基于可微拓扑搜索与可信动量优化的半监督3D牙齿分割方法
Semi-Supervised 3D Tooth Segmentation via Differentiable Topology Search and Trusted Momentum Optimization
摘要: 本文提出DiNTS-TMO，一种结合可微拓扑搜索与可信动量优化的半监督3D牙齿分割框架。针对现有NAS方法在半监督场景下易受伪标签噪声干扰的问题，DiNTS-TMO采用DiNTS可微拓扑搜索自动发现网络结构，并引入可信动量优化(TMO)机制，通过对齐门控筛选无标签梯度，在搜索阶段同时约束网络权重与架构参数的更新。基于STS-Tooth数据集两个子集的五折交叉验证结果，DiNTS-TMO在ROI子集上取得了最优的DSC和HD95，在Integrity子集上取得了最优的DSC，但在边界距离指标上仍存在改进空间。实验结果表明，该方法在低标注3D牙齿分割任务中具有较强的竞争力，并为结合NAS与半监督学习提供了一种可行思路。
Abstract: We propose DiNTS-TMO, a semi-supervised 3D tooth segmentation framework that combines differentiable topology search with Trusted Momentum Optimization (TMO). To address the sensitivity of NAS to noisy pseudo labels, DiNTS-TMO uses DiNTS to automatically discover network topologies and introduces an alignment-gated TMO mechanism to filter unlabeled gradients, jointly constraining the updates of network weights and architecture parameters during search. Five-fold cross-validation on two STS-Tooth subsets shows that DiNTS-TMO achieves the best DSC and HD95 on the ROI subset and the best DSC on the Integrity subset, while still leaving room for improvement on boundary-distance metrics in the full-scan setting. These results indicate that DiNTS-TMO is a competitive approach for low-label 3D tooth segmentation and provide a practical route to combining NAS with semi-supervised learning.
文章引用：林泽峰. 基于可微拓扑搜索与可信动量优化的半监督3D牙齿分割方法[J]. 计算机科学与应用, 2026, 16(4): 498-510. https://doi.org/10.12677/csa.2026.164148

参考文献

[1] Ronneberger, O., Fischer, P. and Brox, T. (2015) U-Net: Convolutional Networks for Biomedical Image Segmentation. In: Lecture Notes in Computer Science, Springer, 234-241. [Google Scholar] [CrossRef] 
[2] Çiçek, Ö., Abdulkadir, A., Lienkamp, S.S., Brox, T. and Ronneberger, O. (2016) 3D U-Net: Learning Dense Volumetric Segmentation from Sparse Annotation. In: Lecture Notes in Computer Science, Springer, 424-432. [Google Scholar] [CrossRef] 
[3] Milletari, F., Navab, N. and Ahmadi, S. (2016) V-Net: Fully Convolutional Neural Networks for Volumetric Medical Image Segmentation. 2016 Fourth International Conference on 3D Vision (3DV), Stanford, 25-28 October 2016, 565-571. [Google Scholar] [CrossRef] 
[4] Isensee, F., Jaeger, P.F., Kohl, S.A.A., Petersen, J. and Maier-Hein, K.H. (2021) NNU-Net: A Self-Configuring Method for Deep Learning-Based Biomedical Image Segmentation. Nature Methods, 18, 203-211. [Google Scholar] [CrossRef] [PubMed]
[5] Liu, H., Simonyan, K. and Yang, Y. (2019) DARTS: Differentiable Architecture Search. Proceedings of the International Conference on Learning Representations, New Orleans, 6-9 May 2019, 1-13.
[6] He, Y., Yang, D., Roth, H., Zhao, C. and Xu, D. (2021) DiNTS: Differentiable Neural Network Topology Search for 3D Medical Image Segmentation. 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Nashville, 20-25 June 2021, 5837-5846. [Google Scholar] [CrossRef] 
[7] Laine, S. and Aila, T. (2017) Temporal Ensembling for Semi-Supervised Learning. Proceedings of the International Conference on Learning Representations, Toulon, 24-26 April 2017, 1-13.
[8] Tarvainen, A. and Valpola, H. (2017) Mean Teachers are Better Role Models: Weight-Averaged Consistency Targets Improve Semi-Supervised Deep Learning. Advances in Neural Information Processing Systems, 30, 1195-1204.
