基于卷积神经网络实现锥形束CT牙齿分割及牙位标定

doi:10.19723/j.issn.1671-167X.2024.04.030

摘要/Abstract

摘要：

目的: 利用卷积神经网络实现基于锥形束CT(cone-beam computed tomography, CBCT)体素数据的牙齿实例分割和牙位标定。方法: 本文所提出的牙齿算法包含三个不同的卷积神经网络，网络架构以Resnet为基础模块，首先对CBCT图像进行降采样，然后确定一个包含CBCT图像中所有牙齿的感兴趣区域(region of interest, ROI)。通过训练模型，ROI利用一个双分支“编码器-解码器”结构网络，预测输入数据中每个体素所对应的相关空间位置信息，进行聚类后实现牙齿的实例分割。牙位标定则通过另一个多类别分割任务设计的U-Net模型实现。随后，在原始空间分辨率下，训练了一个用于精细分割的U-Net网络，得到牙齿的高分辨率分割结果。本实验收集了59例带有简单冠修复体及种植体的CBCT数据进行人工标注作为数据库，对牙齿算法的预测结果使用实例Dice相似系数(instance Dice similarity coefficient, IDSC)用来评估牙齿分割结果，使用平均Dice相似系数(the average Dice similarity coefficient，ADSC)评估牙齿分割及牙位标定的共同结果并进行评定。结果: 量化指标显示，IDSC为89. 35%，ADSC为84. 74%。剔除了带有修复体伪影的数据后生成了有43例样本的数据库，训练网络得到了更优良的性能，IDSC为90. 34%，ADSC为87.88%。将得到的结果进行可视化分析，牙齿算法不仅可以清晰地分割出CBCT中牙齿的形态，而且可以对牙齿的分类进行准确的编号。结论: 该牙齿算法不仅可以成功实现三维图像的牙齿及修复体分割，还可以准确标定所有恒牙的牙位，具有临床实用性。

关键词: 卷积神经网络, 锥形束CT, 牙齿实例分割, 牙位标定

Abstract:

Objective: To propose a novel neural network to achieve tooth instance segmentation and recognition based on cone-beam computed tomography (CBCT) voxel data. Methods: The proposed methods included three different convolutional neural network models. The architecture was based on the Resnet module and built according to the structure of "Encoder-Decoder" and U-Net. The CBCT image was de-sampled and a fixed-size region of interest (ROI) containing all the teeth was determined. ROI would first through a two-branch "encoder and decoder" structure of the network, the network could predict each voxel in the input data of the spatial embedding. The post-processing algorithm would cluster the prediction results of the relevant spatial location information according to the two-branch network to realize the tooth instance segmentation. The tooth position identification was realized by another U-Net model based on the multi-classification segmentation task. According to the predicted results of the network, the post-processing algorithm would classify the tooth position according to the voting results of each tooth instance segmentation. At the original spatial resolution, a U-Net network model for the fine-tooth segmentation was trained using the region corresponding to each tooth as the input. According to the results of instance segmentation and tooth position identification, the model would process the correspon-ding positions on the high-resolution CBCT images to obtain the high-resolution tooth segmentation results. In this study, CBCT data of 59 cases with simple crown prostheses and implants were collected for manual labeling as the database, and statistical indicators were evaluated for the prediction results of the algorithm. To assess the performance of tooth segmentation and classification, instance Dice similarity coefficient (IDSC) and the average Dice similarity coefficient (ADSC) were calculated. Results: The experimental results showed that the IDSC was 89.35%, and the ADSC was 84. 74%. After eliminating the data with prostheses artifacts, the database of 43 samples was generated, and the performance of the training network was better, with 90.34% for IDSC and 87.88% for ADSC. The framework achieved excellent performance on tooth segmentation and identification. Voxels near intercuspation surfaces and fuzzy boundaries could be separated into correct instances by this framework. Conclusions: The results show that this method can not only successfully achieve 3D tooth instance segmentation but also identify all teeth notation numbers accurately, which has clinical practicability.

Key words: Convolutional neural network, Cone-beam computed tomography, Tooth instance segmentation, Tooth identification

中图分类号:

R780.1

薄士仕,高承志. 基于卷积神经网络实现锥形束CT牙齿分割及牙位标定[J]. 北京大学学报（医学版）, 2024, 56(4): 735-740.

Shishi BO,Chengzhi GAO. Tooth segmentation and identification on cone-beam computed tomography with convolutional neural network based on spatial embedding information[J]. Journal of Peking University (Health Sciences), 2024, 56(4): 735-740.

