Pith. sign in

REVIEW 4 major objections 6 minor 58 references

SSS: Semi-Supervised SAM-2 with Efficient Prompting for Medical Imaging Segmentation

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A SAM-2-based semi-supervised framework, SSS, reports state-of-the-art Dice scores of 53.15 on BHSD and 89.34/91.21 on ACDC.

desk verdict The DFE module is a no-op as written (Eq. (15) subtracts the cosine similarity of a vector with itself), so the headline gains are unsupported; the PCSW prompt generator is the one piece that might survive. read the letter →

arxiv 2506.08949 v1 pith:ZNOFKZFW submitted 2025-06-10 cs.CV

Hongjie Zhu , Xiwei Liu , Rundong Xue , Zeyu Zhang , Yong Xu , Daji Ergu , Ying Cai , Yang Zhao This is my paper
classification cs.CV
keywords semi-supervisedlearningmedicalimagesegmentationSAM-2foundationmodelsconsistencyregularizationpromptgenerationbrainhemorrhagecardiacMRI
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

The paper proposes SSS, a semi-supervised medical image segmentation framework built on the SAM-2 visual foundation model, and claims it extracts useful signal from unlabeled scans while using very few pixel-level labels. The method extends a single-stream weak-to-strong consistency-regularization scheme with a Discriminative Feature Enhancement (DFE) module intended to exploit feature differences across augmentation views, and with a Physical Constraints with Sliding Window (PCSW) prompt generator that turns encoder features into reliable prompts for SAM-2. On the BHSD brain hemorrhage CT dataset, SSS reports a Dice score of 53.15 with 500 unlabeled volumes, beating the previous best semi-supervised method by +3.65 Dice; on ACDC cardiac MRI it reports 89.34 Dice with 5% labeled data and 91.21 Dice with 10%. A sympathetic reader would care because this is a step toward making foundation-model segmentation practical in clinical settings where annotations are scarce.

What carries the argument

The two load-bearing components are the Discriminative Feature Enhancement (DFE) mechanism and the Physical Constraints with a Sliding Window (PCSW) prompt generator. DFE is intended to compute, at each feature scale, the cosine similarity between globally pooled feature vectors from two strongly augmented views, fuse them via a sigmoid-weighted blend with MLP projections, and add a residual correction; this is supposed to amplify discriminative regions before pseudo-mask generation. PCSW generates pseudo-masks from multi-scale encoder features via convolution and softmax, then slides a window over consecutive slices of a 3D volume, applies depth-first search with eight-connectivity, and accepts the window only if the largest connected component covers at least a threshold fraction of the foreground, thereby producing prompts for SAM-2. The framework wraps these around a single-stream weak-to-strong consistency-regularization training loop with complementary dropout and an EMA teacher.

What would settle it

Compute the DFE residual in a forward pass: since $S_{\text{Cosine}}(F_{s2}^{(i)}, F_{s2}^{(i)}) = 1$ for any feature map, $\Delta F^{(i)} = 0$ always, so the module cannot alter the features; consequently the ablation's jump from 48.43 to 49.17 Dice cannot be produced by the mechanism as written, and a corrected implementation or re-measured ablation would be needed to confirm the reported gain.

Watch

Extended reading notes

Core claim

The central claim is that combining a SAM-2/MedSAM-2 backbone with a weak-to-strong consistency loss and automatically generated prompts achieves state-of-the-art semi-supervised medical image segmentation. Specifically, the paper reports 53.15 average Dice on BHSD, surpassing Cross Pseudo Supervision with nnU-Net at 49.50 by +3.65, and 89.34/91.21 Dice on ACDC under 5% and 10% labeled data. The intended mechanism is that the DFE module reconstructs features from multi-scale augmented views to sharpen salient regions, while the PCSW generator uses slice-to-slice continuity and connectivity analysis to select trustworthy pseudo-mask slices as input prompts, satisfying SAM-2's need for prompts on unlabeled data.

Load-bearing premise

The claimed DFE gain rests on the premise that the residual feature adjustment actually changes the features; as written, the residual is identically zero because it is multiplied by the cosine similarity of a feature vector with itself.

