Application of Mean Teacher Method in Semi-supervised Medical Image Segmentation

doi:10.12068/j.issn.1005-3026.2026.20250103

Abstract

Abstract:

Medical image segmentation is an important technical basis for clinical diagnosis and treatment. Accurate medical image segmentation contributes to improving the accuracy and efficiency of disease diagnosis. Deep learning methods have made significant progress in this field. However， these methods are heavily dependent on manually labeled data， and the high cost of obtaining high-quality segmentation labels limits their practical application. Semi-supervised learning effectively alleviates the label scarcity problem by combining a small amount of labeled data with a large amount of unlabeled data. Mean teacher （MT） is a mainstream semi-supervised learning method. It leverages an exponential moving average to extract information from unlabeled data， enhancing model accuracy and generalization performance. Currently， MT has been widely adopted in medical image segmentation. In this paper， MT was comprehensively reviewed， focusing on its application and improvement in the medical image segmentation field from aspects such as consistency regularization， uncertainty， attention mechanism， multi-task learning， auxiliary correction， and model variants. Furthermore， trends in the application and enhancement of MT were briefly analyzed， and commonly used comparative experimental methods， datasets， backbone networks of MT， and evaluation metrics in medical image segmentation were summarized. Finally， existing challenges and potential future research directions in applying MT to medical image segmentation were discussed.

Key words: mean teacher, medical image segmentation, deep learning, semi-supervised learning, consistency regularization

CLC Number:

TP 751

Jin-zhu YANG, Mei WEI, Qi YU, Song SUN. Application of Mean Teacher Method in Semi-supervised Medical Image Segmentation[J]. Journal of Northeastern University(Natural Science), 2026, 47(1): 20-30.

Figures/Tables 7

Fig.1 Annual publication trend of papers related to MT improvements from 2018 to February 2025

Fig.2 Flowchart of MT method

Table 1 Commonly used comparative experimental methods used in medical image segmentation papers based

方法名称	主要内容及优势
UA-MT^[24]	通过不确定性感知机制提高分割性能的3D磁共振左心房分割方法
DTC^[50]	基于双任务一致性的分割框架，引入任务级正则化和几何约束，通过联合预测像素级分割图和基于几何的水平集表示来降低计算成本
SASSNet^[51]	基于形状感知的3D语义分割方法，设计同时预测语义分割图和符号距离图的多任务深度网络，充分利用像素级信息，结合几何形状约束，提升网络捕捉复杂形状的能力
URPC ^[52]	用于鼻咽癌实体肿瘤靶区的分割框架，通过多尺度金字塔预测网络和不确定性校正模块，充分利用无标签数据中的丰富信息，降低对标签数据的需求
CPS^[53]	交叉伪标签监督的语义分割方法，在2个具有相同结构但不同初始化的分割网络之间施加一致性约束，利用伪标签相互监督，提高模型对无标签数据的利用效率
ICT^[7]	插值一致性算法，通过在无标签数据的插值上施加一致性约束，推动决策边界向低密度区域移动，减少过拟合并提高泛化能力
MC-Net ^[54]	相互一致性左心房分割框架，通过强调无标签数据中的小分支或模糊边缘等挑战性区域，结合不确定性估计和循环伪标签方案实现相互一致性训练
CCT^[55]	交叉一致性训练语义分割方法.通过在编码器的输出上施加不同扰动，并强制主解码器与辅助解码器之间的预测一致性，增强模型对无标签数据的学习能力
DAN^[56]	设计分割网络和评估网络形成对抗训练机制，设置超参数优化模型权重，加快收敛速度
SS-Net^[57]	结合像素级平滑和类别间分离，提升模型对抗扰动输出一致性和模糊区域特征提取能力
BCP^[58]	通过双向复制粘贴操作混合标签和无标签图像来增强模型对无标签数据的分割能力

Table 1 Commonly used comparative experimental methods used in medical image segmentation papers based

