Deep convolutional neural network designed for age assessment based on orthopantomography data

In this paper, we proposed an age assessment method evaluated on Malaysian children aged between 1 and 17. The approach is based on global fuzzy segmentation, local feature extraction using a projection-based feature transform and a designed deep convolutional neural networks (DCNNs) model. In the first step, a global labelling process was achieved based on fuzzy segmentation, and then, the first-to-third molar teeth were segmented. The deformation invariant features were next extracted based on an intensity projection technique. This technique provided high-order features which were invariant to rotation and partial deformation changes. Finally, the designed DCNN model extracts a large set of features in the hierarchical layers which provided scale, rotation and deformation invariance. The method using this approach was evaluated using a comprehensive and labelled orthopantomographs of 456 patients, which were captured in the Department of Dentistry and Research at Universiti Sains Islam Malaysia. The results from the analysis have suggested that the method can classify the images with high performance, which enabled automated age estimation with high accuracy.