Editorial extensions

If this is right

  • If SSS is right, semi-supervised medical segmentation can ride on a foundation-model backbone rather than training a network from scratch, giving a practical route to high Dice with 5–10% labels.
  • The PCSW prompt generator provides an annotation-free way to satisfy SAM-2's prompt requirement, so the approach extends to other prompt-hungry foundation models.
  • On BHSD, the reported +3.65 Dice over the best prior semi-supervised method makes the 500-unlabeled-volume setting a new reference point for future comparisons.
  • Because the method is built on UniMatch V2's single-stream weak-to-strong framework, its components can be dropped into other fully-supervised segmenters, not only SAM-2.

Reading between the lines

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

  • An implication left implicit is that PCSW, not DFE, likely drives the measured gain: the ablation shows baseline 48.43, +DFE 49.17, +PCSW 51.26, so the prompt generator accounts for the larger jump.
  • The PCSW connectivity prior assumes lesions are spatially continuous across adjacent slices; on datasets with small, dispersed, or slice-independent findings, the sliding-window threshold would need re-tuning or would suppress true positives.
  • The large gap between binary SSS (53.15) and multi-class SSS* (44.50) on BHSD suggests that pooling all hemorrhage types into one foreground class boosts the score, so comparisons against other methods should keep the class definition in mind.
  • A natural testable extension is to replace the self-cosine term in DFE with a cross-view cosine, e.g., $S(F_{s1}, F_{s2})$; if the reported gains persist, the mechanism's value would be verified, otherwise PCSW is the primary driver.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes SSS, a semi-supervised medical image segmentation framework built on the SAM-2/MedSAM-2 backbone. It combines a weak-to-strong consistency regularization scheme in the style of UniMatch V2 with a Discriminative Feature Enhancement (DFE) mechanism and a Physical Constraints with Sliding Window (PCSW) prompt generator. Experiments on ACDC and BHSD report state-of-the-art Dice scores, and an ablation study attributes gains to DFE and PCSW.

Significance. If the reported results were fully supported, SSS would be a useful demonstration that a SAM-2 backbone can be integrated into semi-supervised medical segmentation, and PCSW is a plausible mechanism for generating prompts from unlabeled data. The paper contains a clear ablation design and a parameter study. However, the DFE mechanism as written is inert because of an error in Eq. (15), so the first contribution and the associated ablation gains are not currently supported. The PCSW contribution itself (48.43 to 51.26 Dice) is not nullified, but the headline state-of-the-art claim depends on both modules and therefore needs correction and re-evaluation.

