Convolutional Neural Networks can Estimate Ages from Panoramic Dental X-ray Images Accurately - RMGS

Description

The chronological age estimation from panoramic dental X-ray image is a fundamental task in forensic sciences. Some existing approaches estimated the age of juvenile patient using statistical methods, and others adopted scoring-based methods to assess the most probable age and provide a minimum age estimate which requires expensive training and is known to have poor inter-rater reliability.Here we try to speculate on the chronological ages of subjects with a wider range based on deep learning algorithms, to improve the forensic evaluation system, and to migrate the new technology of deep neural networks (DNNs) to the treatment of medical effects. We applied two types of convolution neural networks (CNNs), a regression method and a multi-class classification method, to analyze the hidden features in dental X-ray images. We collected 1,760 clinical panoramic dental X-ray images between 2 and 85 years of age to develop the model. The models were evaluated with the mean absolute error (MAE) and root-mean-squared error (RMSE) between the estimated ages and chronological age. For all age groups on test samples, the mean MAE/RMSE between the automatic estimates using the proposed two deep learning models were 5.38/5.0037and 4.3082/4.0027 years, respectively. Specifically, we are able to distinguish accurately adolescent age younger than 25 years old using regression model (MAE=2.0300), as for the other age groups, we can control the error within 1-9 years by using multi-class classification model, which approaches the state-of-the-art performance.