Semantic Graph Attention with Explicit Anatomical Association Modeling for Tooth Segmentation from CBCT Images

Semantic Graph Attention with Explicit Anatomical Association Modeling for Tooth Segmentation from CBCT Images

Pengcheng Li^1,2, Yang Liu^3,4, Zhiming Cui⁵, Feng Yang^1,2,
Yue Zhao^1,2, Chunfeng Lian⁶, Chenqiang Gao^1,2

¹ School of Communication and Information Engineering, Chongqing University of Posts and Telecommunications,
² Chongqing Key Laboratory of Signal and Information Processing,
³ Department of Orthodontics, Stomatological Hospital of Chongqing Medical University,
⁴ Chongqing Key Laboratory for Oral Diseases and Biomedical Sciences,
⁵ School of Biomedical Engineering, ShanghaiTech University,
⁶ School of Mathematics and Statistics, Xi’an Jiaotong University

IEEE Transactions on Medical Imaging, 2022
[Paper], [bibtex]

Abstract

Accurate tooth identification and delineation in dental CBCT images are essential in clinical oral diagnosis and treatment. Teeth are positioned in the alveolar bone in a particular order, featuring similar appearances across adjacent and bilaterally symmetric teeth. However, existing tooth segmentation methods ignored such specific anatomical topology, which hampers the segmentation accuracy. Here we propose a semantic graph-based method to explicitly model the spatial associations between different anatomical targets (i.e., teeth) for their precise delineation in a coarse-to-fine fashion. First, to efficiently control the bilaterally symmetric confusion in segmentation, we employ a lightweight network to roughly separate teeth as four quadrants. Then, designing a semantic graph attention mechanism to explicitly model the anatomical topology of the teeth in each quadrant, based on which voxel-wise discriminative feature embeddings are learned for the accurate delineation of teeth boundaries. Extensive experiments on a clinical dental CBCT dataset demonstrate the superior performance of the proposed method compared with other state-of-the-art approaches.

Network Overview

Fig.1 The schematic diagram of the proposed method. The quadrant-segmentation network in the first stage is employed to divide all teeth into four quadrants. The SGANet in the second stage is utilized to identify and delineate each tooth.The final results fusion in the third stage merged the results of all the four quadrants based on the recorded indices (from the first stage) to reconstruct the segmentation in the original image space.

Result Gallery

Fig.2 Qualitative comparison with state-of-the-art methods. The first and second row shows the challenging canine (dislocation and oblique growth) segmentation results and corresponding 3D visualization results. The third and fourth row shows the segmentation results when missing the central incisor. In the competing methods, the missing tooth category is added to its adjacent teeth (i.e., one tooth is marked in 2 colors), and our method maintains accurate identification and segmentation results.

Table 1 The quantitative five-fold cross-validation segmentation results (mean $\pm$ standard deviation) comparison with both CNN and GCN-based methods.