Teeth Detection Algorithm and Teeth Condition Classification Based on Convolutional Neural Networks for Dental Panoramic Radiographs

In this paper, algorithms for automatic detection of teeth and teeth condition classification are proposed for DPR Dental Panoramic Radiographs) which are taken by dental X-ray camera sensors. The images used in the paper were from the DPR database provided by Taoyuan Chang Gung Memorial Hospital in Taiwan. For teeth detection method, initial locations of each tooth are estimated, and the following horizontal and vertical calibrations are used to improve the detection results. Based on the experimental results of the 32 teeth locations of the DPR, there are 25 cases that achieved more than 90% accuracy using the algorithm, and the accuracy can be as high as 96%. Furthermore, a teeth condition classification method is proposed using CNN (Convolutional Neural Networks) model. The teeth conditions in the database are discussed and categorized for later classification. The amount of data should be very large when using a CNN model, therefore data augmentation methods such as flipping and random cropping are used. The basic CNN model was used, and then improved by applying some techniques such as model tuning and dental image contrast enhancement. With this modification, the accuracy goes from 64.08% to 90.23% which showed good performance of the methods designed in this paper. Therefore, the algorithm designed in this paper can perform well with good detection and classification accuracy.