MultiBoneX.zip

The MultiBoneX ver1 dataset is a curated collection of 22,270 high-quality X-ray images covering 11 anatomical classes of human upper and lower limbs for bone fracture detection. The dataset was constructed by integrating multiple public sources (MURA, FrcAtlas, LERA, HBFMID) and processed through a multi-stage pipeline including data cleaning, annotation verification by expert radiologists, preprocessing, enhancement, augmentation, and class balancing. Each image is labeled hierarchically by both skeletal region and fracture status (fractured/normal), enabling use in both binary and multi-class classification tasks.
The dataset reflects real-world radiology variability, with images sourced from different X-ray machines and conditions to ensure robustness across heterogeneous clinical settings. It provides comprehensive coverage of bone structures such as elbow, forearm, wrist, hand, humerus, hip, knee, ankle, and foot.
MultiBoneX ver1 is intended to support research in deep learning, medical imaging, and computer-aided diagnosis, and is suitable for developing clinically relevant, generalizable fracture detection systems for both educational and diagnostic applications.