Opportunistic CT sarcopenia screening: multicenter validation of sex-specific T12-to-L3 conversion models

Measuring muscle parameters at the third lumbar vertebra (L3) is a standard approach for assessing sarcopenia, but these data are not always available on routine CT scans. The twelfth thoracic vertebra (T12) is a promising surrogate, but reliable conversion formulas are lacking. This study aimed to develop and validate formulas for estimating L3 muscle parameters from T12 measurements. This retrospective multicenter study included a total of 3,698 patients aged ≥ 60 years, comprising a training cohort ( n = 3,073), an internal validation cohort ( n = 216), and two external validation cohorts ( n = 163 and n = 246). Deep learning software was used to quantify skeletal muscle index (SMI), normal-attenuation SMI (NSMI), skeletal muscle density (SMD), and intramuscular adipose tissue percentage (IMAT%) at the T12 and L3 levels. Agreement was evaluated via Bland-Altman analysis and concordance correlation coefficients (CCCs). Sex-specific basic (T12-only) and advanced (incorporating age, body mass index [BMI], and interaction terms) linear regression models were developed, with regression coefficients strictly stabilized utilizing a 1000-iteration bootstrap resampling method. Significant systematic biases were observed between T12 and L3 measurements (e.g., male SMI bias: -13 cm²/m²), although agreement remained good to excellent (CCCs: 0.82–0.9). The advanced models, incorporating age and BMI, demonstrated high predictive accuracy (training R 2 : 0.68–0.83) and consistently outperformed basic models in all validation cohorts (all p < 0.001), with NSMI and SMD showing the highest magnitude of improvement. We developed reliable, sex-specific conversion formulas to estimate L3 muscle parameters from T12 CT images. These validated tools facilitate opportunistic sarcopenia screening from routine CT scans where the L3 vertebra is unavailable, promoting standardized diagnosis and broader clinical application.