Automatic classification of distal radius fracture using a two-stage ensemble deep learning framework.

Area of Science:

• Orthopedic surgery
• Radiology
• Artificial intelligence in medicine

Background:

• Distal radius fractures (DRFs) are common wrist injuries.
• Intra-articular DRFs involve the joint surface and are more complex to treat than extra-articular fractures.
• Accurate differentiation is crucial for appropriate treatment planning.

Purpose of the Study:

• To develop and evaluate a two-stage deep learning framework for automatic classification of intra- and extra-articular DRFs.
• To assess the performance of the framework on posteroanterior (PA) view wrist X-rays.

Main Methods:

• A two-stage ensemble deep learning approach was employed.
• The first stage used YOLOv5 networks to detect the distal radius region of interest (ROI).
• The second stage utilized EfficientNet-B3 networks for classifying fractures within the detected ROIs as intra- or extra-articular.

Main Results:

• The framework achieved an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.82.
• Classification accuracy was 0.81, with a true positive rate of 0.83 and a false positive rate of 0.27.
• Specificity for differentiating fracture types was 0.73.

Conclusions:

• Deep learning demonstrates significant potential for automated characterization of distal radius fractures from radiographs.
• The proposed framework provides a baseline for future research, including multi-view analysis for enhanced fracture classification.