图/表 6

图1

图2

表1

图3

图4

图5

参考文献 15

1	Liao FZ , Liang M , Li Z , et al. Evaluate the malignancy of pulmonary nodules using the 3-D deep leaky noisy-OR network[J]. IEEE Trans Neural Netw Learn Syst, 2019, 30 (11): 3484- 3495. doi: 10.1109/TNNLS.2019.2892409
2	Long J , Shelhamer E , Darrell T . Fully convolutional networks for semantic segmentation[J]. IEEE Trans Pattern Anal Mach Intell, 2017, 39 (4): 640- 651. doi: 10.1109/TPAMI.2016.2572683
3	Ronneberger O, Fischer P, Brox T. U-Net: Convolutional networks for biomedical image segmentation: International conference on medical image computing and computer-assisted intervention[C]. Cham: Springer, 2015.
4	Neven D, Brabandere BD, Proesmans M, et al. Instance segmentation by jointly optimizing spatial embeddings and clustering bandwidth: 2019 IEEE/CVF conference on computer vision and pattern recognition (CVPR)[C]. Long Beach, CA: IEEE, 2019.
5	Pei YR , Ai XS , Zha HB , et al. 3D exemplar-based random walks for tooth segmentation from cone-beam computed tomography images[J]. Med Phys, 2016, 43 (9): 5040- 5050. doi: 10.1118/1.4960364
6	Patil S , Kulkarni V , Bhise A . Algorithmic analysis for dental caries detection using an adaptive neural network architecture[J]. Heliyon, 2019, 5 (5): e1579.
7	Zhang KL , Wu J , Chen H , et al. An effective teeth recognition method using label tree with cascade network structure[J]. Comput Med Imaging Graph, 2018, 68, 61- 70. doi: 10.1016/j.compmedimag.2018.07.001
8	Hosntalab M , Aghaeizadeh ZR , Abbaspour TA , et al. Classification and numbering of teeth in multi-slice CT images using wavelet-Fourier descriptor[J]. Int J Comput Assist Radiol Surg, 2010, 5 (3): 237- 249. doi: 10.1007/s11548-009-0389-8
9	Hwang JJ , Jung YH , Cho BH , et al. An overview of deep learning in the field of dentistry[J]. Imaging Sci Dent, 2019, 49 (1): 1- 7. doi: 10.5624/isd.2019.49.1.1
10	Chen H , Zhang KL , Lyu PJ , et al. A deep learning approach to automatic teeth detection and numbering based on object detection in dental periapical films[J]. Sci Rep, 2019, 9 (1): 3840. doi: 10.1038/s41598-019-40414-y
11	Miki Y , Muramatsu C , Hayashi T , et al. Classification of teeth in cone-beam CT using deep convolutional neural network[J]. Comput Biol Med, 2017, 80, 24- 29. doi: 10.1016/j.compbiomed.2016.11.003
12	Cui ZM, Li CJ, Wang WP. ToothNet: Automatic tooth instance segmentation and identification from cone beam CT images: 2019 IEEE/CVF conference on computer vision and pattern recognition (CVPR)[C]. Long Beach, CA: IEEE, 2019.
13	He KM, Zhang XY, Ren SQ, et al. Deep residual learning for image recognition: 2016 IEEE conference on computer vision and pattern recognition (CVPR)[C]. Las Vegas: IEEE, 2016.
14	Dentistry-designation system for teeth and areas of the oral cavity: ISO 3950: 2009[S/OL]. [2016-03-01]. https://www.iso.org/standard/68292.html.
15	Dice LR . Measures of the amount of ecologic association between species[J]. Ecology, 1945, 26 (3): 297- 302. doi: 10.2307/1932409

相关文章 13

[1]	章锦花,潘洁,孙志鹏,王霄. 不同根管内容物对口腔颌面锥形束CT诊断牙根纵裂准确性的影响[J]. 北京大学学报（医学版）, 2023, 55(2): 333-338.
[2]	潘孟乔,刘建,徐莉,徐筱,侯建霞,李小彤,王晓霞. 牙周-正畸-正颌联合治疗骨性安氏Ⅲ类错畸形患者下前牙牙周表型的长期观察[J]. 北京大学学报（医学版）, 2023, 55(1): 52-61.
[3]	叶佳学,梁宇红. 牙髓专科医师应用锥形束CT的现况调查[J]. 北京大学学报（医学版）, 2023, 55(1): 114-119.
[4]	高娟,吕航苗,马慧敏,赵一姣,李小彤. 锥形束CT三维体积测量评估骨性Ⅲ类错正畸正颌治疗后的牙根吸收[J]. 北京大学学报（医学版）, 2022, 54(4): 719-726.
[5]	刘伟涛,王怡然,王雪东,周彦恒. 锥形束CT研究上颌反复扩缩前方牵引后上颌骨缝的三维变化[J]. 北京大学学报（医学版）, 2022, 54(2): 346-355.
[6]	孟圆,张丽琪,赵雅宁,柳登高,张祖燕,高岩. 67例上颌根尖周囊肿的三维影像特点分析[J]. 北京大学学报（医学版）, 2021, 53(2): 396-401.
[7]	曹畅,王菲,王恩博,刘宇. β-磷酸三钙用于下颌第三磨牙拔除术后骨缺损修复的自身对照研究[J]. 北京大学学报（医学版）, 2020, 52(1): 97-102.
[8]	颜野,夏海缀,李旭升,何为,朱学华,张智荧,肖春雷,刘余庆,黄华,何良华,卢剑. 基于U型卷积神经网络学习的前列腺癌影像重建模型在手术导航中的应用[J]. 北京大学学报（医学版）, 2019, 51(3): 596-601.
[9]	谢晓艳,贾淑梅,孙志辉,张祖燕. 分辨率设置与锥形束CT检测牙根外吸收的可靠性[J]. 北京大学学报（医学版）, 2019, 51(1): 75-79.
[10]	赵一姣，刘怡，孙玉春，王勇. 一种基于曲率连续算法的冠、根三维数据融合方法[J]. 北京大学学报(医学版), 2017, 49(4): 719-723.
[11]	钟金晟, 欧阳翔英, 柳登高, 曹采方. 锥形束CT测量离体下颌磨牙Ⅱ°根分叉病变效果的评价[J]. 北京大学学报（医学版）, 2010, 42(1): 41-45.
[12]	刘怡, James MAH, 许天民. 锥形束计算机断层扫描中牙齿的分割精度[J]. 北京大学学报（医学版）, 2010, 42(1): 98-102.
[13]	王瑞永, 马绪臣, 张万林, 柳登高. 健康成年人颞下颌关节间隙锥形束计算机体层摄影术测量分析[J]. 北京大学学报（医学版）, 2007, 39(5): 503-506.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

Dataset	ADSC/%	IDSC/%
A	84.74±1.28	89.35±1.26
B	87.88±1.13	90.34±1.13