pith. sign in

arxiv: 2607.02051 · v1 · pith:ELDCYUPQnew · submitted 2026-07-02 · 💻 cs.CV

Embracing Intra-Class Heterogeneity for Semi-Supervised Medical Image Segmentation: From Diversity to Precision

Yuqi Liu , Yufei Chen , Wei Fu , Xiaodong Yue , Shuo Li This is my paper

Pith reviewed 2026-07-03 15:55 UTC · model grok-4.3

classification 💻 cs.CV
keywords semi-supervised medical image segmentationintra-class heterogeneitymultiple prototypescontrastive learningintensity alignmentlimited labeled dataprototype optimization
0
0 comments X

The pith

MPCL generates multiple intensity-aligned prototypes to model intra-class heterogeneity and improve precision in semi-supervised medical image segmentation.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper seeks to establish that semi-supervised medical image segmentation improves when intra-class heterogeneity is explicitly modeled rather than averaged into uniform representations. Many anatomical structures show varying intensity patterns within the same class, and scarce labels make this harder to capture, leading current methods to imprecise boundaries. MPCL introduces three components that create diverse prototypes aligned to intensity, optimize their space for discriminability, and transfer that knowledge to the segmentation network. If this holds, segmentation accuracy rises especially when only a tiny fraction of data is labeled. A reader would care because expert annotations are costly in medicine, so better use of unlabeled scans could directly aid clinical tasks like tumor outlining.

Core claim

MPCL is an SSMIS framework that achieves better diversity and precision by generating multiple prototypes aligned with intensity characteristics through IHPG, systematically optimizing a more discriminative prototypical space through PSO, and promoting knowledge transfer to the segmentation network through DKA, resulting in significant outperformance over existing methods on three datasets with marked intra-class heterogeneity, particularly under extremely limited labeled data.

What carries the argument

Multiple Prototype Contrastive Learning (MPCL) framework that generates and aligns multiple intensity-based prototypes to represent heterogeneous patterns within the same class.

If this is right

  • Segmentation boundaries become more accurate for structures that vary in intensity within the same class.
  • Performance gains are largest when labeled data is reduced to very small fractions of the training set.
  • The prototypical space becomes both more discriminative across classes and more generalizable within classes.
  • Knowledge from the optimized prototypes transfers directly into higher-precision pixel predictions by the network.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same prototype-generation logic could be tested on non-medical images that also contain intra-class texture or lighting variation.
  • Combining MPCL with existing consistency-regularization techniques might compound the gains under label scarcity.
  • If the intensity alignment step is replaced by learned feature alignment, the method might extend to modalities where intensity is less dominant.
  • Deployment in clinical pipelines would require checking whether the added prototype computation increases inference latency.

Load-bearing premise

Intra-class heterogeneity appears mainly as intensity differences that multiple aligned prototypes can capture and transfer into the segmentation network without creating overfitting or boundary artifacts.

What would settle it

On the three medical image datasets, train MPCL and baseline SSMIS methods with 5 percent or fewer labeled samples and measure Dice or Hausdorff scores; if MPCL shows no consistent gain, the claim that the three components produce better precision from heterogeneity fails.

Figures

Figures reproduced from arXiv: 2607.02051 by Shuo Li, Wei Fu, Xiaodong Yue, Yufei Chen, Yuqi Liu.

