Age Estimation from Pediatric Panoramic Dental Images with CNNs and LightGBM

Abstract

Age estimation of children from dental images is one of the most reliable methods and Demirjian's approach is commonly used for this purpose. Demirjian method estimates the tooth development state and age according to the predefined tables and rules. In this study, we propose to estimate the tooth development stages with Convolutional Neural Networks (CNN) and age with machine learning. A CNN is trained from scratch to classify the stages of hidden teeth at the left mandibula and age is estimated with LightGBM regressor according to the stages. The numerical results show that the proposed method outperforms Demirjian method with age estimation with 10.85 months error.