The use of artificial intelligence in the detection of proximal femur fractures in the emergency department

Objective: This study aims to evaluate the accuracy and effectiveness of a deep learning model based on artificial intelligence in the detection of proximal femur fractures in the emergency department and to conduct a comparative analysis of its diagnostic performance against that of emergency medicine physician assistant. Material and Method: Radiographic images of the pelvis and femur obtained from patients aged 18 to 100 years who presented with trauma to the Emergency Department of Afyonkarahisar Health Sciences University Faculty of Medicine between August 15, 2024, and August 15, 2025, were included in this study. The collected images were categorized as "fracture present," "no fracture," and further classified according to fracture type. The images were labeled using the bounding box method. After labeling, the model was trained using the Yolov11 algorithm. This trained model was compared with the emergency medicine physician assistant and the gold standard. Result: A total of 4,000 images were reviewed, and 2,000 suitable images were used to train the artificial intelligence model. In the testing phase, a set of 400 images was used to compare the "fracture present/absent" assessments made by the artificial intelligence model emergency medicine physician assistant through Cohen's Kappa analysis. The resulting Kappa coefficient was 0.980, indicating a statistically significant agreement (p < 0.001). Regarding fracture classification, the artificial intelligence model achieved an accuracy of 100% for shaft fractures, 78% for femoral neck fractures, 88.8% for intertrochanteric fractures, and 59.6% for subtrochanteric fractures, compared to the assessments made by the the emergency medicine physician assistant. Conclusion: The YOLOv11-based artificial intelligence model demonstrated high accuracy in distinguishing the presence or absence of fractures and was generally successful in classifying fracture types. The relatively lower sensitivity observed in subtrochanteric fractures is attributed to the limited amount of data available for this category, highlighting the study's potential contribution to the existing literature. These findings suggest that such artificial intelligence -assisted systems may serve as supportive diagnostic tools in emergency department settings where rapid decision-making is crucial. Keywords: Artificial intelligence, Emergency Department, Femur, Fracture, YOLOv11