Figure 1
Figure 1. Figure 1: Paradigms of (a) Consistency Regularization-based (CR) SSMIS methods and (b) Prototype Learning-based (PL) SSMIS methods. improves proportionally with the discriminative power of the learned prototypes. : Preprint submitted to Elsevier Page 1 of 15 [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Limitations of the existing PL methods and the advantages of our MPCL that break through the bottleneck of the PL paradigm and achieve complete and precise segmentation results from limited labeled data. (a) and (c) are the limitations of the existing PL method. (b) and (d) are the corresponding solutions and advantages of our MPCL, respectively. (a) Existing methods only have single prototype for one repr… view at source ↗
Figure 3
Figure 3. Figure 3: Overflow of our proposed method Multiple Prototypes Contrastive Learning (MPCL), which consists of three designs, including Intensity-aligned Heterogeneous Prototype Generation (IHPG) providing representations with better diversity via effectively modeling intra-class heterogeneity guided by intensity characteristics, Prototypical Space Optimization (PSO) systematically establishing a solid foundation for … view at source ↗
Figure 4
Figure 4. Figure 4: To demonstrate the improvement on segmenta￾tion precision via embracing intra-class heterogeneity, exper￾iments are conducted on three datasets (LA, Pan-NIH, and BraTS2019) with intra-class heterogeneity on both foreground and background. volumes. We utilized T2-FLAIR for whole brain tumor segmentation due to its optimal visualization of the entire tumor region. 4.2. Implementation Details and Metrics 1) I… view at source ↗
Figure 5
Figure 5. Figure 5: Radar maps comparison shows the superiority of our MPCL in all metrics, especially in boundary precision, on LA, Pan-NIH and BraTS2019 dataset under 5% and 10% labeled data setting. All metrics are normalized with respect to the baseline V-NET. 4.3.2. Qualitative Evaluation for Visual Superiority [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Visualization comparison indicates more precise segmentation of our MPCL on examples of 2D slices on LA, Pan-NIH and BraTS2019 dataset under 5% labeled data setting. Most other methods struggle to process heterogeneous foreground and to distinguish similar foreground and background, thus causing under-segmentation in the foreground and over-segmentation (yellow arrow) in the background. a more robust and g… view at source ↗
Figure 8
Figure 8. Figure 8: t-SNE visualization of ForeGround voxel features and corresponding prototype features under varying prototype numbers (𝐾 = 2 to 𝐾 = 5) on the same BraTS2019 sample as [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗
Figure 7
Figure 7. Figure 7: GMM decomposition of the intensity distribution under different component number settings reveals that in￾creasing the component number beyond 3 leads to redun￾dant subdivision within existing intensity subregions, where individual components lose their representativeness of distinct intensity-based patterns. IHPG effectively captures the intrinsic intra-class hetero￾geneity through intensity characteristi… view at source ↗
Figure 9
Figure 9. Figure 9: Grad-CAM visualization comparison between our MPCL and other PL SSMIS methods (including MPL methods MPER [10] and BaPC [9]; SPL methods CPCL [8] and UPCoL [7]) on BraTS2019 demonstrates that MPCL attends to more diverse intra-class heterogeneity structures while other MPL and SPL methods only capture partial structures. Ground truth is shown in yellow contours in the first column. Dic e S c o r e (%) ↑ [… view at source ↗
Figure 10
Figure 10. Figure 10: Dice score (%) comparison of prototype generation strategies on three datasets under 10% labeled data setting. IHPG with intensity-driven initialization (INT) consistently outperforms both random selection and random initialization (IHPG w/o INT), demonstrating the effectiveness of leveraging intensity characteristics for prototype generation. arbitrary prototype placement fails to capture meaningful stru… view at source ↗
read the original abstract

Due to the scarcity of expert-annotated data, Semi-Supervised Medical Image Segmentation (SSMIS) has emerged as a promising approach. Many anatomical structures in medical images exhibit significant intra-class heterogeneity, with different regions showing heterogeneous intensity patterns within the same structure. However, existing methods inadequately exploit this intensity-manifested intra-class heterogeneity, resulting in uniform structural representations and imprecise segmentation. Furthermore, the scarcity of labeled data makes it more difficult to effectively capture such complex heterogeneity. To address this, we propose Multiple Prototype Contrastive Learning (MPCL), an SSMIS framework that possesses better diversity and better precision. It consists of three novel designs: First, we provide structural representations with better diversity and propose Intensity-aligned Heterogeneous Prototype Generation (IHPG) that effectively models intra-class heterogeneity by generating multiple prototypes aligned with intensity characteristics. Second, we further enhance more diverse structural representations and build a solid foundation for more precise segmentation through Prototypical Space Optimization (PSO) that systematically optimizes a more discriminative and generalizable prototypical space. Finally, we achieve segmentation results with better precision through Dual-branch Knowledge Alignment (DKA) that efficiently promotes intra-class heterogeneity knowledge transfer from prototypical space to the segmentation network. Extensive experiments on three medical image datasets with significant intra-class heterogeneity demonstrate that MPCL significantly outperforms existing methods, especially under extremely limited labeled data.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes Multiple Prototype Contrastive Learning (MPCL) for semi-supervised medical image segmentation to better exploit intra-class heterogeneity manifested in intensity patterns. The method introduces three components: Intensity-aligned Heterogeneous Prototype Generation (IHPG) to create multiple aligned prototypes, Prototypical Space Optimization (PSO) to build a more discriminative prototypical space, and Dual-branch Knowledge Alignment (DKA) to transfer heterogeneity knowledge to the segmentation head. It claims superior performance over existing methods on three medical image datasets, especially under extremely limited labeled data.

