Deeptoothduo: Multi-Task Age-Sex Estimation and Understanding Via Panoramic Radiograph

Abstract

We proposed DeepToothDuo, a Deep Convolutional Neural Network trained with a multi-task approach to estimate age and sex from a panoramic radiograph. This reduced the number of parameters required when training the model to predict age and sex separately. Moreover, we showed that training model to simultaneously predict age and sex provided better network understanding results from SHAP. Our proposed network could predict sex with 87.38% accuracy and estimate age within 1.96 years error. The model understanding study showed that the network considered anatomical features aligned with existing human dental and anatomical studies.