Deep Learning-Based Automatic Glenohumeral Joint Segmentation for Determining Whether the Hill-Sachs Lesion Is...

Area of Science:

• Orthopedic Surgery
• Radiology
• Artificial Intelligence in Medicine

Background:

• Accurate glenoid track quantification is crucial for managing anterior shoulder instability.
• Conventional CT-based methods are time-consuming (2 hours manual segmentation) and have inconsistent interobserver reliability.
• Deep learning offers potential for improving medical image analysis efficiency and accuracy.

Purpose of the Study:

• To develop and validate a deep learning framework for automated CT segmentation of shoulder dislocations.
• To quantify bone defects and glenoid parameters efficiently and consistently.
• To enhance diagnostic workflow for anterior shoulder instability.

Main Methods:

• A deep learning model adapted from TotalSegmentator was used for automated segmentation and 3D reconstruction of CT scans from 43 patients.
• Glenoid track width (GTW) and Hill-Sachs interval (HSI) were measured using the Two-Thirds Glenoid Height Technique.
• Segmentation accuracy was assessed using Dice similarity coefficient; measurement reliability was evaluated using intraclass correlation coefficient (ICC).

Main Results:

• The deep learning model achieved high segmentation accuracy (Dice > 0.95 for scapula and humerus) in just 30 seconds per case.
• GTW measurements showed excellent intra- and interobserver agreement (ICC > 0.90).
• HSI measurements demonstrated high intraobserver reliability (ICC > 0.90) and substantial interobserver agreement (ICC ≥ 0.80).

Conclusions:

• Deep learning integration streamlines the diagnostic workflow for shoulder dislocations, enabling rapid and accurate bone defect quantification.
• The Two-Thirds Glenoid Height Technique is reliable for measuring glenoid parameters on 3D models generated by the deep learning framework.
• This approach provides an efficient tool for surgical planning in anterior shoulder instability.