major comments (4)
  1. [Section 3.3, Eq. (15)] The residual term is defined as ΔF(i) = (1 − S_Cosine(F_s2, F_s2)) · (F_s2 − F_s1). Since the cosine similarity of a vector with itself is exactly 1, the prefactor vanishes identically for every scale i, so ΔF(i) = 0 and Eq. (16) reduces to F_adjusted = F_fused. The DFE module therefore has no effect as written, which contradicts the +0.74 Dice gain in Table 3 (48.43 → 49.17) and the ACDC improvements in Table 4. Please correct the formula, presumably to use S_Cosine(F_s1, F_s2) if the intended residual is a similarity-weighted difference, rerun the affected experiments, and report whether the DFE gains persist. If the corrected DFE cannot reproduce the reported gains, the DFE contribution and its related claims should be removed or substantially revised.
  2. [Section 4.3, Tables 1 and 2] The main comparisons are not backbone-matched. The ACDC baselines use U-Net or nnU-Net backbones while SSS uses MedSAM-2/SAM-2, and the BHSD comparisons are all nnU-Net based. Without a same-backbone baseline, such as a supervised-only or UniMatch V2 run with the MedSAM-2 backbone under identical settings, the reported gains cannot be attributed to the proposed modules rather than to the change of backbone. Please add matched-backbone baselines, or at least an ablation that isolates the effect of the backbone.
  3. [Section 4.4, Table 4 and Fig. 3] The ACDC ablation results are presented without error bars or significance tests, so small per-class differences such as 89.34 versus 89.22 Dice for RV are not interpretable. In addition, the connectivity threshold τ appears to be selected based on the test-set Dice values in Fig. 3; the authors should specify a validation-based selection procedure to avoid test-set overfitting. Please report means and standard deviations over multiple seeds and clarify how τ was chosen.
  4. [Section 4.3, Table 2] The paper defines Jaccard, 95HD, and ASD in Section 4.1, but Table 2 reports only Dice for BHSD. Moreover, the row SSS* (= 44.50 ± 0.17) is presented in the BHSD table while the text in the ACDC paragraph attributes a Dice of 44.50 to multi-class segmentation under 'the same settings.' This is ambiguous. Please clarify what SSS* is, which dataset and setting it corresponds to, and why the BHSD evaluation does not report the other metrics defined in the protocol.
minor comments (6)
  1. [Section 4.3 and Table 1] The labeled-data ratio for ACDC is described as '3%' in the text and '5% / 3 labeled data' in the Table 1 header; the result paragraph also says '3% labeled data setting.' Please harmonize these numbers.
  2. [Equation (15)] The displayed equation contains a stray vertical bar after 'F_s2 − F_s1', which appears to be a typesetting error and should be removed.
  3. [Figure 1] The caption contains corrupted characters such as 'čaĎ', 'čbĎ', and 'čcĎ', which should be fixed.
  4. [Section 3.3] The text says x_s1 and x_s2 are 'subjected to weak and strong augmentations respectively,' but Eq. (5) applies weak augmentation to produce x_w and strong augmentation to produce x_s1 and x_s2. Please correct the wording.
  5. [Table 4] The ACDC ablation does not include a PCSW-only condition, although Table 3 does; adding it would make the component analysis consistent across the two datasets.
  6. [Section 4.4 and Fig. 3] The sentence 'Based on the experimental results, setting τ achieves a favorable balance' does not state the chosen threshold value. Please give the exact value used in the main experiments.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central claims are empirical benchmark results on held-out test sets; the DFE residual in Eq. (15) is identically zero as written, but that is an internal mathematical inconsistency, not a circular reduction.

full rationale

Score 0: no circularity. The paper's central claims are empirical Dice/Jaccard/95HD/ASD results on held-out ACDC and BHSD test sets, compared against published baselines. The self-training loop (weak-to-strong pseudo-labels in Eqs. (6)-(7), and PCSW prompts derived from the model's own encoder features in Eqs. (17)-(18)) is the intended SSL mechanism, not a disguised restatement of the output; the external test evaluation breaks any derivation-level circularity. There is no fitted parameter renamed as a prediction: hyperparameters such as tau=0.8 and EMA momentum are standard tuning choices. Self-citations (BHSD dataset [36], MedSAM-2 [58], and assorted prior works) are not load-bearing; they provide the benchmark and backbone, whose validity is independent of the present results. Major caveat, not a circularity: Eq. (15) defines DeltaF(i) = (1 - S_Cosine(F_s2, F_s2))*(F_s2 - F_s1), and since cosine similarity with itself is identically 1, DeltaF(i)=0 and Eq. (16) collapses to Eq. (14). The reported DFE ablation gain (48.43 to 49.17 Dice in Table 3) is therefore unsupported by the stated formula; this is an internal mathematical inconsistency or typo, not a circular derivation. The main SSS result also relies on PCSW, so the benchmark claim is not fully reduced to the inert DFE term, but the paper needs a corrected Eq. (15) or a removal of the DFE claim.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The central claim rests on standard SSL consistency regularization, a spatial-continuity assumption about lesions, transferability of SAM-2 to medical imaging, and a self-referential prompt loop. The DFE mechanism contains a mathematical error in Eq. (15) that nullifies its residual term. The free parameters tau and the window bounds are tuned on validation data.

