Hybrid ViT-XGBoost Model Classifies Hip Fractures on Pelvic X-Rays with 88.31% Accuracy

Hybrid ViT-XGBoost model classified hip fractures on pelvic X-rays with 88.31% accuracy (95% CI 79.3–93.7%) on 77 test radiographs from a Jordanian centre. Outperformed CNN and transformer baselines, but needs external validation.

• Sensitivity 87.75%, specificity 94.28%, and F1-score 87.75% for three-class classification (normal, intertrochanteric, femoral neck).
• Outperformed all evaluated CNN (MobileNetV2, DenseNet121) and transformer (ViT, DeiT, Swin) baselines.
• First hip fracture classification AI from the Middle East; single-centre, external validation required.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.