Using Kvaal method and machine learning to improve adult dental age estimation with CBCT.
Authors
Affiliations (5)
Affiliations (5)
- Key laboratory of Shaanxi Province for Craniofacial Precision Medicine Research, College of Stomatology, Xi'an Jiaotong University, No. 98 XiWu Road, Xi'an, Shaanxi 710004, China; Department of Orthodontics, College of Stomatology, Xi'an Jiaotong University, No. 98 XiWu Road, Xi'an, Shaanxi 710004, China.
- Key laboratory of Shaanxi Province for Craniofacial Precision Medicine Research, College of Stomatology, Xi'an Jiaotong University, No. 98 XiWu Road, Xi'an, Shaanxi 710004, China.
- Department of Orthodontics, Shanxi Provincial People's Hospital, The Fifth Clinical Medical College of Shanxi Medical University, and School of Stomatology, Shanxi Medical University, Taiyuan, China.
- Key laboratory of Shaanxi Province for Craniofacial Precision Medicine Research, College of Stomatology, Xi'an Jiaotong University, No. 98 XiWu Road, Xi'an, Shaanxi 710004, China; Department of Orthodontics, College of Stomatology, Xi'an Jiaotong University, No. 98 XiWu Road, Xi'an, Shaanxi 710004, China. Electronic address: [email protected].
- Key laboratory of Shaanxi Province for Craniofacial Precision Medicine Research, College of Stomatology, Xi'an Jiaotong University, No. 98 XiWu Road, Xi'an, Shaanxi 710004, China. Electronic address: [email protected].
Abstract
Accurate age estimation in adults remains challenging in forensic practice. This study aimed to improve dental age estimation by using Kvaal method and machine learning with cone-beam computed tomography (CBCT). CBCT scans of 400 Northern Chinese individuals aged 21-70 years were analyzed, and Kvaal-derived indices were measured for selected teeth. Sex-specific multiple linear regression models were developed as baseline methods and compared with two machine-learning regressors: Random Forest (RF) and Extreme Gradient Boosting (XGBoost). Data measurement reliability was excellent (ICC values exceeding 0.95). Width-related indices showed stronger correlations with age than length-related indices, and Kvaal's indices exhibited significant sex differences. On the independent test set, traditional linear regression models yielded mean absolute error (MAE) of 8.16-11.22 years. Both RF and XGBoost clearly outperformed linear regression, reducing MAE by about 15-25% in most tooth position. The best-performing models for the RF model and the XGBoost model were male maxillary second premolars (tooth 15/25; MAE 6.72 years) and the male mandibular lateral incisors (tooth 32/42; MAE 7.18 years), respectively. These findings indicate that CBCT-based dental age estimation combined with machine learning modestly improves accuracy over the conventional Kvaal approach, better capturing complex patterns in age-related dental changes and providing more reliable age estimates for forensic applications in the studied population. Further validation across diverse populations is necessary. However, the proposed strategy would be a complementary option within forensic dental age estimation.