Segmentation and 3D-modelling of single-rooted teeth from CBCT data: An automatic strategy based on dental pulp segmentation and surface deformation

Abstract

Individual three-dimensional (3D) models of the teeth obtained from Cone Beam Computed Tomography (CBCT) images are needed in orthodontics and maxillofacial surgery for treatment planning and simulation purposes. Such models can be obtained with the help of segmentation algorithms. In order to comply with clinical needs, the segmentation process should not rely on human intervention, while providing reliable patient-specific models. In this research, a fully automatic segmentation method based on surface deformation of in situ reconstructed models of dental pulps is proposed. A volume partitioning step defines separating planes between each tooth on both superior and inferior dental arches. A pulp segmentation strategy followed by a hierarchical surface deformation scheme, allows surface evolution until the tooth boundary is reached. Accuracy of the method is assessed by comparison of 26 single-rooted teeth randomly selected in 9 Cone Beam CT scans with ground truths obtained from manual segmentation. Experimental results show a Dice Similarity Coefficient of 92.19 ± 2.31%, a Jaccard Similarity Coefficient of 85.59 ± 3.89%, an Hausdorff distance of 1.54 ± 0.52 mm, and an average symmetric surface distance of 0.24 ± 0.05 mm.