Hip Imaging Dataset for Femoral Neck Fracture Detection and Classification

10,010 hip images from 806 patients collected between 2018 and 2024. Xueyang Han authored this retrospective multicenter study using data from four Chinese hospitals to develop and validate a deep learning model for fracture classification. The dataset includes training, validation, and independent test sets used to benchmark model performance against 12 physicians.

Use Cases

• Train image classification models to detect femoral neck fractures from hip radiography or CT scans.
• Develop models for Garden classification (I-IV) using labeled fracture images to automate a subjective clinical task.
• Benchmark model performance against physician diagnostic accuracy using metrics like sensitivity, specificity, and AUC from the study results.
• Generate model interpretability heatmaps to understand features used for fracture localization in hip images.
• Use the patient cohort data for external validation studies on the robustness of automated fracture detection systems.

Strengths

• 10,010 annotated hip images provide a substantial sample for model training.
• Multicenter data from four hospitals enhances generalizability and reduces single-site bias.
• Performance was validated against 12 physicians of varying experience, providing a clinical benchmark.

Limitations

• Dataset scope is limited to 487.8 KB, indicating it contains study documentation and results, not the primary image files.
• Geographic bias is present as all data comes from four Chinese hospitals.
• The retrospective study design may introduce selection biases not present in prospective data.

Provenance

Source

figshare, authored by Xueyang Han.

Collection Method

Retrospective collection of hip images (radiography/CT) from four hospitals.

Time Range

2018–2024

Freshness

Data covers up to 2024, with the study record last updated in April 2026.

Geography

Four hospitals in China.