AI Summary & Key Takeaways
AI InsightFurther validation in larger, diverse cohorts and integration with other biomarkers are needed to assess its utility in legal and administrative frameworks.
📌 3 Key Takeaways
Abstract
Aims: This study aimed to evaluate whether L3 vertebra bone density measured using lumbar computed tomography (CT) can serve as a potential biomarker for estimating the chronological age of adults over 20 years of age. Methods: We retrospectively analyzed lumbar CT images of 925 individuals (485 men and 440 women) from 2024 to 2025 from three medical centers and measured bone density at the axial midpoint of the L3 vertebral body using CT. Pearson’s correlation and piecewise regression were used to evaluate age-CT density relationships and identify physiological breakpoints. Based on these, we developed a logistic regression model to classify individuals across sex-specific thresholds (49 years old for women and 56 years old for men). To evaluate the legal or administrative applicability of age estimation, four classification models (logistic regression, random forest, support vector cliassifier (SVC), and XGBoost) were compared using a legally significant 65-year cutoff. For continuous age estimation, five regression algorithms-XGBoost, random forest, support vector regressor (SVR) with radial basis function (RBF), ridge, and linear regression-were trained and validated using a five-fold cross-validation scheme, an internal test set, and an independent external dataset. To capture the stability of our findings, 95% confidence intervals were computed for all validation phases. Results: A strong negative correlation was observed between chronological age and CT density in both sexes (r=-0.75 in men, r=-0.85 in women). Logistic regression demonstrated good categorical classification performance, with accuracies of 82% in men and 92% in women. For the female cohort, the best predictive performance was achieved by SVR (RBF) in cross validation (CV) (MAE: 7.734), linear regression in the single test (MAE: 8.032), and linear regression in external validation (MAE: 6.368, R2: 0.717). For the male cohort, random forest led in CV (MAE: 9.388), and SVR (RBF) excelled in both the single test (MAE: 9.703) and external validation (MAE: 7.521, R2: 0.721). Conclusion: The density of the L3 vertebral body measured using routine CT scans is a promising biomarker for age estimation, showing strong categorical classification and predictive utility, especially around the 65-year legal threshold for women. Although its use as a standalone tool for definitive estimation is limited, CT density shows promise as a complementary parameter for age estimation. Further validation in larger, diverse cohorts and integration with other biomarkers are needed to assess its utility in legal and administrative frameworks.