Combining machine learning algorithms to construct a new method for inferring dental age of children with missing teeth in southern China.

Abstract

Age inference is a key focus of forensic work, and traditional dental age inference methods require individuals to have a complete dental arch. However, congenital or acquired tooth loss may lead to random tooth loss in individuals, resulting in bias in age prediction. To address this issue, we validated and modified Bedek's tooth age inference method (a method for inferring the age of a population with missing teeth) for the first time in the Chinese population of children with complete dentition, congenital tooth loss, and acquired tooth loss, and constructed two new machine learning based tooth age inference methods (unilateral mandible and bilateral mandible tooth age estimation models) in this population. The unilateral mandible model was constructed using the remaining five teeth of the left mandible, excluding the lateral incisor and the second premolar of congenital tooth loss, and the first premolars and first molars of the acquired tooth loss, to estimate chronological age (the two most common types of missing teeth in the Chinese population, respectively). However, the actual types of missing teeth in the population are varied, and the information on the location of missing teeth is often replaced by the developmental morphology of the contralateral teeth. In order to augment the predictive information available to model, we further constructed a bilateral mandible model containing 14 individual mandibular teeth by filling in missing values using datawig. In the male agenesis validation group, the MAE values of the best bilateral, unilateral mandible model, and modified Bedek model were 0.641, 0.715, and 0.920, respectively. In females, the MAE values were 0.763, 0.785, and 0.990, respectively. In the male acquired tooth loss validation group, the MAE values of the three models were 0.793, 0.728, and 1.376, respectively. In females, the MAE values were 0.744, 0.779, and 1.094, respectively. Collectively, these novel odontological age-estimation frameworks provide robust, flexible solutions for forensic casework involving partial dentitions. By accommodating variable patterns of congenital and acquired tooth loss without sacrificing predictive precision, they constitute a critical advancement in the forensic identification of unknown or disputed-age individuals.