基于网格特征的自动排牙方法

doi:10.3969/j.issn.1004-132X.2025.11.032

摘要/Abstract

摘要：

针对基于点云的深度学习自动排牙方法数据依赖性强、咬合准确度低等问题，提出了一种基于网格特征的深度学习自动排牙方法。设计的模型包括形状编码器、全局特征编码器、特征解码与映射器以及牙齿咬合生成网络。形状编码器从牙齿模型表面的三角网格数据中提取牙齿形状特征，全局特征编码器从简化后的牙齿点云中提取牙列全局特征，特征解码与映射器则对牙齿全局特征、牙齿局部特征进行融合降维，生成最终的排牙结果，减少了数据依赖性，牙齿咬合生成网络基于颌骨空间位置关系和牙齿特征生成上下牙咬合面，提高了上下牙咬合准确性。为进一步提高模型性能，在损失函数中引入了相似性损失函数，有助于防止过拟合，提高了自动排牙的质量。实验结果表明，与四种现有方法相比，该方法在ADD指标上均有降低，显著提高了深度学习自动排牙的准确度。

关键词: 自动排牙, 网格特征, 形状编码器, 牙齿咬合生成网络

Abstract:

Aiming at the problems of strong data dependence and low occlusion accuracy of point cloud-based deep learning automatic tooth alignment method， a deep learning automatic tooth alignment method was proposed based on mesh features. The design model included shape encoder， global feature encoder， feature decoder and mapper， and tooth occlusion generation network. The shape encoder extracted the tooth shape features from the triangular mesh data on the surfaces of the tooth model， the global feature encoder extracted the global features of the tooth set from the simplified tooth point clouds， and the feature decoder and mapper fused and reduced the dimension of the global features and local features of the tooth to generate the final tooth arrangement results， reducing the data dependence. The tooth occlusion network generated the upper and lower occlusal surfaces based on the spatial position of the jaw and the characteristics of the teeth， which improved the accuracy of the upper and lower occlusal surfaces. In order to further improve the performance of the model， the similarity loss function was introduced into the loss function， which helped to prevent overfitting and improve the quality of automatic tooth alignment. The experimental results show that compared with four existing methods， the proposed method reduces the ADD index， and significantly improves the accuracy of deep learning automatic tooth alignment.

Key words: automatic tooth alignment, mesh feature, shape encoder, tooth occlusion generating network

中图分类号:

TP391

胡梦杰, 方宇航, 秦绪佳, 吴正强. 基于网格特征的自动排牙方法[J]. 中国机械工程, 2025, 36(11): 2738-2746.

Mengjie HU, Yuhang FANG, Xujia QIN, Zhengqiang WU. Automatic Tooth Alignment Method Based on Mesh Features[J]. China Mechanical Engineering, 2025, 36(11): 2738-2746.

导出引用管理器 EndNote|Ris|BibTeX

链接本文: https://www.cmemo.org.cn/CN/10.3969/j.issn.1004-132X.2025.11.032

https://www.cmemo.org.cn/CN/Y2025/V36/I11/2738

图/表 11

图1 Mbtanet模型结构

Fig.1 Structure of Mbtanet model

图2 形状编码器

Fig.2 Shape encoder

图3 牙齿咬合生成网络

Fig.3 Tooth occlusion generating network

图4 增强模型输出结果

Fig.4 Results of enhance model

图5 不同方法下精度对比图

Fig.5 Comparison of precision under different methods

表1 不同方法的性能比较

Tab.1 Performance comparison of different methods

方法	ADD	PA-ADD	CSA	MErot	FDcur	Mean TRE
PSTN	1.793	1.447	0.785	6.103	2.565	0.3426
TAligNet	1.761	1.425	0.789	5.922	2.678	0.3354
TANet	1.597	1.333	0.823	5.585	2.321	0.3201
TADPM	1.582	1.316	0.859	5.581	2.229	0.3125
Mbtanet（本文）	1.562	1.299	0.878	5.585	2.207	0.3001

