Pith. sign in

REVIEW 5 major objections 5 minor 23 references

C3S3: Complementary Competition and Contrastive Selection for Semi-Supervised Medical Image Segmentation

T0 review · 5 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read C3S3 claims that combining two competing backbones that generate each other's pseudo-labels with a contrastive module trained on the intersection and union of their predicted masks improves boundary delineation enough to set new…

desk verdict Solid incremental combination of co-training and contrastive learning for semi-supervised segmentation; the abstract oversells the boundary gains and the ODCL masks lack a specified threshold, but the core idea is coherent and the ablations support it. read the letter →

arxiv 2506.07368 v2 pith:YOTUOHHZ submitted 2025-06-09 cs.CV cs.AI

Jiaying He , Yitong Lin , Jiahe Chen , Honghui Xu , Jianwei Zheng This is my paper · ORCID
classification cs.CVcs.AI
keywords semi-supervisedlearningmedicalimagesegmentationcontrastivepseudo-labelingconsistencyregularizationboundarydelineationleftatriumMRIpancreasCT
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Semi-supervised medical image segmentation must learn precise organ boundaries from mostly unlabeled volumes, and the paper argues the missing ingredient is a contrastive learning scheme driven by the model's own predicted masks rather than by ground-truth supervision. To this end, C3S3 couples two backbones, VNet and ResVNet, whose predictions compete through a weighted mechanism to generate pseudo-labels while a consistency loss keeps them aligned. A second module, the Outcome-Driven Contrastive Learning module, binarizes the two backbones' feature maps, treats their intersection as confident foreground and the region outside their union as confident background, and trains a dual-space intersection-union contrastive loss on those samples. On the left atrium MRI and Pancreas-CT benchmarks with only 20% of volumes labeled, the paper reports the best boundary metrics among compared methods, with 95HD and ASD improved by at least 6% relative to prior state of the art. The case for the paper is that fine boundary detail, not just regional overlap, can be recovered from unlabeled data by this combination of competitive pseudo-labeling and mask-driven contrastive selection.

What carries the argument

The load-bearing machinery is the Outcome-Driven Contrastive Learning module (ODCL) together with the Dynamic Complementary Competition module (DCC), mounted on two backbones, VNet and ResVNet. ODCL first projects each backbone's features into low-dimensional maps and binarizes them; the intersection of the two binary maps defines the confident foreground mask $M_\cap$, and the complement of their union defines the confident background mask, so voxels sampled from $M_\cap$ act as positive pairs while voxels outside are negatives. A dual-space intersection-union contrastive loss $L_{\mathrm{iuc}}$, computed across the four cross-backbone feature-map pairs (Eqs. 7-10), then pulls anchors toward their same-position positive and pushes them away from negatives, with the procedure repeated for background voxels using the union mask. DCC computes a weighted competition score $\alpha L_{\mathrm{ce}} + (1-\alpha)L_{\mathrm{dice}}$ on the labeled data, lets the higher-scoring backbone generate pseudo-labels for the other, and enforces cosine consistency between the two augmented predictions of each backbone. The total loss $L_{\mathrm{total}} = L_{\mathrm{seg}} + L_{\mathrm{cos}} + L_{\mathrm{unsup}}$ combines supervised segmentation, consistency, pseudo-label, and contrastive terms.

What would settle it

A reader could settle the claim by measuring mask quality directly: on the labeled volumes, compute the precision and recall of the intersection and union masks against ground truth, and then sweep the binarization threshold over a range such as 0.1 to 0.9. If the reported 95HD and ASD gains (at least 6% over TraCoCo and CauSSL) hold only inside a narrow threshold window, or if the masks' precision falls far below the final Dice, then the conclusion that the contrastive selection sharpens boundaries is not robust.

Watch

Extended reading notes

Core claim

The paper's central claim is that C3S3 sets a new state of the art for semi-supervised 3D medical image segmentation on both evaluated benchmarks. On the left atrium dataset, C3S3 reaches 91.24% Dice and 84.01% Jaccard with a 95HD of 5.14 voxels and ASD of 1.57 voxels, exceeding every compared semi-supervised method and nearly matching fully supervised VNet trained on all 80 volumes. On Pancreas-CT, it reports the best 95HD (6.96) and ASD (1.74) among semi-supervised methods while reaching the highest Dice (80.93%). The authors attribute the boundary gains to two designed modules: the Dynamic Complementary Competition module, which selects the currently better backbone to produce pseudo-labels that supervise the other, and the Outcome-Driven Contrastive Learning module, whose spatial binary masks extract positive and negative voxel pairs from the intersection and complement of the union of the two backbones' feature maps so that a dual-space contrastive loss pulls the backbones' representations into agreement at precisely those boundary-critical positions. Ablations show each module contributes independently: removing ODCL raises 95HD from 5.14 to 6.97, and removing DCC raises it to 6.27.

