A multi-view imaging dataset for Garden classification of femoral neck fractures

Accurate Garden classification of femoral neck fractures is essential for clinical decision-making and treatment planning. However, the lack of public, expert-annotated, multi-view imaging datasets has limited the development of reproducible artificial intelligence (AI) models and cross-institutional validation studies. Here, we present a fully anonymized, clinically validated multi-view imaging dataset for femoral neck fracture Garden classification, comprising 19,940 CT-derived images from 1,579 patients distributed across six image categories: axial CT views, coronal CT views, and three-dimensional volume-rendered (3D VR) views, each provided as paired original and internal-fixator-occlusion-removed images. The dataset was constructed from retrospective spiral CT examinations collected under ethics approval No. 2024 (Research)-024, with standardized inclusion and exclusion criteria, patient-level partitioning to prevent data leakage, and independent annotations performed by two senior orthopedic surgeons with discrepancies resolved by a chief orthopedic surgeon. We systematically document the imaging acquisition protocol, annotation workflow, preprocessing procedures, and technical validation, including high inter-annotator agreement (Cohen’s κ = 0.82, 95% CI: 0.78–0.86), balanced cohort characteristics across dataset splits, accurate occlusion-region segmentation (Dice = 0.962), and expert-confirmed anatomical plausibility of all occlusion-removed images. Benchmark classification experiments on the coronal-view occlusion-removed subset demonstrated that representative convolutional and transformer-based architectures achieved area under the receiver operating characteristic curve (AUC) values exceeding 0.96, confirming the discriminative utility of the dataset for automated Garden classification. The complete dataset is publicly available through Zenodo under a CC BY 4.0 license 1 . This resource fills a critical gap in publicly available orthopedic imaging data and provides a standardized foundation for the development, benchmarking, and clinical translation of AI-assisted femoral neck fracture classification systems.