AI-assisted age estimation from first molars: an integrated evaluation of regression, machine learning, and 1D-CNN models using CBCT-derived pulp-dentin index

Abstract Background Accurate age estimation is crucial for forensic identification, and teeth serve as durable biological markers due to their post-mortem stability. While cone-beam computed tomography (CBCT) enables precise three-dimensional assessment of age-related pulp changes, most existing studies rely on traditional regression models that do not adequately capture the non-linear complexity of dental aging. This study developed and compared a multi-model framework for adult age estimation using CBCT-derived pulp-dentin index (PDI) of first molars. Results PDI values of all four first molars from 2000 Sichuan Han adults (aged 18–65 years) showed strong negative correlations with chronological age ( r = − 0.798 to − 0.866, p < 0.05), with significant variations by sex and by maxillary versus mandibular position. The dataset was randomly partitioned into training, validation, and test sets (8:1:1 ratio). Machine learning (ML) models substantially outperformed traditional regression. Among all models, the Extreme Gradient Boosting (XGBoost) model using tooth-specific PDI achieved the best overall performance, with a mean absolute error (MAE) of 2.51 years and R² = 0.907. This represents a 51.4% reduction in prediction error compared to models using averaged PDI. The 1D-CNN model achieved an MAE of 4.24 years, outperforming the best traditional regression model. Conclusions The CBCT-derived first molar PDI is a reliable forensic age indicator. Utilising multi-tooth data significantly enhances accuracy, with the XGBoost model achieving the best predictive performance. The 1D-CNN offers a promising automated alternative that, with further optimization and larger datasets, may serve as a pragmatic tool for forensic practice, and this multi-model framework provides enhanced methodological rigour and flexibility for practical dental age assessment.