Development and diagnostic accuracy of a ratio-based multivariable CT imaging model for regional periarticular knee...

To compare the diagnostic performance of CT-derived attenuation-based and ratio-based imaging biomarkers for region-specific knee osteoporosis.

This retrospective diagnostic accuracy study included consecutive adults with knee CT and standing long-leg radiographs from a single academic institution. Exclusion criteria included prior ipsilateral fracture or surgery, incomplete periarticular coverage, non-standard acquisition parameters, age <50 years, or metabolic bone disease. CT-derived imaging biomarkers were evaluated against regional osteoporosis classifications defined by previously established knee CT attenuation thresholds calibrated to central DXA. Mean trabecular attenuation was measured in predefined femoral and tibial compartments. Two ratio-based biomarkers were derived: distal femur metaphysis-to-proximal tibial epiphysis ratio (DFM/PTE) and proximal tibial medial-to-lateral ratio (PTE-MLR). Region-specific multivariable logistic regression models were constructed. Apparent discrimination was assessed using receiver operating characteristic analysis with AUC comparison and net reclassification improvement.

The cohort included 288 adults (mean age 67.1 ± 9.0 years; 50% female); 42.0% had osteoporosis in at least one region. The combined attenuation-plus-ratio model (Model 1) demonstrated outstanding apparent discrimination for any-region (AUC 0.940), distal femoral epiphysis (0.927), and proximal tibial epiphysis osteoporosis (0.957), with significantly higher AUCs than ratio-only models (p < .001). Ratio-only models retained excellent apparent discrimination across regions (AUC 0.829-0.903) with consistent femoral and tibial performance. Net reclassification analyses showed higher apparent reclassification when DFM-HU was added.

Adding DFM attenuation to ratio-based CT models improved apparent discrimination in this cohort, whereas ratio-only models preserved substantial regional diagnostic information. External, multi-scanner validation is needed before clinical implementation.