Glenoid segmentation from computed tomography scans based on a 2-stage deep learning model for glenoid bone loss...

Area of Science:

• Orthopedic Surgery
• Radiology
• Artificial Intelligence in Medicine

Background:

• Current computed tomography (CT) methods for measuring glenoid bone defects have limitations.
• Accurate measurement of glenoid bone loss is crucial for treating shoulder instability.

Purpose of the Study:

• To develop an automated 2-stage deep learning model for glenoid segmentation from CT scans.
• To accurately quantify glenoid bone defect and bone loss.

Main Methods:

• A 2-stage deep learning model (ResNet for location, U-Net for segmentation) was developed.
• The model was trained and tested on CT scans from patients with and without shoulder dislocations.
• Performance was assessed using accuracy, intersection-over-union, volume error, R², and Lin concordance correlation coefficient.

Main Results:

• The 2-stage model achieved high accuracy (99.28%) in glenoid location and segmentation (IoU 0.96).
• Glenoid volume error was low (9.33%).
• High correlations were found between predicted and true values for glenoid volume (R²=0.87, CCC=0.93) and bone loss (R²=0.91, CCC=0.95).

Conclusions:

• The deep learning model effectively segments the glenoid bone from CT scans.
• The model accurately quantifies glenoid bone loss, offering a valuable tool for clinical decision-making.