A Hybrid Approach to Extract Tooth Models from CT Volumes

摘要

As the tooth root has similar bone density to the jaw where it is embedded, its complete boundaries are either missing or at low contrast in the computed tomography (CT) volume data. This paper proposes a hybrid method to create a 'best-fit' polygonal surface of the patient-specific tooth. First, a level-set based shape prior segmentation procedure is employed to extract a coarse whole tooth surface model from CT volume. The surface model produced captures the smooth root part, while losing details of the tooth crown. So, a post process - thin-plate splines transform, involving a consistent semi-automatic landmarks selection and re-placing procedure - is used to warp the crown part of the coarse surface to recover the patient-specific local details of the crown.