Automated CNN-Based Tooth Segmentation in Cone-Beam CT for Dental Implant Planning

Abstract

Accurate tooth segmentation is an essential step for reconstructing the three-dimensional tooth models used in various clinical applications. In this paper, we propose a convolutional neural network (CNN) based method for fully-automatic tooth segmentation with multi-phase training and preprocessing. For multi-phase training, we defined and used sub-volumes of different sizes to produce stable and fast convergence. To deal with the cone-beam computed tomography (CBCT) images from various CBCT scanners, we used a histogram-based method as a preprocessing step to estimate the average gray density level of the bone and tooth regions. Also, we developed a posterior probability function. Regularizing the CNN models with spatial dropout layers and replacing the convolutional layers with dense convolution blocks further improved the segmentation performance. Experimental results showed that the proposed method compared favorably with existing methods.

Keywords

• Cone-beam computed tomography
• convolutional neural network
• network regularization
• posterior probability
• tooth segmentation