Abstract
Supracondylar humerus fractures (SHFs) are the most common type of elbow fracture in children, typically resulting from a fall onto an outstretched hand. Accurate and timely diagnosis is critical to avoid severe complications such as neurovascular injury, compartment syndrome, and malunion. To support the automated and precise diagnosis of pediatric SHFs, we present PediaSHF-DX, a high-quality benchmark dataset comprising 10,325 de-identified elbow X-ray images from 5,163 pediatric patients. Among them, 2,015 images were carefully annotated by two experienced orthopedic surgeons using a double-blind, cross-review protocol to ensure labeling accuracy and clinical reliability. We propose an improved YOLOv11-based detection model that incorporates a LocalAttention-enhanced Bottleneck module and an optimized transmission structure to enhance small-fracture sensitivity and improve fine-grained feature extraction. The model demonstrates high performance on a separate test set of 8,310 images, achieving a precision of 0.96 and showing strong generalization and robustness across various imaging conditions. PediaSHF-DX is publicly available on Figshare and serves as a valuable resource for developing AI-driven diagnostic tools for pediatric orthopedic care.