Significance. If validated with rigorous experiments, the work addresses a practically relevant challenge in medical imaging where anatomical structures show high intra-class variation and annotations are scarce. Modeling heterogeneity via multiple prototypes could lead to more precise segmentations in low-label regimes.

major comments (2)
  1. [Abstract] Abstract: The central claim of significant outperformance on three datasets is asserted without any quantitative results, baselines, error bars, or experimental protocol details, preventing verification of the contribution.
  2. [Method (PSO)] Method (PSO component): The claim that PSO produces a more discriminative and generalizable prototypical space is load-bearing for the precision gains, yet under extremely limited labels the optimization risks collapsing to sample-specific intensity modes; no regularization, temperature scheduling, or validation against overfitting is described to ensure generalization.
minor comments (1)
  1. [Abstract] Abstract: Phrases such as 'better diversity and better precision' are imprecise; quantitative or comparative language would improve clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address each major comment point by point below and will incorporate revisions to strengthen the presentation.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim of significant outperformance on three datasets is asserted without any quantitative results, baselines, error bars, or experimental protocol details, preventing verification of the contribution.

    Authors: We agree that the abstract would be strengthened by including quantitative support for the performance claims. In the revised manuscript, we will update the abstract to report key metrics (e.g., mean Dice scores across the three datasets), reference the main baselines, and note the use of error bars along with the evaluation protocol. revision: yes

  2. Referee: [Method (PSO)] Method (PSO component): The claim that PSO produces a more discriminative and generalizable prototypical space is load-bearing for the precision gains, yet under extremely limited labels the optimization risks collapsing to sample-specific intensity modes; no regularization, temperature scheduling, or validation against overfitting is described to ensure generalization.

    Authors: We appreciate this observation on the PSO component. The current description emphasizes optimization for a discriminative and generalizable space but does not explicitly detail safeguards against overfitting to sample-specific modes under limited labels. We will revise the PSO subsection to include the contrastive loss formulation (with temperature parameter), any regularization applied, and evidence from ablations or validation that the space generalizes. revision: yes

Circularity Check

0 steps flagged

No circularity; derivation self-contained with no equations or self-referential reductions visible

full rationale

The abstract presents MPCL as a new framework with three components (IHPG for generating intensity-aligned prototypes, PSO for optimizing the prototypical space, and DKA for knowledge transfer) to address intra-class heterogeneity in SSMIS. No equations, parameter-fitting steps, or citations appear in the provided text. No load-bearing premise reduces to a self-definition, fitted input renamed as prediction, or self-citation chain. Claims rest on experimental results rather than a closed derivation, so no circular steps can be identified. This is the expected honest non-finding when the source contains no inspectable mathematical chain.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities can be extracted or verified.

pith-pipeline@v0.9.1-grok · 5784 in / 1146 out tokens · 28968 ms · 2026-07-03T15:55:41.829911+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

42 extracted references · 42 canonical work pages

  1. [1]

    Semi- supervised information fusion for medical image analysis: Recent progress and future perspectives.Information Fusion, 106:102263, 2024

    Ying Weng, Yiming Zhang, Wenxin Wang, and Tom Dening. Semi- supervised information fusion for medical image analysis: Recent progress and future perspectives.Information Fusion, 106:102263, 2024

    work page 2024

  2. [2]

    Deep semi-supervised learning for medical image segmentation: A review.Expert Systems with Applications, page 123052, 2024

    Kai Han, Victor S Sheng, Yuqing Song, Yi Liu, Chengjian Qiu, Siqi Ma, and Zhe Liu. Deep semi-supervised learning for medical image segmentation: A review.Expert Systems with Applications, page 123052, 2024

    work page 2024

  3. [3]

