Ứng dụng mô hình học sâu để phát hiện và phân loại gãy cổ xương đùi và gãy liên mấu chuyển trên X-quang

Abstract

Femoral neck and intertrochanteric fractures are common injuries in hip trauma. Accurate detection and classification play an important role in selecting appropriate treatment methods but largely depend on physician experience. This study utilized the RF-DETR model as the foundation for developing a deep learning model to detect and classify femoral neck fractures and intertrochanteric fractures on radiographic images. The RF-DETR model was trained and internally validated on 1,483 publicly available radiographic images. External evaluation was performed on 93 hospital radiographs, with the reference standard defined by the consensus of two orthopedic specialists. On the internal validation set, the model achieved an mAP@50 of 93.9%, precision of 91.8%, and recall of 90.6%. On the hospital dataset, the model achieved an accuracy of 95.7% (95% CI: 89.5 – 98.3%), higher compared to general practitioners at 78.5% (95% CI: 69.1 – 85.6%), p < 0.001 (McNemar test), with a kappa coefficient of 0.936. Aforementioned results suggest the RF-DETR deep learning model demonstrated high performance and shows potential for supporting clinical diagnosis.