Tooth Segmentation from Cone Beam Computed Tomography using Skeleton-guided V-Net

Wenjing Lv, Qunwen Niu, Lanying Wang, Shuang Song, Jiahao Huo, Lin Wang, Ruoxiu Xiao

Abstract: Automatic and accurate tooth segmentation from cone beam computed tomography (CBCT) can provide important support for computer-assisted planning in dental implant surgery. Traditional methods for tooth segmentation often ignore the rich structural features of teeth, which may lead to misdiagnosis or missed diagnosis of underlying diseases. Preserving the accurate geometry and root details of the tooth remains a challenge. In this paper, we proposed a skeleton-guided V-Net network for the segmentation of teeth in CBCT images. In this network, we first use binary classification network to extract the region of interest (ROI) of teeth region. Next, a skeleton extraction network is designed to compute the skeletons of each tooth. Finally, a skeleton-guided V-Net is established to segment the teeth by accurately preserving the geometrical information. The segmentation performance of our proposed network is excellent, the average dice of tooth segmentation is 83.26%, and the precision is about 93.82%. Extensive evaluation, ablation, and comparison experiments demonstrate that our method exhibits state-of-the-art segmentation performance and accurate instance segmentation results, reflecting their potential applicability in clinical medicine.