Load-bearing premise

The whole contrastive branch assumes that the intersection of the two backbones' binarized feature maps reliably marks true foreground and that voxels outside their union reliably mark true background; the paper never specifies the binarization threshold or checks the masks against ground truth, so with noisy predictions the positive and negative samples that drive the boundary gains would be mislabeled and the module could degrade performance instead of improving it.

Editorial extensions

If this is right

  • With only 20% of volumes labeled, C3S3's boundary metrics (95HD 5.14, ASD 1.57 on LA) are reported to be competitive with fully supervised training, which would let clinical segmentation models be built with far fewer manual annotations.
  • The at-least-6% improvement on 95HD and ASD across both MRI and CT domains suggests the boundary-refinement mechanism transfers across imaging modalities and organs.
  • The ablation results show each module is independently beneficial, and the paper claims ODCL and DCC can act as plug-and-play components for other frameworks.
  • The drop in 95HD from 6.97 to 5.14 when both modules are added indicates the method's main payoff is in reducing large boundary errors, the regime most relevant to diagnostic accuracy.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • I would expect the intersection/union masking to be most effective in early-to-mid training, when the two backbones' errors are decorrelated; late in training the masks converge to the consensus error and the contrastive signal weakens, so curriculum-style threshold or augmentation scheduling might extend the gains.
  • The same mask-driven contrastive scheme could transfer to other dense prediction tasks where boundaries are hard to annotate, such as instance boundaries in natural images, since nothing in Eqs. 7-10 is modality-specific; this is my extension, not the paper's claim.
  • The reported results compare against a fixed set of baselines under one 20% labeled split; a testable extension would be evaluating C3S3 under multiple label ratios (5%, 10%, 50%) to see whether the boundary advantage persists as supervision grows.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The paper proposes C3S3, a semi-supervised 3D medical image segmentation method combining two sub-networks (VNet and ResVNet) with a Dynamic Complementary Competition (DCC) module for pseudo-label generation and an Outcome-Driven Contrastive Learning (ODCL) module intended to refine boundary localization. On the LA (MRI) and Pancreas-CT benchmarks with 20% labeled data, C3S3 reports the best or second-best results among compared methods on most metrics, with particularly strong gains on the boundary-sensitive 95HD and ASD metrics. The authors also provide an ablation study showing that both DCC and ODCL contribute to the final performance, and they release code on GitHub.

Significance. If the reported results are reproducible, C3S3 would represent a useful advance in boundary-sensitive semi-supervised medical image segmentation, improving on recent methods such as TraCoCo and MCF on two standard benchmarks. The manuscript benefits from a clear motivation, a modular design that could be transferred to other segmentation frameworks, and the release of code. However, the main empirical claim is weakened by the absence of error bars or repeated-run statistics, by an ambiguity in the ODCL formulation (the binarization threshold is never specified), and by an abstract claim of 'at least 6%' improvement that is not supported by the Pancreas-CT numbers. These issues are local and fixable in a revision, so the work is not fatally compromised, but the current evidence is not yet sufficient to support the strongest conclusions.