    Learning with limited annotations: a survey on deep semi-supervised learning for medical image segmentation

    RushiJiao,YichiZhang,LeDing,BingsenXue,JicongZhang,Rong Cai, and Cheng Jin. Learning with limited annotations: a survey on deep semi-supervised learning for medical image segmentation. Computers in Biology and Medicine, page 107840, 2023

    work page 2023

  4. [4]

    Uncertainty-aware self-ensembling model for semi- supervised3dleftatriumsegmentation

    Lequan Yu, Shujun Wang, Xiaomeng Li, Chi-Wing Fu, and Pheng- Ann Heng. Uncertainty-aware self-ensembling model for semi- supervised3dleftatriumsegmentation. InMedicalimagecomputing and computer assisted intervention–MICCAI 2019: 22nd interna- tional conference, Shenzhen, China, October 13–17, 2019, proceed- ings, part II 22, pages 605–613. Springer, 2019

    work page 2019

  5. [5]

    Semi- supervised medical image segmentation through dual-task consis- tency

    Xiangde Luo, Jieneng Chen, Tao Song, and Guotai Wang. Semi- supervised medical image segmentation through dual-task consis- tency. InProceedings of the AAAI conference on artificial intelli- gence, volume 35, pages 8801–8809, 2021

    work page 2021

  6. [6]

    Triple-task mutual consistency for semi-supervised 3d medical image segmentation.Computers in Biology and Medicine, 175:108506, 2024

    Yantao Chen, Yong Ma, Xiaoguang Mei, Lin Zhang, Zhigang Fu, and Jiayi Ma. Triple-task mutual consistency for semi-supervised 3d medical image segmentation.Computers in Biology and Medicine, 175:108506, 2024

    work page 2024

  7. [7]

    Upcol: uncertainty-informed prototype con- sistency learning for semi-supervised medical image segmentation

    Wenjing Lu, Jiahao Lei, Peng Qiu, Rui Sheng, Jinhua Zhou, Xinwu Lu, and Yang Yang. Upcol: uncertainty-informed prototype con- sistency learning for semi-supervised medical image segmentation. InInternational Conference on Medical Image Computing and Computer-Assisted Intervention, pages 662–672. Springer, 2023

    work page 2023

  8. [8]

    All-around reallabelsupervision:Cyclicprototypeconsistencylearningforsemi- supervisedmedicalimagesegmentation.IEEEJournalofBiomedical and Health Informatics, 26(7):3174–3184, 2022

    Zhe Xu, Yixin Wang, Donghuan Lu, Lequan Yu, Jiangpeng Yan, Jie Luo,KaiMa,YefengZheng,andRaymondKai-yuTong. All-around reallabelsupervision:Cyclicprototypeconsistencylearningforsemi- supervisedmedicalimagesegmentation.IEEEJournalofBiomedical and Health Informatics, 26(7):3174–3184, 2022

    work page 2022

  9. [9]

    Boundary-aware prototype in semi-supervised medical image seg- mentation.IEEE Transactions on Image Processing, 2024

    Yongchao Wang, Bin Xiao, Xiuli Bi, Weisheng Li, and Xinbo Gao. Boundary-aware prototype in semi-supervised medical image seg- mentation.IEEE Transactions on Image Processing, 2024

    work page 2024

  10. [10]

    InICASSP 2025 - 2025 IEEE International Conference on Acoustics,SpeechandSignalProcessing(ICASSP),pages1–5,2025

    YaliBi,EnyuChe,YinanChen,YuanpengHe,andJingweiQu.Multi- prototype-based embedding refinement for medical image segmen- tation. InICASSP 2025 - 2025 IEEE International Conference on Acoustics,SpeechandSignalProcessing(ICASSP),pages1–5,2025

    work page 2025

  11. [11]

    Im- proving segmentation and detection of lesions in ct scans using intensity distribution supervision.Computerized Medical Imaging and Graphics, 108:102259, 2023

    Seung Yeon Shin, Thomas C Shen, and Ronald M Summers. Im- proving segmentation and detection of lesions in ct scans using intensity distribution supervision.Computerized Medical Imaging and Graphics, 108:102259, 2023

    work page 2023

  12. [12]

