Age estimation of children and adolescents from mandibles using machine learning.

Age estimation is a crucial step in forensic identification, particularly in scenarios where dental structures may be absent. This study aimed to develop and evaluate supervised machine learning models to predict chronological age based on mandibular morphometric measurements in children and adolescents. A sample of lateral cephalometric radiographs from 401 orthodontic patients aged between 6 and 16 years was analysed. Linear and angular mandibular measurements including the total mandibular length (Co-Pog), mandibular ramus height (Co-Go), mandibular body length (Go-Gn), and the gonial angle (Ar-Go-Me) were analysed. Eight supervised machine learning algorithms were trained to predict chronological age based on these measurements and sex. The dataset was split into training (80%) and test (20%) sets, with stratified 5-fold cross-validation to prevent overfitting. Model performance was evaluated using mean absolute error (MAE), mean squared error (MSE), root mean squared error (RMSE), and coefficient of determination (R²), with 95% confidence intervals estimated via bootstrapping. The models based on mandibular morphometric features and sex achieved a minimum MAE of 1.54 years (95% CI: 1.33-1.76) and RMSE of 1.93 (95% CI: 1.66-2.18) on the test set. Cross-validation confirmed model stability, with the Gradient Boosting Regressor achieving the best performance, showing a MAE of 1.21 (95% CI: 1.09-1.32) and R² of 0.56 (95% CI: 0.46-0.64). Total mandibular length (Co-Pog) and mandibular ramus height (Co-Go) were the most important predictors. Pairwise comparisons revealed statistically significant differences favoring ensemble methods over linear and simpler tree models. Supervised machine learning models demonstrated promising accuracy for age estimation based on mandibular measurements in growing individuals. Gradient Boosting emerged as the most effective algorithm. However, the generalizability of the models may be influenced by population-specific characteristics and the need for prior knowledge of certain predictor variables. Further external validations are recommended to enhance model applicability across diverse forensic contexts.