Detecting Distal Radius Fractures Using a Segmentation-Based Deep Learning Model.

Area of Science:

• Radiology
• Artificial Intelligence
• Orthopedic Surgery

Background:

• Accurate detection and assessment of distal radius fractures are crucial for effective treatment.
• Current radiographic assessment can be challenging, necessitating improved diagnostic tools.

Purpose of the Study:

• To develop and validate a deep learning model for precise distal radius fracture detection using pixel-level annotations.
• To evaluate the model's performance in a multi-hospital setting with diverse radiographic data.

Main Methods:

• A deep learning model was developed using pixel-level annotations on 3399 wrist radiographs.
• The model was trained and validated on a dataset of 3785 emergency wrist radiograph examinations from six hospitals.
• Model performance was assessed using the area under the ROC curve, with consensus from three hand surgeons serving as the gold standard.

Main Results:

• The deep learning model achieved high accuracy in detecting distal radius fractures, with an area under the ROC curve of 0.97 (CI 0.95-0.98) for examinations without a cast.
• Fracture detection was more accurate on postero-anterior radiographs compared to lateral radiographs.
• The model demonstrated robust performance across multiple hospitals and radiographic system manufacturers.

Conclusions:

• Segmentation-based deep learning models show significant potential for enhancing the accuracy of distal radius fracture detection.
• The developed model performed well in a real-world, multi-institutional setting.
• Further research involving algorithm comparison and external validation is recommended to solidify clinical utility.