图6 牙列结果对比图

Fig.6 Comparison of dental alignment results

图7 不同咬合问题的输出结果

Fig.7 Results of different occlusion problems

表2 消融实验结果

Tab.2 Results of ablation experiment

损失	ADD	PA-ADD	CSA	MErot	FDcur	Mean TRE
$L C D$	1.648	1.456	0.849	5.658	2.378	0.3194
$L p o s$	1.635	1.359	0.834	5.669	2.285	0.3207
$L d o f$	1.728	1.556	0.836	5.681	2.364	0.3266
$L C D$ + $L p o s$ + $L d o f$	1.571	1.347	0.852	5.640	2.247	0.3139
$L$	1.562	1.299	0.878	5.585	2.207	0.3001

表2 消融实验结果

Tab.2 Results of ablation experiment

损失	ADD	PA-ADD	CSA	MErot	FDcur	Mean TRE
$L C D$	1.648	1.456	0.849	5.658	2.378	0.3194
$L p o s$	1.635	1.359	0.834	5.669	2.285	0.3207
$L d o f$	1.728	1.556	0.836	5.681	2.364	0.3266
$L C D$ + $L p o s$ + $L d o f$	1.571	1.347	0.852	5.640	2.247	0.3139
$L$	1.562	1.299	0.878	5.585	2.207	0.3001

