Machine learning assisted 5-part tooth segmentation method for CBCT-based dental age estimation in adults.

Area of Science:

• Forensic Dentistry
• Radiology
• Biometrics

Background:

• Dental age estimation (DAE) in adults using volumetric data from cone-beam computed tomography (CBCT) is an evolving field.
• The 5-Part Tooth Segmentation (SG) method enhances DAE accuracy.
• Supervised machine learning models were explored for DAE.

Purpose of the Study:

• To evaluate the effectiveness of supervised machine learning models for adult DAE using CBCT data.
• To compare Support Vector Regression (SVR) and regression tree models against multiple linear regression.
• To assess the utility of volumetric tooth measurements and sex as predictors of chronological age.

Main Methods:

• CBCT scans from 99 patients (aged 20-59.99) were analyzed.
• Eighty teeth (maxillary canine, lateral incisor, central incisor) were segmented.
• Enamel-dentine volume ratio, pulp-dentine volume ratio, tooth volume ratio, and sex were used as independent variables.

Main Results:

• No multicollinearity was detected among the predictor variables.
• Support Vector Regression (SVR) with a polynomial kernel using the maxillary lateral incisor yielded the best performance (R² = 0.73).
• The optimal model achieved a mean average error of 4.86 years and a root mean squared error of 6.05 years, though segmentation was complex and time-consuming.

Conclusions:

• Machine learning, particularly SVR with a polynomial kernel, shows promise for adult DAE using CBCT volumetric data.
• Maxillary lateral incisors are effective predictors of chronological age in this population.
• Further refinement is needed to optimize segmentation techniques and reduce the labor time for clinical application.