    Comprehensive evaluation of op- timization algorithms for medical image segmentation.Scientific Reports, 15(1):37190, 2025

    Nijad A Al-Najdawi, Ali F Al-Shawabkeh, Sara Tedmori, Ibrahim I Ikhries, and Osama Dorgham. Comprehensive evaluation of op- timization algorithms for medical image segmentation.Scientific Reports, 15(1):37190, 2025

    work page 2025

  13. [13]

    Gaussian mixture models.Encyclopedia of biometrics, 741(659-663):3, 2009

    Douglas A Reynolds et al. Gaussian mixture models.Encyclopedia of biometrics, 741(659-663):3, 2009

    work page 2009

  14. [14]

    Constrained gaussianmixturemodelframeworkforautomaticsegmentationofmr brain images.IEEE transactions on medical imaging, 25(9):1233– 1245, 2006

    Hayit Greenspan, Amit Ruf, and Jacob Goldberger. Constrained gaussianmixturemodelframeworkforautomaticsegmentationofmr brain images.IEEE transactions on medical imaging, 25(9):1233– 1245, 2006

    work page 2006

  15. [15]

    Cross-domain medicalimagetranslationbysharedlatentgaussianmixturemodel

    YingyingZhu,YoubaoTang,YuxingTang,DanielCElton,Sungwon Lee, Perry J Pickhardt, and Ronald M Summers. Cross-domain medicalimagetranslationbysharedlatentgaussianmixturemodel. In Internationalconferenceonmedicalimagecomputingandcomputer- assisted intervention, pages 379–389. Springer, 2020

    work page 2020

  16. [16]

    Rednet: Reliable evidential discounting network for multi- modality medical image segmentation.IEEE Transactions on Medi- cal Imaging, 2025

    Shichen Sun, Yufei Chen, Xiaodong Yue, Chao Ma, and Xiahai Zhuang. Rednet: Reliable evidential discounting network for multi- modality medical image segmentation.IEEE Transactions on Medi- cal Imaging, 2025

    work page 2025

  17. [17]

    Target-aware u-net with fuzzy skip connec- tions for refined pancreas segmentation.Applied Soft Computing, 131:109818, 2022

    Yufei Chen, Chang Xu, Weiping Ding, Shichen Sun, Xiaodong Yue, and Hamido Fujita. Target-aware u-net with fuzzy skip connec- tions for refined pancreas segmentation.Applied Soft Computing, 131:109818, 2022. :Preprint submitted to Elsevier Page 14 of 15

    work page 2022

  18. [18]

    Medical image segmentation review: The success of u-net.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024

    RezaAzad,EhsanKhodapanahAghdam,AmelieRauland,YiweiJia, Atlas Haddadi Avval, Afshin Bozorgpour, Sanaz Karimijafarbigloo, Joseph Paul Cohen, Ehsan Adeli, and Dorit Merhof. Medical image segmentation review: The success of u-net.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024

    work page 2024

  19. [19]

    D-edl: Differ- ential evidential deep learning for robust medical out-of-distribution detection.Medical Image Analysis, page 103888, 2025

    Wei Fu, Yufei Chen, Yuqi Liu, and Xiaodong Yue. D-edl: Differ- ential evidential deep learning for robust medical out-of-distribution detection.Medical Image Analysis, page 103888, 2025

    work page 2025

  20. [20]

    Category-specific unlabeled data risk minimization for ultrasound semi-supervised segmentation.Medical Image Analysis, page 103773, 2025

    Lu Xu, Mingyuan Liu, Boxuan Wei, Yihua He, Zhifan Gao, Hong- bin Han, and Jicong Zhang. Category-specific unlabeled data risk minimization for ultrasound semi-supervised segmentation.Medical Image Analysis, page 103773, 2025

    work page 2025

  21. [21]

    Semisam+: Rethinking semi-supervised medical image segmentation in the era of foundation models.arXiv preprint arXiv:2502.20749, 2025

    Yichi Zhang, Bohao Lv, Le Xue, Wenbo Zhang, Yuchen Liu, Yu Fu, Yuan Cheng, and Yuan Qi. Semisam+: Rethinking semi-supervised medical image segmentation in the era of foundation models.arXiv preprint arXiv:2502.20749, 2025

    work page arXiv 2025

  22. [22]

