Fracture detection of distal radius using deep-learning-based dual-channel feature fusion algorithm

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Orthopedics

Background:

• Distal radius fractures are common, requiring accurate preoperative diagnosis for recovery.
• Deep learning shows promise for fracture diagnosis, but faces challenges with accuracy and occult fractures.

Purpose of the Study:

• To design an improved deep-learning model for faster and more accurate distal radius fracture diagnosis.
• To enhance detection accuracy, especially for subtle or occult fractures.

Main Methods:

• Developed a dual-channel feature fusion network based on Faster R-CNN.
• Integrated Residual Network 50 with Deformable and Separable Attention for lateral X-ray analysis.
• Employed focal loss to address sample imbalance and fused features from anteroposterior and lateral views.

Main Results:

• The dual-channel network achieved high performance: 98.5% AP50, 78.4% AP75, 96.5% accuracy, and 94.7% recall.
• Outperformed traditional models (Faster R-CNN, YOLOv4, YOLOv5s) in key metrics like precision, recall, and AP scores.
• Demonstrated superior accuracy and reliability in identifying both apparent and occult distal radius fractures.

Conclusions:

• Combining anteroposterior and lateral X-ray images improves deep learning model accuracy for distal radius fracture detection.
• The proposed model offers a valuable tool for clinical applications, enhancing diagnostic efficiency and precision.