Whole mandibular canal segmentation using transformed dental CBCT volume in Frenet frame

Abstract

Accurate segmentation of the mandibular canal is essential in dental implant and maxillofacial surgery, which can help prevent nerve or vascular damage inside the mandibular canal. Achieving this is challenging because of the low contrast in CBCT scans and the small scales of mandibular canal areas. Several innovative methods have been proposed for mandibular canal segmentation with positive performance. However, most of these methods segment the mandibular canal based on sliding patches, which may adversely affect the morphological integrity of the tubular structure. In this study, we propose whole mandibular canal segmentation using transformed dental CBCT volume in the Frenet frame. Considering the connectivity of the mandibular canal, we propose to transform the CBCT volume to obtain a sub-volume containing the whole mandibular canal based on the Frenet frame to ensure complete 3D structural information. Moreover, to further improve the performance of mandibular canal segmentation, we use clDice to guarantee the integrity of the mandibular canal structure and segment the mandibular canal. Experimental results on our CBCT dataset show that integrating the proposed transformed volume in the Frenet frame into other state-of-the-art methods achieves a $0.5\text{\%}\sim 12.1\text{\%}$ improvement in Dice performance. Our proposed method can achieve impressive results with a Dice value of 0.865 (±0.035), and a clDice value of 0.971 (±0.020), suggesting that our method can segment the mandibular canal with superior performance.

Keywords: CBCT; Deep learning; Frenet frame; Mandibular canal; Segmentation.

Figure 1

Mandibular canal segmentation in CBCT images. (a) represents the CBCT image in the axial view. The green curve indicates the ground truth of the mandibular canal and the red curve indicates the mandibular canal mask segmented using our method. (b) A 3D visualization of the mandibular canal (indicated in red) in the mandible (indicated in yellow) via the proposed method.

Figure 2

Workflow of the proposed segmentation of the mandibular canal on CBCT. * represents the coordinates of the mandibular foramen and mental foramen. The green arrow indicates the process of obtaining the mask of the mandibular mask and the coordinates of mandibular foramen and mental foramen. The blue arrow indicates the sub-volume mask conversion to the original image, and the orange box shows the prediction details.

Figure 3

(a) Diagram of the mandibular centerline under the Frenet frame; (b) Combined with the double reflection method; (c) Illustration of the point P. The gray areas indicate the surface perpendicular to the tangent vector at point P.

Figure 4

Illustration of the transformed volume.

Figure 5

Illustration of mandibular centerline extraction.

Figure 6

Quantitative evaluation of the segmentation results for independent test sets. ASSD and HD95 show the results of the left and right mandibular canals respectively.

Figure 7

An example of mandible and mandibular canal segmentation results.

Figure 8

Examples of 3D visualization with different segmentation methods.

Figure 9

Illustration of the segmentation results in the coronal view. Green represents the ground truth and red indicates the results predicted by different methods. Only green indicates that the mandibular canal is not segmented here.

Figure 10

Comparison of Dice values for different segmentation networks. (w) indicates the use of the transformed volume method, (w/o) indicates no use.

Figure 11

3D visualization of the segmentation results with or without the clDice module.

Figure 12

Examples of poor segmentation results. The green indicates the labeling of experts and the red represents the segmentation results. The last row is the position of the shown slice in 3D.