Recognition and reconstruction of 3D tooth-alveolar bone complex in CBCT scans


Cone-beam computed tomography (CBCT) is commonly used nowadays for orthodontic diagnosis, treatment planning, and outcome assessment because of its low dose of radiation. For the evaluation of orthodontic treatment, CBCT scans are acquired at different timepoints. These CBCT scans are subsequently segmented into different regions of interest, such as mandible, maxilla and teeth. 3-dimensional (3D) surface models of mandible, maxilla, and teeth at the different timepoints are then created and superimposed to visually and quantitively demonstrate the orthodontic changes. Moreover, these 3D surface models allow for finite element analysis to study the stress distribution in the mandible, maxilla and teeth. However, it is also due to CBCT’s low radiation that the scans have noise and low contrast in visual impression. Currently, the most difficult step in the orthodontic evaluation is the image segmentation. There is no clear edge between teeth root and alveolar bone and their gray values are in the similar level. Therefore, it is not easy to segment teeth and alveolar bone. In CBCT scans, the adjacent teeth are mostly connective to each other which makes segmentation of individual tooth more difficult. To date no fully-automated algorithm has been developed that can reliably segment the different bony structures in CBCT scans. Therefore, a large amount of manual work is required to reconstruct 3D surface models of mandible, maxilla, and teeth. Consequently, the patient-specific orthodontic treatment is hampered by this time-consuming task. In order to reduce the work of image segmentation, we aimed to build the triangle mesh model to distinguish the morphology of teeth, develop a 3D point cloud registration method to recognize mandible and maxilla, and train a convolutional neural network (CNN) to segment mandible, maxilla, and teeth in CBCT scans.

The Institute of Information Technology, Zhejiang Shuren University has a software development center and a big data engineering research center. Prof. Fengjun Hu is an expert in computer graphics from this institute and his research interests also include intelligent robot and machine learning. Prof. Hu’s group recently started the project “the graphics study of bony structures in craniofacial region”. The Institute of Information Technology has the intent to start the new subject “medical computer science”. Our 3D Innovation Lab is now focus on the research project “CBCT image segmentation of mandible, maxilla and teeth for orthodontic treatment using artificial intelligence”. Based on these two research topics, both sides would like to communicate and share each other’s expertise in this field. We also like to discuss the current problems and challenges in image segmentation and talk about further collaboration. The main purpose of this seminar is to get to know each other by presenting researches, visiting labs and discussing questions and then promote the further cooperation.


Project number


Main applicant

T. Forouzanfar

Affiliated with

Amsterdam UMC - Locatie VUmc, Oral Diseases and dental surgery


21/10/2019 to 26/10/2019