方法名称	主要内容及优势
UA-MT^[24]	通过不确定性感知机制提高分割性能的3D磁共振左心房分割方法
DTC^[50]	基于双任务一致性的分割框架，引入任务级正则化和几何约束，通过联合预测像素级分割图和基于几何的水平集表示来降低计算成本
SASSNet^[51]	基于形状感知的3D语义分割方法，设计同时预测语义分割图和符号距离图的多任务深度网络，充分利用像素级信息，结合几何形状约束，提升网络捕捉复杂形状的能力
URPC ^[52]	用于鼻咽癌实体肿瘤靶区的分割框架，通过多尺度金字塔预测网络和不确定性校正模块，充分利用无标签数据中的丰富信息，降低对标签数据的需求
CPS^[53]	交叉伪标签监督的语义分割方法，在2个具有相同结构但不同初始化的分割网络之间施加一致性约束，利用伪标签相互监督，提高模型对无标签数据的利用效率
ICT^[7]	插值一致性算法，通过在无标签数据的插值上施加一致性约束，推动决策边界向低密度区域移动，减少过拟合并提高泛化能力
MC-Net ^[54]	相互一致性左心房分割框架，通过强调无标签数据中的小分支或模糊边缘等挑战性区域，结合不确定性估计和循环伪标签方案实现相互一致性训练
CCT^[55]	交叉一致性训练语义分割方法.通过在编码器的输出上施加不同扰动，并强制主解码器与辅助解码器之间的预测一致性，增强模型对无标签数据的学习能力
DAN^[56]	设计分割网络和评估网络形成对抗训练机制，设置超参数优化模型权重，加快收敛速度
SS-Net^[57]	结合像素级平滑和类别间分离，提升模型对抗扰动输出一致性和模糊区域特征提取能力
BCP^[58]	通过双向复制粘贴操作混合标签和无标签图像来增强模型对无标签数据的分割能力

Fig.3 Common comparison methods and their frequencies

Table 2 Performance evaluation of common semi-supervised methods on 2018 Left Atrium Segmentation

方法名称	评价指标				标签数据数量
方法名称	Dice/%	Jaccard/%	ASD	95HD	labeled	unlabeled
MT^[5]	88.23	79.29	2.73	10.64	16	64
UA-MT^[24]	88.88	80.21	2.26	7.32	16	64
UA-MT^[24]	87.79	78.39	2.12	8.68	8	72
DTC^[50]	89.42	80.98	2.10	7.32	16	64
DTC^[50]	87.51	78.17	2.36	8.23	8	72
SASSNet^[51]	89.27	80.82	3.13	8.83	16	64
SASSNet^[51]	87.54	78.05	2.59	9.84	8	72
URPC^[52]	86.92	77.03	2.28	11.13	8	72
MC-Net^[54]	90.34	82.48	1.77	6.00	16	64
MC-Net^[54]	87.62	78.25	1.82	10.03	8	72
CCT^[55]	88.83	80.06	2.49	8.44	16	64
DAN^[56]	87.52	78.29	2.42	9.01	16	64
SS-Net^[57]	88.55	79.62	1.90	7.49	8	72
BCP^[58]	89.62	81.31	1.76	6.81	8	72

Table 3 Lists of common medical image segmentation datasets

数据集名称	器官	模态	规模/张	数据集描述
Spinal Cord Grey Matter Segmentation Challenge	脊髓	MRI	80	脊髓T2加权MRI，用于脊柱椎体分割 http://cmictig.cs.ucl.ac.uk/niftyweb/
2018 Left Atrium Segmentation Challenge	心脏	MRI	154	3D LGE-MRI完整心房扫描图像 https://www.cardiacatlas.org/atriaseg2018-challenge/ atria-seg-data/
Automatic Cardiac Diagnosis Challenge (ACDC)		MRI	150	心脏图像及左、右心室和心肌的标签 http://acdc.creatis.insa-lyon.fr/
MM-WHS 2017		MRI，CT	120	用于多模态全心分割 https://zmiclab.github.io/zxh/0/mmwhs/
CAMELYON16	细胞	WSI	400	用于乳腺癌淋巴结转移检测和病理分析 https://camelyon16.grand-challenge.org/
MoNuSeg			44	用于细胞核分割和病理图像分析 https://monuseg.grand-challenge.org/
CRAG			213	用于结直肠腺癌腺体和息肉分割 https://kddcup24.github.io/
DSB2018			1 039	用于细胞核分割 https://www.kaggle.com/c/data-science-bowl-2018/data

