Sex-specific skeletal ageing trajectories and reference centiles across vertebrae and ribs in over 78,000 patients

Background Age-related bone loss is well established, but CT attenuation trajectories across the skeleton are poorly characterised. Automated segmentation makes every bone measurable, yet site-specific trajectories, and interchangeability with the conventionally used L1, remain unclear. Methods In this retrospective multicentre cross-sectional study we analysed non-contrast torso CT performed during 2024 at nine Japanese academic hospitals in the Japan Medical Image Database. Patients aged 15-100 years with abdominopelvic coverage were eligible, one examination each. Mean attenuation of 25 vertebrae and 24 ribs was measured from TotalSegmentator v2.10.0 masks. The main outcome was the age- and sex-specific distribution of attenuation at each site. We fitted sex-specific spline mean curves and quantile-regression centiles, tested the sex-by-age interaction, and assessed each site as an L1 surrogate by age-adjusted correlation, errors-in-variables conversion, and discrimination of low-for-age L1. Findings Of 239 766 examinations from 135 425 patients, 79 482 remained after restricting to torso coverage and one examination per patient, and 78 620 after 862 exclusions for age or sex (41 056 men, 37 564 women; mean age 65.3 [SD 15.8] years). Attenuation fell with age at all 49 sites in women and 46 of 49 in men, with a sex-dependent shape everywhere (all FDR-adjusted p<0.0001). Female decline was steepest at a median of 52.0 years, and the 40-80-year change was a median -34.1% versus -12.8% in men. L2 reproduced L1 closely (limits of agreement +/-45 HU); ribs did not (+/-68 to +/-82 HU). Interpretation These centiles place an automated attenuation measurement at any of 49 sites against age- and sex-matched peers. Vertebrae adjacent to L1 may replace it with a stated precision; ribs may not.