Application of a fully deep convolutional neural network to the automation of tooth segmentation on panoramic radiographs

Objectives. To evaluate a fully deep learning mask region-based convolutional neural network (R-CNN) method for automated tooth segmentation using individual annotation of panoramic radiographs. Study Design. In total, 846 images with tooth annotations from 30 panoramic radiographs were used for training, and 20 panoramic images as the validation and test sets. An oral radiologist manually performed individual tooth annotation on the panoramic radiographs to generate the ground truth of each tooth structure. We used the augmentation technique to reduce overfitting and obtained 1024 training samples from 846 original data points. A fully deep learning method using the mask R-CNN model was implemented through a fine-tuning process to detect and localize the tooth structures. For performance evaluation, the F1 score, mean intersection over union (IoU), and visual analysis were utilized. Results. The proposed method produced an H score of 0.875 (precision: 0.858, recall: 0.893) and a mean IoU of 0.877. A visual evaluation of the segmentation method showed a close resemblance to the ground truth. Conclusions. The method achieved high performance for automation of tooth segmentation on dental panoramic images. The proposed method might be applied in the first step of diagnosis automation and in forensic identification, which involves similar segmentation tasks.