free parameters (2)
  • Connectivity threshold tau = The chosen value is 0.8.
    Selected by the validation sweep in Fig. 3 over 0.1 to 0.9; it controls pseudo-mask acceptance in Eq. (21) and affects the reported Dice.
  • Sliding-window length bounds alpha and beta = Set to one-third and one-half of the slice count.
    Chosen ad hoc in Section 3.4 to define the dynamic window size N; no derivation or ablation for these bounds.
assumptions (5)
  • domain assumption Weak-to-strong consistency regularization with pseudo-labels from weakly augmented views improves segmentation on unlabeled medical images.
    Borrowed from UniMatch V2 [44] and used as the base framework in Section 3.2; it is a standard SSL assumption not re-validated for the SAM-2 backbone.
  • domain assumption Lesion regions in 3D volumetric medical images are spatially continuous across adjacent slices, so connectivity filtering in sliding windows selects reliable pseudo-masks.
    Stated in Section 3.4 as the physical constraint that motivates PCSW.
  • domain assumption SAM-2 and MedSAM-2 features pre-trained on SA-1B transfer usefully to medical image segmentation after fine-tuning.
    Relies on MedSAM-2 [58] and SAM-2 [28] as the backbone in all experiments.
  • ad hoc to paper The model's own pseudo-masks, after PCSW filtering, are reliable enough to serve as prompts for SAM-2.
    Self-referential, since prompt generation in Section 3.4 uses predictions from the same encoder with no independent source for prompt ground truth.
  • standard math Cosine similarity of a feature vector with itself equals 1, which makes the residual term in Eq. (15) exactly zero.
    This standard identity is true and exposes the derivation error in the DFE residual calculation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SSS: Semi-Supervised SAM-2 with Efficient Prompting for Medical Imaging Segmentation." pith.science (2026). https://pith.science/paper/ZNOFKZFW

@misc{pith2026250608949,
  author       = {Pith},
  title        = {Pith review of: SSS: Semi-Supervised SAM-2 with Efficient Prompting for Medical Imaging Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZNOFKZFW}},
  note         = {Machine review of arXiv:2506.08949}
}
read the original abstract

In the era of information explosion, efficiently leveraging large-scale unlabeled data while minimizing the reliance on high-quality pixel-level annotations remains a critical challenge in the field of medical imaging. Semi-supervised learning (SSL) enhances the utilization of unlabeled data by facilitating knowledge transfer, significantly improving the performance of fully supervised models and emerging as a highly promising research direction in medical image analysis. Inspired by the ability of Vision Foundation Models (e.g., SAM-2) to provide rich prior knowledge, we propose SSS (Semi-Supervised SAM-2), a novel approach that leverages SAM-2's robust feature extraction capabilities to uncover latent knowledge in unlabeled medical images, thus effectively enhancing feature support for fully supervised medical image segmentation. Specifically, building upon the single-stream "weak-to-strong" consistency regularization framework, this paper introduces a Discriminative Feature Enhancement (DFE) mechanism to further explore the feature discrepancies introduced by various data augmentation strategies across multiple views. By leveraging feature similarity and dissimilarity across multi-scale augmentation techniques, the method reconstructs and models the features, thereby effectively optimizing the salient regions. Furthermore, a prompt generator is developed that integrates Physical Constraints with a Sliding Window (PCSW) mechanism to generate input prompts for unlabeled data, fulfilling SAM-2's requirement for additional prompts. Extensive experiments demonstrate the superiority of the proposed method for semi-supervised medical image segmentation on two multi-label datasets, i.e., ACDC and BHSD. Notably, SSS achieves an average Dice score of 53.15 on BHSD, surpassing the previous state-of-the-art method by +3.65 Dice. Code will be available at https://github.com/AIGeeksGroup/SSS.

Figures

Figures reproduced from arXiv: 2506.08949 by the authors.

