Age Detection by Deep Learning from Dental Panoramic Radiographs

Abstract

The use of deep learning approaches is growing day by day in the solution of various real-world problems in engineering science. Health sciences problems are also one of the areas in that deep learning is frequently applied. Especially in digital forensics cases and anthropology, when determining the identification of living individuals or corpses, the most important specification is to state the age of the person. At the stage of determining the age, the analysis of bone or tooth development of people is two of the most trustworthy methods. Moreover, there is a distinctive interrelation between the eruption of permanent and primary teeth and the chronological age of the individual. In this study, a deep learning approach is suggested as an alternative to age determination using traditional methods. A total of 627 dental orthopantomographic images gathered from individuals between the ages of 2 and 21 were employed in this study. The data set consists of two different classes, individuals under the age of 13 and individuals aged 13 and over, who have completed the eruption of their permanent number 7 teeth. Firstly, feature extraction was operated on the data by using the Convolutional Neural Network (CNN) architecture, which is one of the deep learning approaches. Afterward the feature extraction phase, the system was completed using four different classifiers. 70% of the dataset is allocated for training while in the rest is reserved for testing. The results achieved using various evaluation metrics are presented in detail with a complexity matrix, tables, and graphs. In this study, 84% accuracy, 85% F-score, and 76% sensitivity values were reached using the Alexnet architecture and k-nearest neighbor (k-NN) algorithm. It is forecasted that the proposed system will ensure age determination in less time and abate the cost compared to traditional age determination methods. Besides, the study will both support dentists in the clinical environment and can be used in education.