Detecting Pediatric Wrist Fractures

This study evaluated the effectiveness of a deep learning object detection model, Faster R-CNN, in classifying pediatric wrist fractures, including subtle buckle fractures. The model achieved high accuracy, sensitivity, and specificity in identifying fractures and significantly improved resident radiologist performance, enhancing accuracy from 80% to 93% in detecting any fracture and from 69% to 92% in detecting buckle fractures. The use of AI predictions also demonstrated that radiologists outperformed the AI in cases of disagreement, highlighting the model’s potential as an augmentation tool.