Forensic Age Estimation using CBCT-Derived Mandibular Morphometrics: A Comparative Study of Regression and Machine-Learning Models

Abstract

Background: Accurate age estimation in adolescents and adults remains challenging in forensic practice once dental development is complete. Cone-beam computed tomography (CBCT) enables three-dimensional evaluation of skeletal structures and may improve age estimation without additional radiation exposure. Aim: To develop and internally validate a CBCT-based multivariate regression model for chronological age estimation using mandibular morphometrics and to compare its performance with a machine-learning approach. Materials and Methods: This retrospective study analyzed 150 CBCT scans of individuals aged 10–70 years. Mandibles were segmented using ITK-SNAP software, and standardized three-dimensional morphometric measurements were obtained. The dataset was randomly divided into a training set (n = 105) and a testing set (n = 45). Pearson’s correlation analysis and stepwise multivariate linear regression were used to develop the regression model. A Random Forest regression model was trained for comparison. Model performance was assessed using the coefficient of determination (R²), root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). Agreement between predicted and chronological age was evaluated using Bland–Altman analysis and intraclass correlation coefficient (ICC). Results: Chronological age showed a strong positive correlation with the gonial angle and a strong negative correlation with the ramus height–to–body length ratio. The final regression model retained gonial angle, bicondylar width, and ramus height–to–body length ratio as significant predictors. In the testing dataset, the regression model demonstrated excellent predictive accuracy (R² = 0.881; RMSE = 6.09 years; MAE = 5.89 years), minimal bias (−0.49 years), and excellent agreement (ICC = 0.90). The Random Forest model showed reasonable performance but did not outperform the regression model. Conclusion: CBCT-derived mandibular morphometrics enable accurate, noninvasive forensic age estimation. The regression model demonstrated superior reliability and interpretability compared with machine-learning, supporting its clinical and medico-legal applicability. Key findings 1. A regression-based model using CBCT-derived mandibular morphometrics achieved high predictive accuracy and agreement for forensic age estimation, outperforming the machine-learning approach in internal validation. 2. Mandibular shape and angular parameters, particularly gonial angle and ramus height–to–body length ratio, were more reliable indicators of chronological age than isolated linear mandibular measurements.