CT-SMA improves trabecular estimation and OA diagnosis in knee osteoarthritis cohorts.

Reliable analysis of subchondral trabecular microstructure is critical for knee osteoarthritis assessment. However, this analysis largely relies on high-resolution MRI acquired using balanced fast field echo (BFFE) sequences, which are rarely included in routine clinical protocols. Clinical CT is widely acquired, yet its limited spatial resolution and soft-tissue contrast makes direct trabecular parameter estimation unreliable. Therefore, it is specifically demanded to enable accurate trabecular microstructural analysis and osteoarthritis diagnosis using routine clinical CT, while also approaching the reliability of MR-based analysis. In this paper, we propose CT-based Subchondral Microstructural Analysis (CT-SMA) method, which utilizes distillation learning technology to transfer high-resolution structural knowledge from MR to CT while enforcing CT-only inference. The core idea of CT-SMA is to transfer microstructural knowledge learned from high-resolution MR to CT through cross-modal knowledge distillation, using a pre-trained MR-based teacher model to supervise CT-based student model on feature maps. To support effective distillation, CT-SMA further introduces a synthesis-based, multi-stage MR–CT registration strategy that establishes patch-level correspondences across modalities, despite substantial differences in resolution, contrast, and appearance. Experiments on a clinical knee imaging cohort demonstrate that CT-SMA substantially improves CT-based trabecular parameter estimation, achieving strong agreement (ICC = 0.742) with MR-derived references across key trabecular biomarkers. Moreover, when aggregated using a Transformer-based model, the regressed CT-derived parameters enable patient-level osteoarthritis diagnosis with an AUC of 0.883, substantially outperforming CT-based prediction without distillation (AUC = 0.778). These results indicate that routine clinical CT can support reliable subchondral bone analysis via proposed CT-SMA, establishing a practical foundation for large-scale studies.