Figure 1
Figure 1. Illustrations (a), (b), and (c) respectively depict existing semi-supervised learn￾ing methods employing consistency regularization with increasing strength, correspond￾ing to FixMatch [30], UniMatch V1 [7], and UniMatch V2 [44]. And (d) illustrates the proposed method, which incorporates the Discriminative Feature Enhancement (DFE) mechanism into UniMatch V2 to exploit feature information from multiple perspec￾tive… view at source ↗
Figure 2
Figure 2. An overview of the proposed SSS framework. SSS first applies strong and weak augmentation pools to the original image to generate corresponding augmented ver￾sions. These three types of input data are then processed by the SAM-2 image encoder. The features extracted from the strongly and weakly augmented images are passed to the Discriminative Feature Enhancement (DFE) mechanism to capture feature discrep￾ancies int… view at source ↗
Figure 3
Figure 3. Impact of varying value of Connectivity Threshold τ on the Dice. 4.5 Qualitative Analysis [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Visual comparison of input images generated using different methods. References 1. Bai, Y., Chen, D., Li, Q., Shen, W., Wang, Y.: Bidirectional copy-paste for semi￾supervised medical image segmentation. In: Proceedings of the IEEE/CVF confer￾ence on computer vision and…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

58 extracted references · 35 canonical work pages

  1. [1]

    In: Proceedings of the IEEE/CVF confer- ence on computer vision and pattern recognition

    Bai, Y., Chen, D., Li, Q., Shen, W., Wang, Y.: Bidirectional copy-paste for semi- supervised medical image segmentation. In: Proceedings of the IEEE/CVF confer- ence on computer vision and pattern recognition. pp. 11514–11524 (2023)

    2023

  2. [2]

    Bernard, O., Lalande, A., Zotti, C., Cervenansky, F., Yang, X., Heng, P.A., Cetin, I., Lekadir, K., Camara, O., Ballester, M.A.G., et al.: Deep learning techniques for automatic mri cardiac multi-structures segmentation and diagnosis: is the problem solved? IEEE transactions on medical imaging37(11), 2514–2525 (2018)

    2018

  3. [3]

    Medical Artificial Intelligence for Early Detection of Lung Cancer: A Survey

    Cai, G., Cai, Y., Zhang, Z., Cao, Y., Wu, L., Ergu, D., Liao, Z., Zhao, Y.: Medical ai for early detection of lung cancer: A survey. arXiv preprint arXiv:2410.14769 (2024)

    work page Pith review arXiv 2024

  4. [4]

    arXiv preprint arXiv:2409.14028 (2024)

    Cai, G., Zhang, R., He, H., Zhang, Z., Ergu, D., Cao, Y., Zhao, J., Hu, B., Liao, Z., Zhao, Y., et al.: Msdet: Receptive field enhanced multiscale detection for tiny pulmonary nodule. arXiv preprint arXiv:2409.14028 (2024)

    arXiv 2024

  5. [5]

    Medical Image Analysis98, 103310 (2024)

    Chen, C., Miao, J., Wu, D., Zhong, A., Yan, Z., Kim, S., Hu, J., Liu, Z., Sun, L., Li, X., et al.: Ma-sam: Modality-agnostic sam adaptation for 3d medical image segmentation. Medical Image Analysis98, 103310 (2024)

    2024

  6. [6]

    In: International conference on medical image com- puting and computer-assisted intervention

    Chen, F., Fei, J., Chen, Y., Huang, C.: Decoupled consistency for semi-supervised medical image segmentation. In: International conference on medical image com- puting and computer-assisted intervention. pp. 551–561. Springer (2023)

    work page 2023

  7. [7]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Chen, X., Yuan, Y., Zeng, G., Wang, J.: Semi-supervised semantic segmentation with cross pseudo supervision. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 2613–2622 (2021)

    2021

  8. [8]

    In: Proceedings SSS: Semi-Supervised SAM-2 17 of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Cheng, Z., Wei, Q., Zhu, H., Wang, Y., Qu, L., Shao, W., Zhou, Y.: Unleashing the potential of sam for medical adaptation via hierarchical decoding. In: Proceedings SSS: Semi-Supervised SAM-2 17 of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 3511–3522 (2024)

    work page 2024

Show all 58 references
  1. [9]

    arXiv preprint arXiv:2504.09203 (2025)

    Dutta, S., Vasim, A., Gole, S., Rezatofighi, H., Banerjee, B.: Aeroseg: Harnessing sam for open-vocabulary segmentation in remote sensing images. arXiv preprint arXiv:2504.09203 (2025)

    2025 arXiv

  2. [10]

