An efficient method to automate tooth identification and 3D bounding box extraction from Cone Beam CT Images

TL;DR

This work tackles automatic tooth identification, localization, and 3D bounding box extraction from CBCT images by splitting the volume into axial slices and applying a single-stage 2D detector (YOLOv7) to label teeth within each slice. A Hungarian assignment-based method links 2D detections across slices to build coherent 3D tooth volumes, supplemented by a graph-inspired reconstruction and an artifact-correction step that can separate fused teeth when interocclusal space is limited. The approach achieves high detection performance (mAP@0.5 = 0.889) and robust reconstruction across two dataset subsets, with strong practical integration into the Dentomo dental analysis tool. Overall, the method offers a resource-efficient, interpretable pipeline that reduces labeling effort and supports downstream dental pathology analysis in clinical settings.

Abstract

Accurate identification, localization, and segregation of teeth from Cone Beam Computed Tomography (CBCT) images are essential for analyzing dental pathologies. Modeling an individual tooth can be challenging and intricate to accomplish, especially when fillings and other restorations introduce artifacts. This paper proposes a method for automatically detecting, identifying, and extracting teeth from CBCT images. Our approach involves dividing the three-dimensional images into axial slices for image detection. Teeth are pinpointed and labeled using a single-stage object detector. Subsequently, bounding boxes are delineated and identified to create three-dimensional representations of each tooth. The proposed solution has been successfully integrated into the dental analysis tool Dentomo.

Figures (6)

• Figure 1: Tooth division and reconstruction dataset. Distribution of the labels in FDI notation (in increasing sub-label order). There are 89 CBCT images, 56 with bite block (left) and 33 without bite block (right).
• Figure 2: Tooth detection in an axial slice. No distinction between top and bottom, or right and left sides.
• Figure 3: Matches in consecutive axial slices. In red, the tooth volumes under construction with a single match. In yellow, the tooth volumes under construction with no matches in that axial slice. In other colors, the "active" tooth volumes (filled) matched with a new detection (not filled) in the axial slice under study.
• Figure 4: Matching two-dimensional bounding boxes to model a three-dimensional volume containing a tooth.
• Figure 5: Teeth reconstruction. Sagittal (a and b) and coronal (c) views of the teeth.