Deep learning for chronological age and sex prediction from dental panoramic radiographs

Abstract

Chronological age estimation and biological sex classification are two relevant tasks in a variety of clinical procedures. Both of them are useful for the identification of human remains, legal age determination, validation of birth certificates, or control of migration flows, among others. Many body indicators have been used for both age and sex estimation. However, experts agree that the oral cavity hosts anatomical structures whose development correlates with chronological age to a large degree, and these structures are different enough across males and females to enable accurate sex determination. Specifically, the teeth and mandible are considered good candidates due to their resistance and the ease of being radiologically observed. In the last decades, researchers and clinical experts developed numerous methods to convert dry-bone or radiologic measurements into an estimation of chronological age or sex. However, the application of these models has a series of disadvantages, such as the inherent subjectivity caused by the expert's measurements or the time required to conduct the process. To alleviate these problems, this PhD Thesis introduces three different approaches based on deep learning techniques for automatic age and sex estimation on dental panoramic radiographs. These methods proved to be useful for assessing the suitability not only of the whole X-ray image but also of specific skeletal structures present in the image, such as the mandible and the teeth.