For pre-operative planning of interventions in dental implantology as well as maxillofacial surgery an accurate geometric reconstruction of the relevant anatomical structures is indispensable. An implant that is placed into the mandibular bone must not damage risk structures like the alveolar nerves or dental roots. Furthermore, a stable placement of the implant inside the bone must be achieved. Planning is performed based on medical images, such as CT or cone-beam CT. The aim is to segment the structures in an automatic fashion from the image data, which is impeded by low contrast or metal artefacts. We have developed a hierarchical approach based on statistical shape models and machine learning techniques in order to segment bone and nerve as well to determine the configuration of teeth. In particular, our bone segmentation has reached an accuracy suitable for clinical use.
