CTooth: A Fully Annotated 3D Dataset and Benchmark for Tooth Volume Segmentation on Cone Beam Computed Tomography Images

3D tooth segmentation is a prerequisite for computer-aided dental diagnosis and treatment. However, segmenting all regions manually subjective time-consuming. Recently, deep learning-based methods produce convincing results reduce manual annotation efforts, but it requires large quantity of ground truth training. To our knowledge, there are few data available the study. In this paper, we establish fully annotated cone beam computed tomography dataset CTooth with gold standard. This contains 22 volumes (7363 slices) fine labels by experienced radiographic interpreters. ensure relative even sampling distribution, variance included in including missing teeth restoration. Several state-of-the-art evaluated on dataset. Afterwards, further summarise apply series attention-based Unet variants volumes. work provides new benchmark volume task. Experimental evidence proves that attention modules UNet structure boost responses areas inhibit influence background noise. The best performance achieved SKNet module, 88.04% Dice 78.71% IOU, respectively. framework outperforms other codebase released here .