Abstract:Semi-supervised learning (SSL) techniques address the high labeling costs in 3D medical image segmentation, with the teacher-student model being a common approach. However, using an exponential moving average (EMA) in single-teacher models may cause coupling issues, where the weights of the student and teacher models become similar, limiting the teacher's ability to provide additional knowledge for the student. Dual-teacher models were introduced to address this problem but often neglected the importance of maintaining teacher model diversity, leading to coupling issues among teachers. To address the coupling issue, we incorporate a double-copy-paste (DCP) technique to enhance the diversity among the teachers. Additionally, we introduce the Staged Selective Ensemble (SSE) module, which selects different ensemble methods based on the characteristics of the samples and enables more accurate segmentation of label boundaries, thereby improving the quality of pseudo-labels. Experimental results demonstrate the effectiveness of our proposed method in 3D medical image segmentation tasks. Here is the code link: <a class="link-external link-https" href="https://github.com/Fazhan-cs/DCP" rel="external noopener nofollow">this https URL</a>.

What problem does this paper attempt to address?

### What problems does this paper attempt to solve? This paper aims to solve the problem of high - labeling cost in 3D medical image segmentation, especially in the semi - supervised learning (SSL) environment. Specifically, the authors focus on: 1. **The coupling problem in the single - teacher model**: In the traditional single - teacher model, using exponential moving average (EMA) may cause the weights of the student model and the teacher model to become similar, thus limiting the ability of the teacher model to provide additional knowledge to the student model. 2. **The lack of diversity in the double - teacher model**: Although the double - teacher model has been introduced to solve the coupling problem of the single - teacher model, existing methods often ignore maintaining the diversity between teacher models, resulting in a coupling problem between the two teachers as well. 3. **The quality problem of pseudo - label generation**: Existing semi - supervised learning methods usually adopt a fixed integration strategy when generating pseudo - labels without considering the characteristics of input samples, which may lead to low - quality pseudo - labels and affect the final segmentation performance. To solve these problems, the authors propose the following innovations: - **Double - Copy - Paste (DCP) technology**: By performing different copy - paste operations on the input data during the training process, the diversity between the double - teacher models is increased, thereby improving the learning effect of the model. - **Staged Selective Ensemble (SSE) module**: Dynamically select the integration strategy according to the prediction similarity of samples to ensure the generation of high - quality pseudo - labels, thereby improving the segmentation performance. These methods work together to fully utilize the limited labeled data and a large amount of unlabeled data, while avoiding the model coupling problem and generating high - quality pseudo - labels, thereby improving the overall performance of 3D medical image segmentation. ### Formula summary - **Mask generation and application in DCP operation**: \[ I_{aug}^{la} = I_{la}\odot M_a+I_{ua}\odot(1 - M_a) \] \[ I_{aug}^{ua}=I_{ua}\odot M_a+I_{la}\odot(1 - M_a) \] - **Pseudo - label generation in the SSE module**: \[ Y_{out1}=labaug_{la}\odot M+pb\odot(1 - M) \] \[ Y_{out2}=labaug_{ua}\odot M+y_b\odot(1 - M) \] - **Loss function**: \[ L_{in1}=L_{seg}(Q_{in1},Y_{out1})\odot Mask+\alpha\cdot L_{seg}(Q_{in1},Y_{out1})\odot(1 - Mask) \] \[ L_{in2}=L_{seg}(Q_{in2},Y_{out2})\odot(1 - Mask)+\alpha\cdot L_{seg}(Q_{in2},Y_{out2})\odot Mask \] \[ L_{all}=L_{in1}+L_{in2} \] Through these methods, the paper proposes a more effective semi - supervised 3D medical image segmentation framework.

Dual-Teacher Ensemble Models with Double-Copy-Paste for 3D Semi-Supervised Medical Image Segmentation

3D Graph-S<SUP>2</SUP>Net: Shape-Aware Self-ensembling Network for Semi-supervised Segmentation with Bilateral Graph Convolution

Robust Semi-supervised 3D Medical Image Segmentation with Diverse Joint-task Learning and Decoupled Inter-student Learning

DSST: A Dual Student Model Guided Student–teacher Framework for Semi-Supervised Medical Image Segmentation

Dual-Decoder Consistency via Pseudo-Labels Guided Data Augmentation for Semi-Supervised Medical Image Segmentation

Semi-supervised medical image segmentation via a tripled-uncertainty guided mean teacher model with contrastive learning

Alternate Diverse Teaching for Semi-supervised Medical Image Segmentation

DHC: Dual-debiased Heterogeneous Co-training Framework for Class-imbalanced Semi-supervised Medical Image Segmentation

Dual-Task Mutual Learning for Semi-Supervised Medical Image Segmentation

Specific Instance and Cross-Prompt-Based Robust 3-D Semi-Supervised Medical Image Segmentation

Dual Teacher Model for Semi-Supervised ABUS Tumor Segmentation.

Dual-scale Enhanced and Cross-generative Consistency Learning for Semi-supervised Medical Image Segmentation

InterTeach: A Novel Approach for Semi-Supervised Medical Image Segmentation Using Cooperative Teacher-Student Networks

Bidirectional Copy-Paste for Semi-Supervised Medical Image Segmentation

Semi-supervised Medical Image Segmentation through Dual-task Consistency

Co-training with High-Confidence Pseudo Labels for Semi-supervised Medical Image Segmentation

Dual-Teacher++: Exploiting Intra-domain and Inter-domain Knowledge with Reliable Transfer for Cardiac Segmentation

Diversity matters: Cross-head mutual mean-teaching for semi-supervised medical image segmentation

Semi-Supervised 3D Medical Image Segmentation Based on Dual-Task Consistent Joint Learning and Task-Level Regularization.

Multi-task contrastive learning for semi-supervised medical image segmentation with multi-scale uncertainty estimation

Competitive Ensembling Teacher-Student Framework for Semi-Supervised Left Atrium MRI Segmentation