Table 4 Lists of common performance metrics for medical image segmentation

性能评价指标	定义	计算公式	取值范围	取值含义
Dice	预测结果与真实值间的相似度	$D i c e = 2 × X ⋂ Y X + Y$	[0,1]	数值越大表明分割效果越好
Jaccard	预测结果与真实值间的交并比	$J a c c a r d = X ⋂ Y X ⋃ Y$
准确率（Accuracy）	分割结果预测正确的像素比例	$A c c u r a c y = T P + T N T P + T N + F P + F N$
召回率（Recall）	正类真实值样本正确预测比例	$R e c a l l = T P T P + F N$
特异性（Specificity）	负类真实值样本正确预测比例	$S p e c i f i c i t y = T N T N + F P$
体积相似度 (VS)	分割结果与真实值体积相似度	$V S = 1 - X - Y X + Y$
Hausdorff距离	预测结果与真实值边界匹配程度	$H D A, B = m a x s u p α ∈ A a, B, s u p b ∈ B b, A$	[0, $∞)$	数值越小表明分割效果越好
ASD	预测结果与真实值平均表面距离	$A S D = 1 X + Y ∑ x ∈ X m i n y ∈ Y d x, y + ∑ y ∈ Y m i n x ∈ X d x, y$	[0, $∞)$	数值越小表明分割效果越好

Table 4 Lists of common performance metrics for medical image segmentation

性能评价指标	定义	计算公式	取值范围	取值含义
Dice	预测结果与真实值间的相似度	$D i c e = 2 × X ⋂ Y X + Y$	[0,1]	数值越大表明分割效果越好
Jaccard	预测结果与真实值间的交并比	$J a c c a r d = X ⋂ Y X ⋃ Y$
准确率（Accuracy）	分割结果预测正确的像素比例	$A c c u r a c y = T P + T N T P + T N + F P + F N$
召回率（Recall）	正类真实值样本正确预测比例	$R e c a l l = T P T P + F N$
特异性（Specificity）	负类真实值样本正确预测比例	$S p e c i f i c i t y = T N T N + F P$
体积相似度 (VS)	分割结果与真实值体积相似度	$V S = 1 - X - Y X + Y$
Hausdorff距离	预测结果与真实值边界匹配程度	$H D A, B = m a x s u p α ∈ A a, B, s u p b ∈ B b, A$	[0, $∞)$	数值越小表明分割效果越好
ASD	预测结果与真实值平均表面距离	$A S D = 1 X + Y ∑ x ∈ X m i n y ∈ Y d x, y + ∑ y ∈ Y m i n x ∈ X d x, y$	[0, $∞)$	数值越小表明分割效果越好

