DentalSegmentator: robust deep learning-based CBCT image segmentation

Delineation of anatomical structures on dento-maxillo-facial (DMF) computed tomography (CT) or cone beam computed tomography (CBCT) scans is greatly needed for an increasing number of digital dentistry tasks. Following this process, called segmentation, three-dimensional (3D) patient-specific models can be exported for visualization, treatment planning, intervention, and follow-up purposes. Although several methods based on deep learning (DL) have been proposed for automating this task, there is no thoroughly evaluated publicly available tool offering segmentation of the anatomical structures needed for digital dentistry workflows. In this work, we propose and evaluate DentalSegmentator, a tool based on the nnU-Net deep learning framework, for fully automatic segmentation of 5 anatomic structures on DMF CT and CBCT scans: maxilla and upper skull, mandible, upper teeth, lower teeth and mandibular canal. A retrospective sample of 470 CT and CBCT scans was used as a training/validation set. The performance and generalizability of the tool was evaluated by comparing segmentations provided by experts and automatic segmentations on 2 hold-out test datasets: an internal dataset of 133 CT and CBCT scans acquired before orthognathic surgery and an external dataset of 123 CBCT scans randomly sampled from routine examinations in 5 institutions. In our internal test dataset (n = 133), the mean overall results were a Dice similarity coefficient (DSC) of 92.2 ± 6.3% and a normalized surface distance (NSD) of 98.2 ± 2.2%. In our external test dataset (n = 123), the mean overall results were a DSC of 94.2 ± 7.4% and a NSD of 98.4 ± 3.6%. The results obtained on this highly diversified dataset demonstrate that our tool can provide fully automatic and robust multiclass segmentation for DMF (CB)CT scans. To encourage the clinical deployment of DentalSegmentator, our pretrained nnU-Net model is made publicly available along with an extension for the 3D Slicer software.