Diagnostic performance of paraspinal muscle fat-free cross-sectional area versus vertebral bone quality score for osteoporosis identification on lumbar MRI

To compare the diagnostic performance of the vertebral bone quality (VBQ) score and quantitative paraspinal muscle parameters derived from lumbar MRI for osteoporosis identification. We further evaluated whether muscle-derived parameters provide incremental diagnostic value beyond clinical variables and whether combined vertebral and muscle measurements improve discrimination. This retrospective cross-sectional study analyzed 150 patients who underwent both lumbar magnetic resonance imaging (MRI) and dual-energy X-ray absorptiometry (DXA). The lowest T-score was defined as the minimum of the femoral neck, total hip, and individual lumbar (L1–L4) T-scores, and osteoporosis was defined as a lowest T-score ≤ − 2.5. MRI-based measurements included the L1–L4 VBQ score and quantitative paraspinal muscle parameters measured on axial T2-weighted MRI, including cross-sectional area (CSA), fat-free cross-sectional area (FCSA), and fat fraction (FF) of the erector spinae plus multifidus (ES + MF) and psoas major (PM). Associations were examined using Spearman correlation and logistic regression analyses. Diagnostic performance was evaluated using receiver operating characteristic (ROC) curve analysis with pairwise comparisons by the DeLong test and internal validation by bootstrap resampling. Measurement reliability was assessed using intra- and inter-observer intraclass correlation coefficients (ICCs). Patients with osteoporosis exhibited higher VBQ scores ( P = 0.019) and significantly lower CSA and FCSA values of both the erector spinae plus multifidus (ES + MF) and psoas major (PM) (all P < 0.001). The lowest DXA T-score showed a weak inverse correlation with VBQ ( ρ = −0.273) but moderate positive correlations with paraspinal muscle CSA and FCSA, with the strongest association observed for PM FCSA ( ρ = 0.518). ES + MF FCSA and PM FCSA were strongly correlated with each other ( ρ = 0.768). In multivariable logistic regression analyses, only FCSA parameters remained independently associated with osteoporosis, whereas VBQ was not independently associated after adjustment for age, sex, and body mass index. ROC analysis demonstrated that PM FCSA (AUC = 0.761) and ES + MF FCSA (AUC = 0.759) provided better discrimination than VBQ (AUC = 0.611). The combined VBQ–PM FCSA model achieved the highest AUC among imaging-based models (0.785); however, its performance did not differ significantly from that of PM FCSA alone ( P = 0.328), and bootstrap internal validation yielded an optimism-corrected AUC of 0.776. When added to a basic clinical model (age, sex, and BMI), PM FCSA significantly improved discrimination (AUC from 0.768 to 0.791; likelihood-ratio test P = 0.002), whereas adding VBQ did not ( P = 0.098). Furthermore, the likelihood-ratio test confirmed that adding VBQ to PM FCSA alone significantly improved model fit ( P = 0.002), and the combined model significantly outperformed VBQ alone (DeLong P < 0.001). Paraspinal muscle fat-free cross-sectional area, particularly of the psoas major, was independently associated with osteoporosis and demonstrated higher diagnostic performance than the VBQ score. Although combined models showed numerically higher AUCs, the incremental value of adding VBQ to FCSA was not statistically significant, and whether these biomarkers provide complementary information requires further validation. FCSA may serve as a useful MRI-derived quantitative imaging biomarker for osteoporosis identification using routinely acquired axial T2-weighted lumbar MRI.