    In: European Conference on Computer Vision

    Gao, N., Zhou, S., Wang, L., Zheng, N.: Pmt: Progressive mean teacher via ex- ploring temporal consistency for semi-supervised medical image segmentation. In: European Conference on Computer Vision. pp. 144–160. Springer (2024)

    2024

  3. [11]

    arXiv preprint arXiv:2408.13491 (2024)

    Ge, J., Zhang, Z., Phan, M.H., Zhang, B., Liu, A., Zhao, Y.: Esa: Annotation-efficient active learning for semantic segmentation. arXiv preprint arXiv:2408.13491 (2024)

    2024 arXiv

  4. [12]

    Hiwase, A.D., Ovenden, C.D., Kaukas, L.M., Finnis, M., Zhang, Z., O’Connor, S., Foo, N., Reddi, B., Wells, A.J., Ellis, D.Y.: Can rotational thromboelastometry rapidly identify theragnostic targets in isolated traumatic brain injury? Emergency Medicine Australasia37(1), e14480 (2025)

    2025

  5. [13]

    In: MIC- CAI Challenge on Fast and Low-Resource Semi-supervised Abdominal Organ Seg- mentation, pp

    Huang, Y., Zhang, H., Yan, Y., Hassan, H.: 3d cross-pseudo supervision (3d-cps): A semi-supervised nnu-net architecture for abdominal organ segmentation. In: MIC- CAI Challenge on Fast and Low-Resource Semi-supervised Abdominal Organ Seg- mentation, pp. 87–100. Springer (2022)

    2022

  6. [14]

    arXiv e-prints pp

    Ji, Y., Saratchandran, H., Gordon, C., Zhang, Z., Lucey, S.: Sine activated low-rank matrices for parameter efficient learning. arXiv e-prints pp. arXiv–2403 (2024)

    2024

  7. [15]

    Computers in Biology and Medicine169, 107840 (2024)

    Jiao, R., Zhang, Y., Ding, L., Xue, B., Zhang, J., Cai, R., Jin, C.: Learning with limited annotations: a survey on deep semi-supervised learning for medical image segmentation. Computers in Biology and Medicine169, 107840 (2024)

    2024

  8. [16]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Kirillov, A., Mintun, E., Ravi, N., Mao, H., Rolland, C., Gustafson, L., Xiao, T., Whitehead, S., Berg, A.C., Lo, W.Y., et al.: Segment anything. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 4015–4026 (2023)

    2023

  9. [17]

    arXiv preprint arXiv:2503.04639 (2025)

    Konwer, A., Yang, Z., Bas, E., Xiao, C., Prasanna, P., Bhatia, P., Kass-Hout, T.: Enhancing sam with efficient prompting and preference optimization for semi- supervised medical image segmentation. arXiv preprint arXiv:2503.04639 (2025)

    2025 arXiv

  10. [18]

    In: Medical Image Computing and Computer Assisted Intervention–MICCAI 2020: 23rd International Conference, Lima, Peru, October 4–8, 2020, Proceedings, Part I 23

    Li, S., Zhang, C., He, X.: Shape-aware semi-supervised 3d semantic segmenta- tion for medical images. In: Medical Image Computing and Computer Assisted Intervention–MICCAI 2020: 23rd International Conference, Lima, Peru, October 4–8, 2020, Proceedings, Part I 23. pp. 552–561. ...

    2020

  11. [19]

    Medical image analysis42, 60–88 (2017)

    Litjens, G., Kooi, T., Bejnordi, B.E., Setio, A.A.A., Ciompi, F., Ghafoorian, M., Van Der Laak, J.A., Van Ginneken, B., Sánchez, C.I.: A survey on deep learning in medical image analysis. Medical image analysis42, 60–88 (2017)

    2017

  12. [20]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Liu, Y., Tian, Y., Chen, Y., Liu, F., Belagiannis, V., Carneiro, G.: Perturbed and strict mean teachers for semi-supervised semantic segmentation. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 4258– 4267 (2022)

    2022

  13. [21]

    Measurement p

    Luo, S., Zhang, Y., Zhang, Z., Guo, B., Lian, J.J., Jiang, H., Zou, S., Wang, W.: Epdd-yolo: An efficient benchmark for pavement damage detection based on mamba-yolo. Measurement p. 117638 (2025)

    2025

  14. [22]

    Luo, X., Liao, W., Chen, J., Song, T., Chen, Y., Zhang, S., Chen, N., Wang, G., Zhang, S.: Efficient semi-supervised gross target volume of nasopharyngeal carcinoma segmentation via uncertainty rectified pyramid consistency. In: Med- ical Image Computing and Computer Assisted ...