References 58

[1]	Ronneberger O， Fischer P， Brox T. U-Net： convolutional networks for biomedical image segmentation［C］//Medical Image Computing and Computer Assisted Intervention—MICCAI 2015. Cham： Springer， 2015： 234-241.
[2]	Long J， Shelhamer E， Darrell T. Fully convolutional networks for semantic segmentation［C］//IEEE Conference on Computer Vision and Pattern Recognition （CVPR）. Boston： IEEE， 2015： 3431-3440.
[3]	Vaswani A， Shazeer N， Parmar N， et al. Attention is all you need［J］. Advances in Neural Information Processing Systems， 2017， 30： 5998-6008.
[4]	Laine S， Aila T. Temporal ensembling for semi-supervised learning［EB/OL］. （2016-10-07）［2024-05-27］..
[5]	Tarvainen A， Valpola H. Mean teachers are better role models： weight-averaged consistency targets improve semi-supervised deep learning results［J］. Advances in Neural Information Processing Systems， 2017， 30： 1195-1204.
[6]	Miyato T， Maeda S I， Koyama M， et al. Virtual adversarial training： a regularization method for supervised and semi-supervised learning［J］. IEEE Transactions on Pattern Analysis and Machine Intelligence， 2019， 41（8）： 1979-1993.
[7]	Verma V， Kawaguchi K， Lamb A， et al. Interpolation consistency training for semi-supervised learning［J］. Neural Networks， 2022， 145： 90-106.
[8]	Fu Y G， Robu M R， Koo B， et al. More unlabelled data or label more data？ a study on semi-supervised laparoscopic image segmentation［C］//Domain Adaptation and Representation Transfer and Medical Image Learning with Less Labels and Imperfect Data. Cham： Springer， 2019： 173-180.
[9]	Zhou Y N， Chen H， Lin H J， et al. Deep semi-supervised knowledge distillation for overlapping cervical cell instance segmentation［C］//Medical Image Computing and Computer Assisted Intervention—MICCAI 2020. Cham： Springer， 2020： 521-531.
[10]	Wang G T， Liu X L， Li C P， et al. A noise-robust framework for automatic segmentation of COVID-19 pneumonia lesions from CT images［J］. IEEE Transactions on Medical Imaging， 2020， 39（8）： 2653-2663.
[11]	Xu X A， Sanford T， Turkbey B， et al. Shadow-consistent semi-supervised learning for prostate ultrasound segmentation［J］. IEEE Transactions on Medical Imaging， 2022， 41（6）： 1331-1345.
[12]	Yang Y L， Song H， Lin Y C， et al. PrixMatch： semi-supervised network for multi-modal medical image segmentation with cross-modal data augmentation and adaptive prior knowledge thresholding［C］//IEEE International Conference on Bioinformatics and Biomedicine （BIBM）. Lisbon： IEEE， 2024： 2766-2771.
[13]	Cai C Y， He J J， Zhang M L， et al. Patch shuffle and pixel contrast： dual consistency learning for semi-supervised lung tumor segmentation［C］// Pattern Recognition and Computer Vision. Singapore： Springer， 2024： 504-516.
[14]	He A L， Li T， Wu Y L， et al. FRCNet： frequency and region consistency for semi-supervised medical image segmentation［C］// Medical Image Computing and Computer Assisted Intervention—MICCAI 2024. Cham： Springer， 2024： 305-315.
[15]	Zhao Z Y， Xu K X， Yeo H Z， et al. MS-MT： multi-scale mean teacher with contrastive unpaired translation for cross-modality vestibular schwannoma and cochlea segmentation［C］// Brainlesion： Glioma， Multiple Sclerosis， Stroke and Traumatic Brain Injuries. Cham： Springer， 2023： 68-78.
[16]	Aralikatti R C， Pawan S J， Rajan J. A dual-stage semi-supervised pre-training approach for medical image segmentation［J］. IEEE Transactions on Artificial Intelligence， 2024， 5（2）： 556-565.
[17]	Xu Z， Wang Y X， Lu D H， et al. Ambiguity-selective consistency regularization for mean-teacher semi-supervised medical image segmentation［J］. Medical Image Analysis， 2023， 88： 102880.
[18]	Wang Y X， Zhang Y， Tian J， et al. Double-uncertainty weighted method for semi-supervised learning［C］//Medical Image Computing and Computer Assisted Intervention—MICCAI 2020. Cham： Springer， 2020： 542-551.
[19]	Zhang Y C， Jiao R S， Liao Q C， et al. Uncertainty-guided mutual consistency learning for semi-supervised medical image segmentation［J］. Artificial Intelligence in Medicine， 2023， 138： 102476.
[20]	Hang W L， Feng W， Liang S， et al. Local and global structure-aware entropy regularized mean teacher model for 3D left atrium segmentation［C］//Medical Image Computing and Computer Assisted Intervention—MICCAI 2020. Cham： Springer， 2020： 562-571.