major comments (5)
  1. [III-C, Eqs. (7)-(9)] The Spatial Position Binary Masking step never states the binarization threshold for the projected feature maps. Since these maps are real-valued, M∩ and M∪ are undefined until a threshold is chosen; different thresholds (e.g., 0, 0.5, mean activation) can make the intersection empty or the union cover the whole volume, which would eliminate one of the two contrastive terms in Eq. (9). The paper also reports no warm-up, fallback, or validation of mask quality. Because the entire ODCL contribution rests on the assumption that the intersection of binarized features is true foreground and the complement of the union is true background, the threshold must be specified and the mask quality checked (e.g., agreement with ground-truth foreground on labeled examples) before the Table IV gains can be attributed to correct contrastive learning.
  2. [III-C, Eqs. (7) and (8)] Equations (7) and (8), as written, use M∩ and (1−M∪) as scalar coefficients in the numerator and denominator without spatial indexing. For a voxel φ1 at position i, the masks should be evaluated at i and the negative sum should be restricted to the complement region; otherwise it is unclear whether negative voxels φn are drawn from inside the foreground, from the background, or from both, and how the masks weight them. This ambiguity prevents exact reproduction of Liuc1 and Liuc2 and should be clarified with explicit spatial indices or a precise sampling procedure.
  3. [Abstract and Tables I-II] The abstract states that 'at least 6%' improvement is achieved on the 95HD and ASD metrics, but this is not supported on Pancreas-CT. Relative to TraCoCo, the improvements are (7.34−6.96)/7.34 ≈ 5.18% for 95HD and (1.84−1.74)/1.84 ≈ 5.43% for ASD; only the LA numbers exceed 6%. Please correct the claim or report per-dataset percentages so that the reader can see exactly where the 6% threshold is and is not met.
  4. [IV, Tables I and II] All results are reported as point estimates without standard deviations, confidence intervals, or significance tests. Given that the largest claimed improvements in the ablation (Table IV, 95HD from 6.97 to 5.14) are comparable in magnitude to typical run-to-run variation in 3D medical segmentation, the paper should provide mean ± std over at least three random seeds and, if feasible, paired significance tests for the key comparisons against TraCoCo and MCF. Without this, the state-of-the-art claim is not robustly established.
  5. [V, Table III] The competition weight α is tuned on the LA dataset (best α = 0.8) and appears to be used for the reported LA and Pancreas results, but the paper provides no sensitivity analysis for Pancreas and does not account for the selection of α on the evaluation benchmark. The same applies to the unspecified λ in Eq. (3) and temperature τ in Eqs. (7)-(8). The reported rankings may therefore be partly a product of hyperparameter fitting rather than an out-of-sample comparison. Please state the values of α, λ, and τ used for each dataset, and show that the conclusions are stable across reasonable choices of these hyperparameters.
minor comments (5)
  1. [Abstract and Section I] The abstract uses 'immanent challenge' where 'imminent' or 'inherent' was likely intended, and Section I has 'two pop techniques' that should read 'two popular techniques'.
  2. [III-C] The text contains a typo: 'vise versa' should be 'vice versa'.
  3. [Tables I and II] Several numeric entries contain stray spaces (e.g., '1 .51', '91 .47', '84 .36'); these should be removed for readability.
  4. [References] Reference [14] is cited merely as 'ArXiv' with no publication venue or year; also, the CORN acronym is not expanded at first use in Section II-A.
  5. [Figure 1 caption] The caption 'where the blue areas represent inaccurately localized segments' is unclear; it should specify whether blue denotes over-segmentation, under-segmentation, or prediction error relative to the ground truth.

Circularity Check

1 steps flagged · score 2.0 of 10

No self-citations and no equation-level circularity; the LA boundary headline is, on the alpha axis, the argmax of a test-metric grid search (Table III) reported as an out-of-sample result in Table I, while the ODCL threshold omission and Eq.-7 zero-loss issue are correctness risks rather than circularity. Score 2.

  1. fitted input called prediction [Section V, Table III (alpha ablation) and Section IV-A, Table I (headline LA results); cf. Eq. (4)]
    "Analysis in the left atrial dataset shows that different values of the hyperparameter alpha have a relatively mild impact on model performance, with the best results achieved when alpha is set to 0.8, as shown in Table III. ... C3S3 achieves a 95HD of 5.14 ... For ASD, C3S3 achieves 1.57"

    Table III varies alpha over {1.0,...,0.0} on the LA dataset and reports the four target metrics for each value; the alpha=0.8 row is the grid maximum (95HD 5.14, ASD 1.57, Dice 91.24, Jaccard 84.01). Section IV-A then reports C3S3 with exactly these same four numbers as the headline comparison against competitors. The 'prediction' that C3S3 is best on LA boundary metrics is thus, on the alpha axis, the argmax of a search evaluated with the same metrics, dataset, and evaluation split, presented as an out-of-sample result rather than a parameter-free claim. The paper's own table shows the effect is not mild (95HD spans 5.14 to 7.14 across the grid), so the selection materially determines the headline figure.

