Dental Anatomy Segmentation from Cone Beam CT Images

Dental cone-beam computed tomography (CBCT) has been commonly used in digital dentistry, measuring complete dental anatomy information for diagnosis and treatment planning. Existing tooth segmentation from CBCT images has made adequate progress. However, there is no study proposed to make dental anatomical segmentation (i.e., enamel, pulp, and dentin), even if it is crucial in digital dentistry. Moreover, the limited CBCT resolution and large shape variance bring additional challenges. In this paper, we propose a novel learning-based method to automatically segment 3D tooth from CBCT images with structurally anatomical parts (i.e., enamel, pulp, and dentin). Furthermore, we utilize the tooth skeleton and Frangi filter to guide pulp segmentation precisely. Extensive experiments on our established dataset of 200 patients demonstrate the effectiveness and advantage of our method, which is helpful for clinical diagnosis and surgical treatment.