    Semi-supervisedmedicalimagesegmentationviauncertainty rectified pyramid consistency.Medical Image Analysis, 80:102517, 2022

    Xiangde Luo, Guotai Wang, Wenjun Liao, Jieneng Chen, Tao Song, Yinan Chen, Shichuan Zhang, Dimitris N Metaxas, and Shaoting Zhang. Semi-supervisedmedicalimagesegmentationviauncertainty rectified pyramid consistency.Medical Image Analysis, 80:102517, 2022

    work page 2022

  23. [23]

    InIn- ternationalConferenceonMedicalImageComputingandComputer- Assisted Intervention, pages 481–491

    JinyiXiang,PengQiu,andYangYang.Fussnet:Fusingtwosourcesof uncertainty for semi-supervised medical image segmentation. InIn- ternationalConferenceonMedicalImageComputingandComputer- Assisted Intervention, pages 481–491. Springer, 2022

    work page 2022

  24. [24]

    Temporal ensembling for semi- supervisedlearning

    Samuli Laine and Timo Aila. Temporal ensembling for semi- supervisedlearning. InInternationalConferenceonLearningRepre- sentations, 2017

    work page 2017

  25. [25]

    AnttiTarvainenandHarriValpola.Meanteachersarebetterrolemod- els: Weight-averaged consistency targets improve semi-supervised deep learning results.Advances in neural information processing systems, 30, 2017

    work page 2017

  26. [26]

    Semi-supervised 3d medical image segmentation based on dual-task consistent joint learning and task-level regularization

    Qi-Qi Chen, Zhao-Hui Sun, Chuan-Feng Wei, Edmond Q Wu, and Dong Ming. Semi-supervised 3d medical image segmentation based on dual-task consistent joint learning and task-level regularization. IEEE/ACMTransactionsonComputationalBiologyandBioinformat- ics, 20(4):2457–2467, 2022

    work page 2022

  27. [27]

    Adaptive feature aggregation based multi-task learning for uncertainty-guided semi-supervised medical image segmentation.Expert Systems with Applications, 232:120836, 2023

    Jun Lyu, Bin Sui, Chengyan Wang, Qi Dou, and Jing Qin. Adaptive feature aggregation based multi-task learning for uncertainty-guided semi-supervised medical image segmentation.Expert Systems with Applications, 232:120836, 2023

    work page 2023

  28. [28]

    Self- supervisedcorrectionlearningforsemi-supervisedbiomedicalimage segmentation

    Ruifei Zhang, Sishuo Liu, Yizhou Yu, and Guanbin Li. Self- supervisedcorrectionlearningforsemi-supervisedbiomedicalimage segmentation. InMedical Image Computing and Computer Assisted Intervention–MICCAI 2021: 24th International Conference, Stras- bourg, France, September 27–October 1, 2021, Proceedings, Part II 24, pages 134–144. Springer, 2021

    work page 2021

  29. [29]

    Semi-supervised left atrium segmentation with mutual consistency training

    YichengWu,MinfengXu,ZongyuanGe,JianfeiCai,andLeiZhang. Semi-supervised left atrium segmentation with mutual consistency training. InInternational Conference on Medical Image Computing and Computer-Assisted Intervention, volume 12902, pages 297–306. Springer, Cham, 2021

    work page 2021

  30. [30]

    Few-shot semantic segmentation with prototype learning

    Nanqing Dong and Eric P Xing. Few-shot semantic segmentation with prototype learning. InBMVC, volume 3, page 4, 2018

    work page 2018

  31. [31]

    Psanet: prototype-guided salient attention for few-shot segmentation.The Visual Computer, pages 1–15, 2024

    Hao Li, Guoheng Huang, Xiaochen Yuan, Zewen Zheng, Xuhang Chen, Guo Zhong, and Chi-Man Pun. Psanet: prototype-guided salient attention for few-shot segmentation.The Visual Computer, pages 1–15, 2024

    work page 2024

  32. [32]

    Kp2l: Knowledge-driven pyramid prototype learning for semi- supervised medical image segmentation.Knowledge-Based Systems, 340:115662, 2026