full rationale

The paper has zero self-citations (references [1]-[24] are all external works), so the self-citation, imported-uniqueness, and ansatz-via-citation patterns do not apply; the VNet/ResVNet dual backbone is adopted 'in accordance with MCF [18]', an external CVPR 2023 paper. The evaluation is self-contained against external benchmarks: every headline metric is measured on held-out volumes against manual ground truth, and the ODCL ablation (Table IV) is checked against that same external ground truth, so the ODCL claim does not reduce to its self-generated masks by construction — the mask-based sampling is the standard semi-supervised self-training loop, and the reported gain (95HD 6.97 to 5.64) is an empirically measured outcome, not a logical identity. Flagged for the record, but as correctness/reproducibility risks rather than circularity: (i) Section III-C states 'we binarize the two feature maps' without ever specifying the threshold, and mask quality is never validated, so ODCL's positive/negative sampling is not independently reproducible; (ii) Eq. (7) as printed places M-cap as a coefficient on the positive term, so for any anchor with M-cap=1 the loss is -log(1)=0, meaning the printed Liuc1 is identically zero and cannot be the source of the reported gain; (iii) the abstract's 'at least 6%' improvement is arithmetically false on Pancreas-CT (vs. TraCoCo: 95HD (7.34-6.96)/7.34 = 5.2%; ASD (1.84-1.74)/1.84 = 5.4%). The one structural reduction approaching pattern 2 is the alpha selection: Table III searches alpha on the LA dataset using exactly the four metrics Table I then reports as C3S3's headline advantage, so on that axis the LA boundary 'prediction' is the selected maximum of the search. The central claim retains independent content because the comparison pits C3S3 against eight external baselines, and the LA-selected alpha=0.8 is carried over to a second dataset (Pancreas-CT) where C3S3 again leads on 95HD and ASD. Verdict: no equation-level or self-citation circularity; one mild in-sample-selection concern is weighted at score 2.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The method does not introduce physical entities. The module names (ODCL, DCC) are constructs, but they have no independent falsifiable handle outside the paper; they are evaluated only through the reported metrics.

free parameters (4)
  • alpha (competition weight) = 0.8 (best on LA)
    Weights CE and Dice terms in Lcompetition (Eq. 4); selected via ablation on LA dataset (Table III), directly influences pseudo-label selection and final metrics.
  • lambda (loss weight) = Not specified
    Balances Lcp and Liuc in Eq. 3; paper says setting is task-dependent but gives no value.
  • tau (contrastive temperature) = Not specified
    Temperature in contrastive losses (Eqs. 7-8); no value reported.
  • binarization threshold = Not specified
    Used to convert feature maps to binary masks M∩ and M∪ in Section III-C; not described, yet it defines positive and negative samples.
assumptions (4)
  • ad hoc to paper Intersection of binarized feature maps is correct foreground; voxels outside union are correct background.
    Invoked in Section III-C Spatial Position Binary Masking; the contrastive loss labels are derived from this assumption rather than from ground truth.
  • domain assumption The backbone with lower supervised competition loss produces reliable pseudo-labels for the other backbone.
    Assumed in Section III-B DCC module; standard co-training assumption but not independently validated here.
  • standard math Contrastive loss draws positive pairs together and pushes negatives apart, improving feature discriminability.
    Standard InfoNCE-style objective in Eq. 7-8; accepted background.
  • domain assumption Semi-supervised training with 20 percent labeled volumes is representative of clinical annotation scarcity for these datasets.
    Section IV uses 16/100 LA and 12/62 Pancreas labeled volumes; assumes this regime is meaningful.

how reviews work

0 comments
Cite this review

Pith. "Pith review of C3S3: Complementary Competition and Contrastive Selection for Semi-Supervised Medical Image Segmentation." pith.science (2026). https://pith.science/paper/YOTUOHHZ

@misc{pith2026250607368,
  author       = {Pith},
  title        = {Pith review of: C3S3: Complementary Competition and Contrastive Selection for Semi-Supervised Medical Image Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YOTUOHHZ}},
  note         = {Machine review of arXiv:2506.07368}
}
read the original abstract

