Deep learning-based automated segmentation and quantification of glenoid and humeral head defects

Area of Science:

• Orthopedic diagnostics
• Medical imaging analysis
• Artificial intelligence in healthcare

Background:

• Recurrent shoulder dislocations cause humeral head and glenoid bone defects, impacting joint stability and complicating surgery.
• Current manual assessment methods are subjective and time-consuming.

Purpose of the Study:

• To develop and validate a deep learning network for automated segmentation and quantification of shoulder bone defects using MRI.
• To improve the precision and efficiency of defect assessment compared to manual methods.

Main Methods:

• A deep defect detection network was developed using YOLOv8-seg backbone with specialized modules (edge-global attention, ASFFHead, vision transformer, dynamic mask generation).
• The model was trained and validated on 1592 MRI scans (glenoid and humeral head).
• Performance was evaluated using IoU, Dice score, precision, and Hausdorff distance, with comparisons to U-Net, R-CNN, U-NeXt, YOLOv8, and YOLOv11.

Main Results:

• The proposed model achieved a mean IoU of 89.77% and a Dice score of 94.71%.
• It demonstrated superior boundary delineation with a Hausdorff distance of 2.3 mm, outperforming baseline models.
• Ablation studies highlighted the contribution of dynamic mask generation and edge-global attention modules for accuracy in complex defect regions.

Conclusions:

• The deep learning network offers an efficient and precise solution for automated shoulder bone defect quantification from MRI.
• The model's ability to handle low contrast and complex morphologies supports personalized surgical planning and objective clinical assessment.