CT-based subchondral bone microstructural analysis in knee osteoarthritis via MR-guided distillation learning

Area of Science:

• Biomedical Imaging
• Radiology
• Machine Learning

Background:

• Subchondral bone analysis is crucial for knee osteoarthritis (OA) assessment.
• High-resolution MRI is ideal but not routinely available.
• Clinical CT has limitations in spatial resolution and soft-tissue contrast for reliable trabecular analysis.

Purpose of the Study:

• To develop a method enabling accurate trabecular microstructural analysis and OA diagnosis using routine clinical CT.
• To achieve reliability comparable to MR-based analysis using CT data.
• To leverage cross-modal knowledge distillation for improved CT-based analysis.

Main Methods:

• Proposed CT-based Subchondral Microstructural Analysis (CT-SMA) method.
• Utilized distillation learning to transfer knowledge from MRI to CT.
• Implemented a synthesis-based, multi-stage MR-CT registration strategy for cross-modal correspondence.
• Employed a Transformer-based model for patient-level OA diagnosis.

Main Results:

• CT-SMA demonstrated strong agreement (ICC=0.742) with MR-derived trabecular parameters.
• CT-derived parameters enabled accurate OA diagnosis with an AUC of 0.883.
• Outperformed CT-based prediction without distillation (AUC=0.778).

Conclusions:

• Routine clinical CT can support reliable subchondral bone analysis using the CT-SMA method.
• This approach provides a practical foundation for large-scale knee osteoarthritis studies.
• CT-SMA bridges the gap between clinical CT limitations and the need for precise microstructural analysis.