Deep convolutional neural network for tooth segmentation in X-ray cone beam CT

Abstract

Dental radiography analysis plays an important role in clinical diagnosis, treatment and surgery such as implant, orthodontics. For dental radiography analysis, 2-Dimensional X-ray panorama image is mainly used and recently the use of dental Cone Beam Computed Tomography (CBCT) is increasing. If a tooth can be segmented from a Cone Beam Computed Tomography image with 3-dimensional information, it is a great help for dental care. Although there are tooth segmentation methods based on conventional image processing algorithms, there is a problem that only the upper part of the tooth is segmented or the tooth segmentation is not completely automated. Therefore, in order to try a new approach unlike the existing methods, we applied the convolutional neural network model, which has been attracting attention in computer vision. In this thesis, a deep convolutional neural network model for tooth segmentation in 3D computed tomography was proposed. The proposed model is an improved structure of neural network for image segmentation. A convolutional neural network model with high tooth segmentation performance was proposed by applying feature map concatenation and residual layer to existing image segmentation model. And, the tooth segmentation performance of the proposed model was compared with that of the conventional model by using the dice coefficient, which is widely used in the performance evaluation of bio-image segmentation.