For the immanent challenge of insufficiently annotated samples in the medical field, semi-supervised medical image segmentation (SSMIS) offers a promising solution. Despite achieving impressive results in delineating primary target areas, most current methodologies struggle to precisely capture the subtle details of boundaries. This deficiency often leads to significant diagnostic inaccuracies. To tackle this issue, we introduce C3S3, a novel semi-supervised segmentation model that synergistically integrates complementary competition and contrastive selection. This design significantly sharpens boundary delineation and enhances overall precision. Specifically, we develop an Outcome-Driven Contrastive Learning module dedicated to refining boundary localization. Additionally, we incorporate a Dynamic Complementary Competition module that leverages two high-performing sub-networks to generate pseudo-labels, thereby further improving segmentation quality. The proposed C3S3 undergoes rigorous validation on two publicly accessible datasets, encompassing the practices of both MRI and CT scans. The results demonstrate that our method achieves superior performance compared to previous cutting-edge competitors. Especially, on the 95HD and ASD metrics, our approach achieves a notable improvement of at least 6%, highlighting the significant advancements. The code is available at https://github.com/Y-TARL/C3S3.

Figures

Figures reproduced from arXiv: 2506.07368 by the authors.

Figure 1
Figure 1. The visual comparison among C3S3 and others, where the [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of our proposal. Upon receiving the input data, C3S3 undergoes two random data augmentations, followed by two separate [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Illustration of ODCL. ODCL identifies positive and negative [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Qualitative segmentation results. (a) Results on the LA dataset. (b) Results on the Pancreas-CT dataset. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

23 extracted references · 23 canonical work pages

  1. [1]

    Directional connectivity-based segmen- tation of medical images,

    Ziyun Yang and Sina Farsiu, “Directional connectivity-based segmen- tation of medical images,” in CVPR, 2023, pp. 11525–11535

    work page 2023

  2. [2]

    Magicnet: Semi- supervised multi-organ segmentation via magic-cube partition and re- covery,

    Duowen Chen, Yunhao Bai, Wei Shen, et al., “Magicnet: Semi- supervised multi-organ segmentation via magic-cube partition and re- covery,” in CVPR, 2022, pp. 23869–23878

    work page 2022

  3. [3]

    Bidirectional copy-paste for semi-supervised medical image segmentation,

    Yunhao Bai, Duowen Chen, Qingli Li, et al., “Bidirectional copy-paste for semi-supervised medical image segmentation,” in CVPR, 2023, pp. 11514–11524

    work page 2023

  4. [4]

    Co-training with high- confidence pseudo labels for semi-supervised medical image segmenta- tion,

    Zhiqiang Shen, Peng Cao, Hua Yang, et al., “Co-training with high- confidence pseudo labels for semi-supervised medical image segmenta- tion,” in IJCAI, 2023, pp. 4199–4207

    work page 2023

  5. [5]

    Transformation-consistent self-ensembling model for semi-supervised medical image segmenta- tion,

    Xiaomeng Li, Lequan Yu, Hao Chen, et al., “Transformation-consistent self-ensembling model for semi-supervised medical image segmenta- tion,” IEEE Transactions on Neural Networks and Learning Systems , pp. 523–534, 2020

    work page 2020

  6. [6]

    Competitive ensembling teacher-student framework for semi-supervised left atrium mri segmen- tation,

    Yuyan Shi, Yichi Zhang, and Shasha Wang, “Competitive ensembling teacher-student framework for semi-supervised left atrium mri segmen- tation,” BIBM, pp. 2231–2234, 2023

    work page 2023

  7. [7]

    Self-paced co-training and foundation model for semi-supervised medical image segmentation,

    Bo Qian, Yang Wen, and Bin Sheng, “Self-paced co-training and foundation model for semi-supervised medical image segmentation,” in ICME, 2024, pp. 1–6

    work page 2024

  8. [8]

    Caussl: Causality-inspired semi-supervised learning for medical image segmentation,

    Juzheng Miao, Cheng Chen, Furui Liu, et al., “Caussl: Causality-inspired semi-supervised learning for medical image segmentation,” in ICCV, Oct. 2023, pp. 21369–21380

    work page 2023

Show all 23 references
  1. [9]

    Pseudo-label guided contrastive learning for semi-supervised medical image segmentation,

    Hritam Basak and Zhaozheng Yin, “Pseudo-label guided contrastive learning for semi-supervised medical image segmentation,” in CVPR, 2023, pp. 19786–19797

    2023

  2. [10]

    C3t: Contrastive consistency cross-network learning for semi-supervised semantic seg- mentation,

