Purpose
Periodontitis is the main cause of tooth loss in adults. It is important to calculate healthy periodontal membrane area to evaluate the severity of periodontitis. The aim of this study is to develop a computer-assisted system based on convolutional neural network (CNN) to segment and calculate the tooth root surface area (RSA) on cone-beam CT(CBCT).
Methods
We presented a deep learning system to automatically identify the alveolar bone and tooth regions by applying an advanced Mask R-CNN segmentation on clinically dataset of 2000 CBCT images. Twenty-four teeth from 20 patients who required tooth extraction were selected. Before extraction, pre-treatment CBCT images of all the patients were recorded. The RSA of each tooth was calculated by CNN. After extraction, all the teeth were scanned by CBCT again. The RSA of each extracted tooth was calculated by CNN again and also calculated by medical image control system (Mimics version 15.01; Materialise, Leuven, Belgium). RSA of 24 teeth calculated using these two measurement methods were analyzed by the paired t-test (P < 0.05) and Bland-Altman plot consistency analysis.
Results
The paired t-test result showed that there was no statistical difference of RSA of 24 teeth calculated by CNN and Mimics (p > 0.05). The Bland-Altman plot test also showed the good consistency.
Conclusion
We applied Mask R-CNN to segment tooth root and calculate the RSA on CBCT. Such approach presents a novel, fast, automatic and accurate approach to measure the RSA and can be used for estimating the non-extracted teeth.