CBCT Based Morphological Segmentation Methods for Dental Age Estimation in Adults

Abstract

The improvement of imaging acquisition technology directly influences the development of non-invasive methods in dental age estimation (DAE). The introduction of three-dimensional imaging — especially the Cone Beam Computed-Tomography (CBCT) — created a new perspective on DAE methods, especially for adults where regressive changes must be examined thoroughly. A preliminary systematic review and meta-analysis found that the CBCT DAE method can contribute to DAE methods in adults, especially if the volumetric analysis is used. Another finding is that a common approach used for the analysis is the quantification of the ratio between pulp and tooth volumetric information. However, the methods do not factor in different regressive changes that occur in the crown and root of a tooth, neither by combination nor separation. Therefore, this study explored the reproducibility and reliability of adult's DAE methods using CBCT imaging through a step-by-step volumetric segmentation approach.

This study was performed on a CBCT scan dataset of the anterior maxillary teeth from patients of Indonesian origin. The methodology was developed using open-source software (ITK-SNAP), and the combination of human intervention and semi-automated segmentation gave the best quantified reproducibility value. Consequently, this approach was used to develop the 4-Part Tooth Segmentation (SG⁴_t) method. The anterior maxillary lateral incisor presented the highest value of R². In this study phase, an additional regression model emerged as a tool for age estimation and can also be used in challenging cases such as when teeth are found fragmented.

The expansion of the SG⁴_t method was concentrated to the crown area, where attrition can be quantified using the enamel volume. This 3-Part Crown Segmentation (SG³cr) method gave the highest R² to the anterior maxillary canine. Furthermore, the 5-Part Tooth Segmentation (SG⁵_t) method was developed by combining the previous significant variables. Additionally, subsequent supervised machine learning (SML) analyses were tested. It was found that the utilisation of SML did not significantly improve the method's reliability as it is comparable to the original multiple linear regression (MLR) model.

In conclusion, the overall volumetric segmentation approach applied in the adult population using CBCT scans provides a better representation of CA without introducing collinearity within the predictive variables. The SG⁴_t approach was better in analysing a whole tooth considering the labour time it takes and the minimal improvement in the model performance when compared to the SG⁵_t method. In contrast, the SG³_cr method is encouraged for age estimation using the anterior maxillary crown as enamel segmentation significantly improves the predictive model performance, albeit with a longer labour time.

Date of Award	2023
Original language	English
Awarding Institution	University of Dundee
Sponsors	Universitas Diponegoro
Supervisor	Scheila Manica (Supervisor), Mark Hector (Supervisor) & Simon Shepherd (Supervisor)