    Yuqi Liu, Yufei Chen, Wei Fu, Xiaodong Yue, and Thierry Denœux. Kp2l: Knowledge-driven pyramid prototype learning for semi- supervised medical image segmentation.Knowledge-Based Systems, 340:115662, 2026

    work page 2026

  33. [33]

    Semi- supervised semantic segmentation with prototype-based consistency regularization.Advances in neural information processing systems, 35:26007–26020, 2022

    Haiming Xu, Lingqiao Liu, Qiuchen Bian, and Zhen Yang. Semi- supervised semantic segmentation with prototype-based consistency regularization.Advances in neural information processing systems, 35:26007–26020, 2022

    work page 2022

  34. [34]

    Adaptiveprototype learningand allocation for few-shot segmentation

    Gen Li, Varun Jampani, Laura Sevilla-Lara, Deqing Sun, Jonghyun Kim,and JoongkyuKim. Adaptiveprototype learningand allocation for few-shot segmentation. InProceedings of the IEEE/CVF confer- ence on computer vision and pattern recognition, pages 8334–8343, 2021

    work page 2021

  35. [35]

    Chunna Tian, Zhenxi Zhang, Xinbo Gao, Heng Zhou, Ran Ran, and ZhichengJiao.Animplicit-explicitprototypicalalignmentframework for semi-supervised medical image segmentation.IEEE Journal of Biomedical and Health Informatics, 28(2):929–940, 2023

    work page 2023

  36. [36]

    Mvpcl: multi-view prototype consis- tencylearningforsemi-supervisedmedicalimagesegmentation.The Visual Computer, pages 1–14, 2024

    Xiafan Li and Hongyan Quan. Mvpcl: multi-view prototype consis- tencylearningforsemi-supervisedmedicalimagesegmentation.The Visual Computer, pages 1–14, 2024

    work page 2024

  37. [37]

    Focal loss for dense object detec- tion

    T-YLPG Ross and GKHP Dollár. Focal loss for dense object detec- tion. Inproceedings of the IEEE conference on computer vision and pattern recognition, pages 2980–2988, 2017

    work page 2017

  38. [38]

    V-net: Fully convolutional neural networks for volumetric medical image segmentation

    Fausto Milletari, Nassir Navab, and Seyed-Ahmad Ahmadi. V-net: Fully convolutional neural networks for volumetric medical image segmentation. In2016 fourth international conference on 3D vision (3DV), pages 565–571. IEEE, 2016

    work page 2016

  39. [39]

    Unitbox:Anadvancedobjectdetectionnetwork

    JiahuiYu,YuningJiang,ZhangyangWang,ZhiminCao,andThomas Huang. Unitbox:Anadvancedobjectdetectionnetwork. InProceed- ings of the 24th ACM international conference on Multimedia, pages 516–520, 2016

    work page 2016

  40. [40]

    A global benchmark of algorithms for segmenting the left atrium from late gadolinium-enhanced cardiac magnetic res- onance imaging.Medical image analysis, 67:101832, 2021

    Zhaohan Xiong, Qing Xia, Zhiqiang Hu, Ning Huang, Cheng Bian, Yefeng Zheng, Sulaiman Vesal, Nishant Ravikumar, Andreas Maier, Xin Yang, et al. A global benchmark of algorithms for segmenting the left atrium from late gadolinium-enhanced cardiac magnetic res- onance imaging.Medical image analysis, 67:101832, 2021

    work page 2021

  41. [41]

    Data from pancreas-ct

    Holger R Roth, Amal Farag, E Turkbey, Le Lu, Jiamin Liu, and Ronald M Summers. Data from pancreas-ct. the cancer imaging archive.IEEE Transactions on Image Processing, 5, 2016

    work page 2016

  42. [42]

    The multimodal brain tumor image segmentation benchmark (brats).IEEE transactions on medical imaging, 34(10):1993–2024, 2014

    Bjoern H Menze, Andras Jakab, Stefan Bauer, Jayashree Kalpathy- Cramer, Keyvan Farahani, Justin Kirby, Yuliya Burren, Nicole Porz, Johannes Slotboom, Roland Wiest, et al. The multimodal brain tumor image segmentation benchmark (brats).IEEE transactions on medical imaging, 34(10):1993–2024, 2014. :Preprint submitted to Elsevier Page 15 of 15

    work page 1993