图8 排牙实例测试结果

Fig.8 Test result of dental alignment example

图9 排牙测试准确率

Fig.9 Accuracy of tooth arrangement testing

参考文献 27

[1]	DOU Y L， SHEN D G， CUI Z M， et al. 3D Structure Guided Network for Tooth Alignment in 2D Photograph［J］. arXiv： 2310. 11106， 2023.
[2]	姜金刚，陈厚鋆，马雪峰，等.个性化正畸弓丝成形规划方法及实验研究［J］. 中国机械工程， 2020， 31（11）： 1323-1330.
	JIANG Jingang， CHEN Houjun， MA Xuefeng， et al. Forming Planning Method and Experimentation of Personalized Orthodontic Archwires［J］. China Mechanical Engineering， 2020， 31（11）： 1323-1330.
[3]	DAI N， YU X， FAN Q， et al. Complete Denture Tooth Arrangement Technology Driven by a Reconfigurable Rule［J］. PLoS ONE， 2018， 13（6）： 452-457.
[4]	WANG C， WEI G， WANG W， et al. Tooth Alignment Network Based on Landmark Constraintsand Hierarchicalgraph Structure［C］∥IEEE Transactions on Visualization and Computer Graphics. Melbour-ne， 2022： 1457-1469.
[5]	AUBRY M， MATURANA D， EFROS A， et al. Seeing 3D Chairs： Exemplar Part-based 2D-3D Alignment Using a Large Dataset of CAD Models［C］∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition（CVPR）.Los Alamitos， 2014： 3762-3769.
[6]	FISHER M， RITCHIE D， SAVVA M， et al. Example-based Synthesis of 3D Object Arrangements［J］. ACM Transactions on Graphics（TOG）， 2012， 31（6）： 135.
[7]	程筱胜，郝国栋，戴宁，等. 三角网格牙颌模型上的牙弓线半自动探测方法［J］. 机械科学与技术， 2008， 27（8）： 1043-1046.
	CHENG Xiaosheng， HAO Guodong， DAI Ning， et al. Semi-automatic Detection Method for Dental Arch Line on Triangular Mesh Dental Model［J］. Mechanical Science and Technology for Aerospace Engineering， 2008， 27（8）： 1043-1046.
[8]	张永德，蒋济雄. 正畸弓丝弯制特性分析及实验研究［J］. 中国机械工程， 2011， 22（15）： 1827-1831.
	ZHANG Yongde， JIANG Jixiong. Bending Property Analysis and Experimental Study of Orthodontic Wires［J］. China Mechanical Engineering， 2011， 22（15）： 1827-1831.
[9]	LIN A C. Integration of 3D CAD， Reverse Engineering and Rapid Prototyping in Fabrication of Invisible Tooth Aligner［C］∥Proceedings of the IEEE International Conference on Systems， Man， and Cybernetics （SMC）. Los Alamitos， 2005： 2431-2436.
[10]	WANG X Z， LI Z K， MA Y Q. A PSO Based Automated Tooth Arrangement Method［J］. Computer Engineeringand Applications， 2012， 48（5）：211-212.
[11]	CHENG C， CHENG X， DAI N， et al. Personalized Orthodontic Accurate Tooth Arrangement System with Complete Teeth Model［J］. Journal of Medical Systems， 2015， 39（9）： 1-12.
[12]	WEI G， CUI Z， ZHU J， et al. Dense Representative Toothlandmark/Axis Detection Network on 3D Model［J］. Computer-Aided Geometric Design， 2022， 94（1）： 10277.
[13]	CHANG Y B， XIA J J， GATENO J， et al. An Automatic and Robust Algorithm of Reestablishment of Digital Dental Occlusion［J］. IEEE Transactions on Medical Imaging， 2010， 29（9）： 1652-1663.
[14]	CUI Z M， LI C J， WANG W P， et al. ToothNet： Automatic Tooth Instance Segmentation and Identification from Cone Beam CT Images［C］∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition（CVPR）. Los Alamitos， 2019： 6368-6377.
[15]	CUI Z M， LI C J， CHEN N L， et al. TSegNet： an Efficient and Accurate Tooth Segmentation Network on 3D Dental Model［J］. Medical Image Analysis， 2021， 69（1）： 101949.
[16]	WU W， QI Z， LI F P. PointConv： Deep Convolutional Networks on 3D Point Clouds［C］∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition （CVPR）. Los Alamitos， 2019： 9621-9630.
[17]	LIU Y， FAN B， XIANG S， et al. Relation Shape Convolutional Neural Network for Point Cloud Analysis［C］∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition （CVPR）. Los Alamitos， 2019： 8895-8904.
[18]	WANG Y， SUN Y， LIU Z， et al. Dynamic Graph CNN for Learning on Point Clouds［J］. ACM Transactions on Graphics， 2019， 38（5）： 1-12.
[19]	WEI G， CUI Z， LIU Y， et al. TANet： Towards Fully Automatic Tooth Arrangement［C］∥Proceedings of the European Conference on Computer Vision （ECCV）. Glasgow， 2020： 481-497.
[20]	QI C R， SU H， MO K， et al. PointNet： Deep Learning on Point Sets for 3D Classification and Segmentation［C］∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition （CVPR）. Los Alamitos， 2017： 652-660.
[21]	LI Y， TARLOW D， BROCKSCHMIDT M， et al. Gated Graph Sequence Neural Networks［J］.arXiv：， 2015.
[22]	LING C， SHI Z F， WANG Y Q， et al. iOrthoPredictor： Model Guided Deep Prediction of Teeth Alignment［J］. ACM Transactions on Graphics， 2020， 39（6）： 216.
[23]	LI X， BI L， KIM J， et al. Malocclusion Treatment Planning via PointNet-based Spatial Transformation Network［C］∥Proceedings of the International Conference on Medical Image Computing and Computer-assisted Intervention （MICCAI）. Cham， 2020： 105-114.
[24]	RUIZHONGTAI Q C， LI Y， WU J， et al. PointNet++： Deep Hierarchical Feature Learning on Point Sets in a Metric Space［C］∥Advances in Neural Information Processing Systems（NeurIPS）. Vancouver， 2017： 5105-5114.
[25]	LEI C S， XIA M F， WANG S F， et al. Automatic Tooth Arrangement with Joint Features of Point and Mesh Representations via Diffusion Probabilistic Models［J］.arXiv：， 2024.
[26]	田素坤，戴宁，袁福来. 基于条件生成对抗网络的缺失牙体功能性咬合面形态设计方法［J］.中国机械工程， 2021， 32（3）： 331-340.
	TIAN Sukun， DAI Ning， YUAN Fulai. Functional Occlusal Surface Morphology Design Method for Missing Teeth Based on CGAN［J］. China Mechanical Engineering， 2021， 32（3）： 331-340.
[27]	LUO Y， DU H， YAN Y， et al. Mesh CNN-based BREP to CSG Conversion Algorithm for 3D CAD Models and Its Application［J］. Nuclear Science and Techniques， 2022， 33（6）： 77-90.