Tooth segmentation from cone-beam CT using graph cut

Abstract

Cone beam computed tomography (CBCT) can provide dentists with accurate 3D diagnostic images of the maxillofacial region at a lower irradiation dose compare to conventional medical CT. Due to low image contrast, higher image noise and missing image boundaries, tooth segmentation in CBCT is difficult even with experienced radiographic interpreters. In this paper, we proposed a graph cuts segmentation approach of obtaining the 3D tooth model from CBCT images. A 3D Markov Random Fields (MRF) is used to model CBCT 3D images. We then used graph cuts to obtain the optimal image segmentation. For a total of 25 teeth data sets, our results shows an average dice similarity coefficient of 0.89.