Osteoporosis remains underdiagnosed and undertreated despite established screening pathways. Many eligible patients are not referred for testing, and fracture risk is often misclassified when bone mineral density is assessed with dual-energy X-ray absorptiometry. Routine CT examinations offer a chance to evaluate trabecular bone without additional imaging, but variation in scanner settings and models has limited wider use. A large retrospective analysis developed and validated an automated approach that quantifies vertebral trabecular attenuation on CT, corrects for protocol differences and establishes normative values and diagnostic thresholds. The findings outline a scalable framework for opportunistic screening that accounts for age, sex, race and ethnicity and technical heterogeneity in real-world practice.
Automated Pipeline and Cross-Scanner Calibration
A deep learning system was trained to automatically place three-dimensional regions of interest within thoracic and lumbar vertebral bodies and quantify trabecular attenuation. Automated placement was validated against radiologist review on 1496 vertebrae and achieved greater than 99% accuracy. The dataset comprised 538 946 non-contrast CT examinations from 283 499 patients acquired on 43 scanner models using six tube voltages. To address technical variability, attenuation values were first normalised to a 120 kVp reference, then calibrated by scanner model.
Must Read: AI Enables Osteoporosis Screening in Low-Dose Chest CT
Tube voltage exerted a larger influence on measured attenuation than scanner model. Values at 80 kVp and 120 kVp differed by 23%, whereas model-specific correction was less than 5% for 31 of 43 models and less than 10% for 38 of 43 models. This two-step harmonisation enabled analysis across diverse protocols without reliance on phantom measurements, which are impractical for retrospective studies. By correcting for protocol and hardware differences, the pipeline supports consistent application in multi-scanner environments and reduces the risk of systematic over- or underestimation of trabecular attenuation.
Normative Values and Diagnostic Thresholds
Normative trabecular attenuation values were generated for each thoracic and lumbar vertebra, stratified by sex, race, ethnicity and five-year age bands. The approach used all eligible CT series per examination to maximise statistical power. To align CT-based attenuation with established categories from dual-energy X-ray absorptiometry, thresholds were derived using the reported World Health Organization prevalences for osteoporosis and osteopenia in women aged 60–69 years. The 22nd and 65th percentiles of attenuation in this reference group were calculated for each vertebra, yielding vertebra-specific cut-offs that correspond to osteoporosis and low bone mass. These thresholds can be used in both women and men, consistent with evidence supporting the application of female reference databases to assess osteoporosis in men.
Representativeness of the CT cohort relative to the broader screened population was examined by comparing osteoporosis prevalence among all women aged 60–69 years who underwent dual-energy X-ray absorptiometry at the study institution with the subset who also had a CT within 12 months. The prevalence values were similar with no significant difference, supporting generalisability of the percentile-based mapping. The analysis also showed that attenuation varies by vertebral level. It was highest at T1 and lowest at L3, with age-related reductions greatest in more caudal vertebrae. In women, mean declines ranged from 2.3 Hounsfield units per year at T1 to 2.9 at L5, while in men the corresponding decreases were 1.2 and 2.1 per year. These patterns reinforce the need for vertebra-specific thresholds rather than a single universal cut-off.
Demographic Patterns Across Age, Sex and Ethnicity
The study included 283 499 patients with a mean age of 59 years ± 15, of whom 51.2% were female. White patients comprised 55.5% of the cohort, the three largest racial groups were Asian, Black and White. For 31.0% of patients, race data were unavailable. Age-related declines in trabecular attenuation were observed in both sexes. In younger adults under 50 years, women had higher attenuation than men, whereas in those older than 50 years attenuation was higher in men. In men, attenuation decreased almost linearly with age. In women, the decline was steeper around menopause, indicating a more pronounced reduction in trabecular bone density in the perimenopausal and postmenopausal period.
Attenuation values differed by race. In both sexes, values were highest in Black patients, followed by Asian patients and lowest in White patients. For example, in women aged 65–69 years, mean L1 attenuation was 131 Hounsfield units ± 44 in Black patients compared with 108 ± 42 in Asian patients and 98 ± 32 in White patients. Ethnicity also influenced attenuation in men: Hispanic men had higher values than non-Hispanic men, whereas attenuation in Hispanic and non-Hispanic women was similar in the studied age group. These findings highlight demographic variation that is relevant when interpreting opportunistic CT measurements and underscore the value of normative datasets stratified by age, sex, race and ethnicity.
An automated deep learning method quantified vertebral trabecular attenuation on routine CT and introduced practical corrections for tube voltage and scanner model, enabling harmonised analysis across diverse protocols. Vertebra-specific normative values and percentile-based diagnostic thresholds aligned with established categories were defined in a large, heterogeneous cohort. Attenuation decreased with age, showed sex-specific trajectories and varied across racial and ethnic groups. These results demonstrate a feasible pathway for opportunistic osteoporosis screening using existing CT data, with potential to improve identification of low bone mineral density where conventional referral and testing rates remain low.
Source: Radiology
Image Credit: iStock
