Automatic quantification of scapular and glenoid morphology from CT scans using deep learning.

Area of Science:

• Orthopedics and Biomedical Engineering
• Artificial Intelligence in Medical Imaging
• Quantitative Anatomy

Background:

• Glenohumeral osteoarthritis diagnosis and management rely on accurate scapular and glenoid morphology assessment.
• Manual quantification of these parameters from CT images is time-consuming and prone to inter-observer variability.
• There is a need for automated, reliable methods to analyze scapular and glenoid morphology.

Purpose of the Study:

• To develop and validate an open-source deep learning model for automated quantification of scapular and glenoid morphology.
• To assess the accuracy of the model in segmenting scapulae and identifying landmarks in both normal and osteoarthritic subjects.
• To measure key morphological parameters including glenoid version, inclination, and critical shoulder angle.

Main Methods:

• Deep learning was employed for automated scapular segmentation and identification of 13 key landmarks.
• A coordinate system unaffected by osteoarthritis was established using 9 landmarks.
• Glenoid size and orientation parameters were calculated, and 5-fold cross-validation was performed on 116 scapulae.

Main Results:

• The deep learning model achieved a Dice similarity coefficient >0.97 for scapular segmentation.
• Automatic landmark positioning error was 1-2.5 mm, comparable to human raters.
• The model demonstrated high accuracy (R² 0.88-0.95) for quantifying glenoid version, inclination, critical shoulder angle, and other parameters.

Conclusions:

• The developed open-source deep learning model reliably quantifies scapular and glenoid morphology from CT scans.
• The model's accuracy is sufficient for clinical application in patients with glenohumeral osteoarthritis.
• This automated approach offers a valuable tool for objective assessment and research in shoulder pathology.