This proof-of-concept study evaluated the feasibility of an AI-based age estimation model using an occlusal tooth wear parameter (ΔFwear) quantified from biofluorescence. Quantitative light-induced fluorescence (QLF) images from 104 adults (20–70 years; 2,733 teeth) were analyzed. To prevent data leakage, the dataset was split at the participant level. A random forest (RF) regressor was optimized, and recursive feature elimination with cross-validation (RFECV) identified efficient tooth subsets. Final models were validated using an independent test set, and correlations between mean ΔFwear and chronological age were assessed. Cross-validation (CV) performance peaked with three teeth; however, independent testing showed that a model incorporating seven key teeth achieved the best generalization performance. This 7-tooth model achieved a mean absolute error (MAE) of 7.49 years (95% CI: 5.90–9.17), comparable to the full 28-tooth model (MAE: 7.27 years; p = 0.79), with a stronger Pearson correlation with age (r = 0.78 vs. 0.71) and an equivalent R2 of 0.61. These findings support the feasibility of integrating ΔFwear with an interpretable machine-learning framework for non-invasive age estimation. While the reduced 7-tooth model offers analytical efficiency, further validation in larger and more diverse cohorts is required to confirm its generalizability for broader forensic or epidemiological applications.
Kim et al. (Thu,) studied this question.