[9] Yu, L., Wang, S., Li, X., Fu, C.W. and Heng, P.A. (2019) Uncertainty-Aware Self-Ensembling Model for Semi-Supervised 3D Left Atrium Segmentation. In: Lecture Notes in Computer Science, Springer, 605-613. [Google Scholar] [CrossRef] 
[10] Chen, X., Yuan, Y., Zeng, G. and Wang, J. (2021) Semi-Supervised Semantic Segmentation with Cross Pseudo Supervision. 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Nashville, 20-25 June 2021, 2613-2622. [Google Scholar] [CrossRef] 
[11] Zhao, W., Zhong, L., Liao, X., Liao, W., Zhang, S., Zhang, S., et al. (2026) MetaSSL: A General Heterogeneous Loss for Semi-Supervised Medical Image Segmentation. IEEE Transactions on Medical Imaging, 45, 751-763. [Google Scholar] [CrossRef] 
[12] Huang, H., Chen, Z., Chen, C., Lu, M. and Zou, Y. (2023) Complementary Consistency Semi-Supervised Learning for 3D Left Atrial Image Segmentation. Computers in Biology and Medicine, 165, Article 107368. [Google Scholar] [CrossRef] [PubMed]
[13] Liang, K., Chen, L., Liu, B. and Liu, Q. (2026) Cautious Optimizers: Improving Training with One Line of Code. Proceedings of the International Conference on Learning Representations, Rio de Janeiro, 23-27 April 2026, 1-26.
[14] Pauletto, L., Amini, M. and Winckler, N. (2022) Se2NAS: Self-Semi-Supervised Architecture Optimization for Semantic Segmentation. 2022 26th International Conference on Pattern Recognition (ICPR), Montreal, 21-25 August 2022, 54-60. [Google Scholar] [CrossRef] 
[15] Chen, R., Nian, F., Cen, Y., Peng, Y., Wang, H., Yu, Z., et al. (2025) L-SSHNN: A Larger Search Space of Semi-Supervised Hybrid NAS Network for Echocardiography Segmentation. Expert Systems with Applications, 276, Article 127084. [Google Scholar] [CrossRef] 
[16] Yu, T., Kumar, S., Gupta, A., Levine, S., Hausman, K. and Finn, C. (2020) Gradient Surgery for Multi-Task Learning. Advances in Neural Information Processing Systems, 33, 5824-5836.
[17] Qiu, Z., Yao, T. and Mei, T. (2017). Learning Spatio-Temporal Representation with Pseudo-3D Residual Networks. 2017 IEEE International Conference on Computer Vision (ICCV), Venice, 22-29 October 2017, 5534-5542.[CrossRef] 
[18] Liu, C., Chen, L., Schroff, F., Adam, H., Hua, W., Yuille, A.L., et al. (2019) Auto-DeepLab: Hierarchical Neural Architecture Search for Semantic Image Segmentation. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Long Beach, 15-20 June 2019, 82-92. [Google Scholar] [CrossRef] 
[19] Wang, Y., Ye, F., Chen, Y., Wang, C., Wu, C., Xu, F., et al. (2025) A Multi-Modal Dental Dataset for Semi-Supervised Deep Learning Image Segmentation. Scientific Data, 12, Article No. 117. [Google Scholar] [CrossRef] [PubMed]
[20] Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., et al. (2019) PyTorch: An Imperative Style, High-Performance Deep Learning Library. Advances in Neural Information Processing Systems, 32, 8024-8035.
[21] MONAI Contributors (2026) Modules—MONAI 1.5.0 Documentation.
https://docs.monai.io/en/1.5.0/modules.html
[22] Oktay, O., Schlemper, J., Folgoc, L.L., Lee, M., Heinrich, M., et al. (2018) Attention U-Net: Learning Where to Look for the Pancreas. Proceedings of the International Conference on Medical Imaging with Deep Learning, Amsterdam, 4-6 July 2018, 1-10.
[23] Hatamizadeh, A., Tang, Y., Nath, V., Yang, D., Myronenko, A., Landman, B., et al. (2022) UNETR: Transformers for 3D Medical Image Segmentation. 2022 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), Waikoloa, 3-8 January 2022, 1748-1758. [Google Scholar] [CrossRef] 
[24] Tang, Y., Yang, D., Li, W., Roth, H.R., Landman, B., Xu, D., et al. (2022) Self-Supervised Pre-Training of Swin Transformers for 3D Medical Image Analysis. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), New Orleans, 18-24 June 2022, 20698-20708. [Google Scholar] [CrossRef] 
[25] Myronenko, A. (2019) 3D MRI Brain Tumor Segmentation Using Autoencoder Regularization. In: Crimi, A., Bakas, S., Kuijf, H., Keyvan, F., Reyes, M. and van Walsum, T., Lecture Notes in Computer Science, Springer, 311-320. [Google Scholar] [CrossRef]