Context-Aware Dual-Task Deep Network for Concurrent Bone Segmentation and Clinical Assessment to Enhance Shoulder...

Area of Science:

• Orthopedic Surgery
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Accurate preoperative planning for shoulder arthroplasty necessitates precise glenohumeral (GH) joint digital models and reliable clinical staging.
• Current methods may lack the precision required for optimal surgical outcomes.

Purpose of the Study:

• To develop and validate a dual-task deep network, xCEL-UNet, for simultaneous 3D bone reconstruction and clinical assessment of the GH joint.
• To enhance preoperative planning for shoulder replacement surgery through improved digital modeling and staging.

Main Methods:

• xCEL-UNet, a deep learning model, was designed for humerus and scapula bone reconstruction from CT scans.
• The network also assessed three clinical conditions: osteophyte size (OS), joint space reduction (JS), and humeroscapular alignment (HSA).
• Transfer learning was employed on a dataset of 571 patients to optimize segmentation and classification.

Main Results:

• The model achieved high accuracy in bone reconstruction, with median RMSEs of 0.31 mm (humerus) and 0.24 mm (scapula), and Hausdorff distances of 2.35 mm and 3.28 mm, respectively.
• Classification accuracy for clinical staging was 91% for OS, 93% for JS, and 85% for HSA.
• GradCAM visualization confirmed the network's interpretability.

Conclusions:

• The xCEL-UNet framework provides accurate 3D bone surface reconstructions of the GH joint.
• It offers dependable clinical assessments, supporting therapeutic decision-making in shoulder arthroplasty.
• This AI-driven approach enhances preoperative planning for shoulder replacement.