Statistical Shape Model for Bone-Based Cartilage Prediction: Applicability in Healthy and Pathological Knees.

Area of Science:

• Orthopedic surgery
• Medical imaging
• Computational anatomy

Background:

• Total Knee Arthroplasty (TKA) aims to improve outcomes in severe knee osteoarthritis.
• Current image-based planning relies on CT scans for bone geometry but lacks cartilage information.
• This limitation can impact intraoperative registration and surgical accuracy in TKA.

Purpose of the Study:

• To evaluate the ability of Statistical Shape Models (SSMs) to predict cartilage from bone shape.
• To assess if SSMs can capture pathological variability, including complex cartilage loss and bone deformation patterns.
• To determine the suitability of SSMs for integration into CT-based patient-specific instrumentation (PSI) and robotic surgical workflows.

Main Methods:

• Coupled bone-cartilage SSMs were trained separately on healthy and pathological knee datasets from the OAI-ZIB database.
• Classification into healthy and pathological groups was based on joint space narrowing.
• Model performance was evaluated by comparing bone fitting and cartilage prediction accuracy.

Main Results:

• Bone fitting errors were 0.27-0.32 mm and cartilage prediction errors were 0.41-0.49 mm (RMSE 0.66-0.79 mm).
• These errors are comparable to existing literature and close to inter-observer variability for MRI manual segmentation.
• SSMs demonstrated similar predictive performance for both healthy and pathological cases, indicating they can capture pathological variability, though osteophytes were not fully captured.

Conclusions:

• The developed SSMs accurately reconstructed bone and predicted cartilage in both healthy and arthritic knees.
• The models show robustness to pathological variability, making them suitable for clinical integration.
• This CT-based approach offers an alternative to MRI and avoids additional surgical steps for cartilage assessment in TKA planning.