehensive solution for automated fracture detection, precise segmentation, accurate classification, reduced diagnostic workload, and enhanced patient outcomes.