A Semi-automatic Method And SoftWare For Tooth Segmentation In CT Images

Constructing accurate 3D representation of the teeth and surrounding bones plays an important role in the pre-operative planning and biomechanical evaluating of the dental surgery. In order to simulate tooth movement and rearrangement individually, it is necessary to segment each tooth from the alveolar bones before the 3D modeling.As current computer-aided systems in dentistry usually obtain the data needed for 3D modeling by digitizing the surface of the solid model (cast or imprints) of teeth and alveolar bones, many special algorithms have been developed according to the characteristics of the obtained data, and these automated or semi-automated algorithms can be used to define the contour of each tooth successfully. However, the constructed 3D surface model is not enough when the analysis requires a detailed model of the teeth and their roots, and this has obstructed the clinical application of the computer-aided systems in dentistry. For this reason, some efforts have been devoted in recent years to construct 3D model from CT slices, which can provide detailed information of the teeth and their roots.The dental CT images have the following distinct characteristics: The interstice between the two neighboring teeth is very small; the neighboring hard tissues have the same or similar gray values; the size, location and gray value of corresponding tooth in the two neighboring CT slices have no obvious difference. According to these characteristics, several special algorithms have been proposed for tooth segmentation in CT images, however, the shape and arrangement characteristics of teeth were not taken into consideration in these algorithms and the accuracy of the segmentation depend much on the setting of some key parameters.In this thesis, a new semi-automated algorithm is developed to segment each tooth from the alveolar bones. This algorithm is based on the characteristics of shape and arrangement of teeth, and the characteristics of CT images.Firstly, CT images are preprocessed with a home-designed program to get rid of noises; then a slice in which all teeth are visible is selected as the reference slice, and each tooth is segmented from the alveolar bones interactively in this slice; after that, the bounding box of each tooth is drawn and mapping to the neighboring slice as the operation region of corresponding tooth, and the segmentation operation is performed tooth by tooth in the operation region by using region growing algorithm or threshold algorithm; the finished slice will be taken as the new reference slice; in this way, tooth segmentation is carried out slice by slice until all slices are finished.This new algorithm proposed here makes use of the characteristics of shape and arrangement of teeth, and the characteristics of CT images. There are no significant differences between the semi-automated algorithms and the manual implementation for the segmentation results. This algorithm can be used to create patient-specific 3D model of teeth and alveolar bones in clinic.