    Yucheng Shu, Jiaxin Xie, Lihong Qiao, et al., “C3t: Contrastive consistency cross-network learning for semi-supervised semantic seg- mentation,” in ICME, 2024, pp. 1–6

    2024

  3. [12]

    Correlation-aware mutual learning for semi-supervised medical image segmentation,

    Shengbo Gao, Ziji Zhang, Jiechao Ma, et al., “Correlation-aware mutual learning for semi-supervised medical image segmentation,” in MICCAI, 2023, pp. 98–108

    2023

  4. [13]

    Translation Consistent Semi- supervised Segmentation for 3D Medical Images,

    Yuyuan Liu, Yu Tian, Chong Wang, et al., “Translation Consistent Semi- supervised Segmentation for 3D Medical Images,” IEEE Transactions on Medical Imaging , pp. 1–1, 2024

    2024

  5. [14]

    Leveraging coral- correlation consistency network for semi-supervised left atrium mri segmentation,

    Xinze Li, Runlin Huang, Zhenghao Wu, et al., “Leveraging coral- correlation consistency network for semi-supervised left atrium mri segmentation,” ArXiv, 2024

    2024

  6. [15]

    Icr-net: Semi- supervised medical image segmentation guided by intra-sample cross reconstruction,

    Xianpeng Cao, Weixing Xie, Xianxing Cao, et al., “Icr-net: Semi- supervised medical image segmentation guided by intra-sample cross reconstruction,” in ICME, 2024, pp. 1–6

    2024

  7. [16]

    Rcps: Rectified contrastive pseudo supervision for semi-supervised medical image seg- mentation,

    Xiangyu Zhao, Zengxin Qi, Sheng Wang, et al., “Rcps: Rectified contrastive pseudo supervision for semi-supervised medical image seg- mentation,” IEEE Journal of Biomedical and Health Informatics , pp. 251–261, 2023

    2023

  8. [17]

    Uncertainty global contrastive learning framework for semi-supervised medical image seg- mentation,

    Hengyang Liu, Pengcheng Ren, Yang Yuan, et al., “Uncertainty global contrastive learning framework for semi-supervised medical image seg- mentation,” IEEE Journal of Biomedical and Health Informatics , pp. 1–10, 2024

    2024

  9. [18]

    Mcf: Mutual correction framework for semi-supervised medical image segmentation,

    Yongchao Wang, Bin Xiao, Xiuli Bi, et al., “Mcf: Mutual correction framework for semi-supervised medical image segmentation,” in CVPR, 2023, pp. 15651–15660

    2023

  10. [19]

    A global benchmark of algorithms for segmenting late gadolinium-enhanced cardiac magnetic resonance imaging,

    Zhaohan Xiong, Qing Xia, Zhiqiang Hu, et al., “A global benchmark of algorithms for segmenting late gadolinium-enhanced cardiac magnetic resonance imaging,” Medical Image Analysis , p. 101832, 2020

    2020

  11. [20]

    Deeporgan: Multi-level deep convolutional networks for automated pancreas segmentation,

    Holger R. Roth, Le Lu, Amal A. Farag, et al., “Deeporgan: Multi-level deep convolutional networks for automated pancreas segmentation,” in MICCAI, 2015, p. 556–564

    2015

  12. [21]

    Uncertainty-aware self- ensembling model for semi-supervised 3d left atrium segmentation,

    Lequan Yu, Shujun Wang, Xiaomeng Li, et al., “Uncertainty-aware self- ensembling model for semi-supervised 3d left atrium segmentation,” in MICCAI, 2019, pp. 605–613

    2019

  13. [22]

    Shape-aware semi- supervised 3d semantic segmentation for medical images,

    Shuailin Li, Chuyu Zhang, and Xuming He, “Shape-aware semi- supervised 3d semantic segmentation for medical images,” in MICCAI, 2020, pp. 552–561

    2020

  14. [23]

    Semi-supervised medical image segmentation through dual-task consistency,

    Xiangde Luo, Jieneng Chen, Tao Song, et al., “Semi-supervised medical image segmentation through dual-task consistency,” in AAAI, 2021, pp. 8801–8809

    2021

  15. [24]

    Exploring smoothness and class-separation for semi-supervised medical image segmentation,

    Yicheng Wu, Zhonghua Wu, Qianyi Wu, et al., “Exploring smoothness and class-separation for semi-supervised medical image segmentation,” in MICCAI, 2022, pp. 34–43

    2022

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.