[21]	Gao H M， Wang H Y， Chen L X， et al. Semi-supervised segmentation of hyperspectral pathological imagery based on shape priors and contrastive learning［J］. Biomedical Signal Processing and Control， 2024， 91： 105881.
[22]	Yang X， Li H M， Wang Y， et al. Contrastive rendering with semi-supervised learning for ovary and follicle segmentation from 3D ultrasound［J］. Medical Image Analysis， 2021， 73： 102134.
[23]	Jin Q G， Cui H， Sun C M， et al. Inter- and intra-uncertainty based feature aggregation model for semi-supervised histopathology image segmentation［J］. Expert Systems with Applications， 2024， 238： 122093.
[24]	Yu L Q， Wang S J， Li X M， et al. Uncertainty-aware self-ensembling model for semi-supervised 3D left atrium segmentation［C］//Medical Image Computing and Computer Assisted Intervention—MICCAI 2019. Cham： Springer， 2019： 605-613.
[25]	Xiang J Y， Qiu P， Yang Y. FUSSNet： fusing two sources of uncertainty for semi-supervised medical image segmentation［C］//Medical Image Computing and Computer Assisted Intervention—MICCAI 2022. Cham： Springer， 2022： 481-491.
[26]	Guo X R， Sun K， Zheng Y J. Multi-scale consistency adversarial learning for semi-supervised 3D medical image segmentation［J］. Biomedical Signal Processing and Control， 2025， 103： 107475.
[27]	Wu Y C， Li X Q， Zhou Y. Uncertainty-aware representation calibration for semi-supervised medical imaging segmentation［J］. Neurocomputing， 2024， 595： 127912.
[28]	Hu L， Li J X， Peng X C， et al. Semi-supervised NPC segmentation with uncertainty and attention guided consistency［J］. Knowledge-Based Systems， 2022， 239： 108021.
[29]	Zhou Q， Feng Y W， Huang Z W， et al. Specific instance and cross-prompt-based robust 3-D semi-supervised medical image segmentation［J］. IEEE Transactions on Instrumentation and Measurement， 2024， 73： 4009414.
[30]	Meng H， Zhao H C， Yu Z N， et al. Uncertainty-aware mean teacher framework with inception and squeeze-and-excitation block for MICCAI FLARE22 challenge［C］//Fast and Low-Resource Semi-supervised Abdominal Organ Segmentation. Cham： Springer， 2022： 245-259.
[31]	Xu A， Wang S Y， Fan J Y， et al. Dual attention based uncertainty-aware mean teacher model for semi-supervised cardiac image segmentation［C］// IEEE International Conference on Progress in Informatics and Computing （PIC）. Shanghai： IEEE， 2021： 82-86.
[32]	Farooq M U， Ullah Z， Gwak J. Residual attention based uncertainty-guided mean teacher model for semi-supervised breast masses segmentation in 2D ultrasonography［J］. Computerized Medical Imaging and Graphics， 2023， 104： 102173.
[33]	Lei T， Zhang D， Du X G， et al. Semi-supervised medical image segmentation using adversarial consistency learning and dynamic convolution network［J］. IEEE Transactions on Medical Imaging， 2023， 42（5）： 1265-1277.
[34]	Wang K P， Zhan B， Zu C， et al. Semi-supervised medical image segmentation via a tripled-uncertainty guided mean teacher model with contrastive learning［J］. Medical Image Analysis， 2022， 79： 102447.
[35]	Lyu J， Sui B， Wang C Y， et al. Adaptive feature aggregation based multi-task learning for uncertainty-guided semi-supervised medical image segmentation［J］. Expert Systems with Applications， 2023， 232： 120836.
[36]	Cheng H T， Yeh C F， Kuo P C， et al. Self-similarity student for partial label histopathology image segmentation［C］//Computer Vision—ECCV 2020. Cham： Springer， 2020： 117-132.
[37]	Huang H B， Cui Y， Shi M X， et al. Semi-supervised cerebrovascular segmentation using TOF-MRA images based on label refinement and consistency regularization［C］//2024 IEEE International Symposium on Biomedical Imaging （ISBI）. Piscataway： IEEE，2024： 1-5.
[38]	Qiu Y， Lu H T， Mei J， et al. Towards semi-supervised multi-modal rectal cancer segmentation： a large-scale dataset and a multi-teacher uncertainty-aware network［J］. Expert Systems with Applications， 2024， 255： 124734.
[39]	Zou J G， Wang Z H， Du X Q. A double-teacher model capable of exploiting isomorphic and heterogeneous discrepancy information for medical image segmentation［J］. Diagnostics， 2023， 13（11）： 1971.
