Tooth Instance Segmentation Algorithm Based on Multi-View Fusion in CBCT Images

Accurate and rapid segmentation of individual teeth from cone beam computed tomography (CBCT) images is of great significance in dental treatment. In order to solve the problems of difficult processing of occlusal surface and low resolution of CBCT image, a two-stage deep learning framework is proposed in this paper. The CBCT data are transformed into multi-view slices, and the axial segmentation results provide segmentation guidance and instance label information for sagittal segmentation by combining the data characteristics of different views. The CBCT data is reconstructed based on the sagittal plane case segmentation results, and the high precision tooth case segmentation results are obtained. The experimental results show that the average Dice coefficients of teeth and occlusal surfaces are 89.74% and 93.56%, respectively, and the average segmentation speed is 9.2 seconds, which realizes the task of high-precision instance segmentation of CBCT data in a short time.