Automatic Segmentation and Radiologic Measurement of Distal Radius Fractures Using Deep Learning

Area of Science:

• Radiology
• Artificial Intelligence
• Orthopedic Surgery

Background:

• Deep learning is increasingly applied in medical imaging analysis.
• Distal radius fractures are common and require accurate radiologic assessment.

Purpose of the Study:

• To develop and validate a deep learning algorithm for measuring key radiologic parameters of distal radius fractures.
• To compare automated measurements with those made by orthopedic hand surgeons.

Main Methods:

• An algorithm using attention U-Net and RetinaNet was developed for radius and ulna segmentation.
• The algorithm was trained and validated on 634 wrist X-rays (AP and lateral views).
• Radiologic parameters (radial inclination, tilt, height) were automatically measured and compared to surgeon measurements.

Main Results:

• High accuracy (99.98%) and Dice Similarity Coefficient (98.07%) for radius segmentation on AP images.
• Excellent correlation coefficients (Pearson's r > 0.94, ICC > 0.96) for radial inclination and tilt.
• Good correlation for radial height (Pearson's r = 0.768, ICC = 0.868).

Conclusions:

• The deep learning algorithm accurately segments distal radius and ulna in X-rays.
• The algorithm provides reliable automatic measurements of critical radiologic parameters for distal radius fractures.