[40]	Li B L， Wang Y， Xu Y M， et al. DSST： a dual student model guided student-teacher framework for semi-supervised medical image segmentation［J］. Biomedical Signal Processing and Control， 2024， 90： 105890.
[41]	Hu G， Zhao F， Houssein E H. Competitive dual-students using bi-level contrastive learning for semi-supervised medical image segmentation［J］. Engineering Applications of Artificial Intelligence， 2025， 144： 110082.
[42]	Li W， Bian R F， Zhao W Y， et al. Diversity matters： cross-head mutual mean-teaching for semi-supervised medical image segmentation［J］. Medical Image Analysis， 2024， 97： 103302.
[43]	Zhu J Y， Bolsterlee B， Chow B V Y， et al. Hybrid dual mean-teacher network with double-uncertainty guidance for semi-supervised segmentation of MRI scans［EB/OL］. （2023-03-09）［2024-05-11］. .
[44]	Zhou Q， Yu B， Xiao F， et al. Robust semi-supervised 3D medical image segmentation with diverse joint-task learning and decoupled inter-student learning［J］. IEEE Transactions on Medical Imaging， 2024， 43（6）： 2317-2331.
[45]	Qiu Y T， Meng J， Li B H. Semi-supervised strong-teacher consistency learning for few-shot cardiac MRI image segmentation［J］. Computer Methods and Programs in Biomedicine， 2025， 261： 108613.
[46]	Li F X， Jiang A L， Li M Y， et al. HPFG： semi-supervised medical image segmentation framework based on hybrid pseudo-label and feature-guiding［J］. Medical & Biological Engineering & Computing， 2024， 62（2）： 405-421.
[47]	Lin C， Liu Y Z， Wang D H， et al. S-ACMT： enhancing semi-supervised medical image segmentation with SAM as a cross-supervised secondary teacher［C］//2024 9th International Conference on Signal and Image Processing （ICSIP）. Piscataway： IEEE， 2024： 774-779.
[48]	Mei C Y， Yang X G， Zhou M， et al. Semi-supervised image segmentation using a residual-driven mean teacher and an exponential Dice loss［J］. Artificial Intelligence in Medicine， 2024， 148： 102757.
[49]	Gao N， Zhou S P， Wang L， et al. PMT： progressive mean teacher via exploring temporal consistency for semi-supervised medical image segmentation［C］//Computer Vision—ECCV 2024. Cham： Springer， 2024： 144-160.
[50]	Luo X D， Chen J N， Song T， et al. Semi-supervised medical image segmentation through dual-task consistency［J］. Proceedings of the AAAI Conference on Artificial Intelligence， 2021， 35（10）： 8801-8809.
[51]	Li S L， Zhang C Y， He X M. Shape-aware semi-supervised 3D semantic segmentation for medical images［C］//Medical Image Computing and Computer Assisted Intervention—MICCAI 2020. Cham： Springer， 2020： 552-561.
[52]	Luo X D， Liao W J， Chen J N， et al. Efficient semi-supervised gross target volume of nasopharyngeal carcinoma segmentation via uncertainty rectified pyramid consistency［C］//Medical Image Computing and Computer Assisted Intervention—MICCAI 2021. Cham： Springer， 2021： 318-329.
[53]	Chen X K， Yuan Y H， Zeng G， et al. Semi-supervised semantic segmentation with cross pseudo supervision［C］//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition （CVPR）. Nashville： IEEE， 2021： 2613-2622.
[54]	Wu Y C， Xu M F， Ge Z Y， et al. Semi-supervised left atrium segmentation with mutual consistency training［C］// Medical Image Computing and Computer Assisted Intervention—MICCAI 2021. Cham： Springer， 2021： 297-306.
[55]	Ouali Y， Hudelot C， Tami M. Semi-supervised semantic segmentation with cross-consistency training［C］//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition （CVPR）. Seattle： IEEE， 2020： 12671-12681.
[56]	Zhang Y Z， Yang L， Chen J X， et al. Deep adversarial networks for biomedical image segmentation utilizing unannotated images［C］// Medical Image Computing and Computer Assisted Intervention—MICCAI 2017. Cham： Springer， 2017： 408-416.
[57]	Wu Y C， Wu Z H， Wu Q Y， et al. Exploring smoothness and class-separation for semi-supervised medical image segmentation［C］// Medical Image Computing and Computer Assisted Intervention—MICCAI 2022. Cham： Springer， 2022： 34-43.
[58]	Bai Y H， Chen D W， Li Q L， et al. Bidirectional copy-paste for semi-supervised medical image segmentation［C］//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition （CVPR）. Vancouver： IEEE， 2023： 11514-11524.