    2021

  15. [23]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Miao, J., Chen, C., Liu, F., Wei, H., Heng, P.A.: Caussl: Causality-inspired semi-supervised learning for medical image segmentation. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 21426–21437 (2023)

    2023

  16. [24]

    Nature Machine Intelligence5(7), 724–738 (2023)

    Peiris, H., Hayat, M., Chen, Z., Egan, G., Harandi, M.: Uncertainty-guided dual- views for semi-supervised volumetric medical image segmentation. Nature Machine Intelligence5(7), 724–738 (2023)

    2023

  17. [25]

    arXiv preprint arXiv:2504.18983 (2025)

    Qi, X., Zhang, Z., Gang, C., Zhang, H., Zhang, L., Zhang, Z., Zhao, Y.: Mediaug: Exploring visual augmentation in medical imaging. arXiv preprint arXiv:2504.18983 (2025)

    2025 arXiv

  18. [26]

    arXiv preprint arXiv:2501.01392 (2025)

    Qi, X., Zhang, Z., Handoko, A.B., Zheng, H., Chen, M., Huy, T.D., Phan, V.M.H., Zhang, L., Cheng, L., Jiang, S., et al.: Projectedex: Enhancing generation in ex- plainable ai for prostate cancer. arXiv preprint arXiv:2501.01392 (2025)

    2025 arXiv

  19. [27]

    arXiv preprint arXiv:2502.00631 (2025)

    Qi, X., Zhang, Z., Zheng, H., Chen, M., Kutaiba, N., Lim, R., Chiang, C., Tham, Z.E., Ren, X., Zhang, W., et al.: Medconv: Convolutions beat transformers on long-tailed bone density prediction. arXiv preprint arXiv:2502.00631 (2025)

    2025 arXiv

  20. [28]

    arXiv preprint arXiv:2408.00714 (2024)

    Ravi, N., Gabeur, V., Hu, Y.T., Hu, R., Ryali, C., Ma, T., Khedr, H., Rädle, R., Rolland, C., Gustafson, L., et al.: Sam 2: Segment anything in images and videos. arXiv preprint arXiv:2408.00714 (2024)

    2024 arXiv

  21. [29]

    In: Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, Oc- tober 5-9, 2015, proceedings, part III 18

    Ronneberger, O., Fischer, P., Brox, T.: U-net: Convolutional networks for biomed- ical image segmentation. In: Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, Oc- tober 5-9, 2015, proceedings, part III 18....

    2015

  22. [30]

    Advances in neural information processing systems33, 596–608 (2020)

    Sohn, K., Berthelot, D., Carlini, N., Zhang, Z., Zhang, H., Raffel, C.A., Cubuk, E.D., Kurakin, A., Li, C.L.: Fixmatch: Simplifying semi-supervised learning with consistency and confidence. Advances in neural information processing systems33, 596–608 (2020)

    2020

  23. [31]

    In: International Conference on Med- ical Image Computing and Computer-Assisted Intervention

    Song, B., Wang, Q.: Sdcl: Students discrepancy-informed correction learning for semi-supervised medical image segmentation. In: International Conference on Med- ical Image Computing and Computer-Assisted Intervention. pp. 567–577. Springer (2024)

    2024

  24. [32]

    arXiv preprint arXiv:2412.19990 (2024)

    Tan,S.,Xue,R.,Luo,S.,Zhang,Z.,Wang,X.,Zhang,L.,Ergu,D.,Yi,Z.,Zhao,Y., Cai, Y.: Segkan: High-resolution medical image segmentation with long-distance dependencies. arXiv preprint arXiv:2412.19990 (2024)

    2024 arXiv

  25. [33]

    arXiv preprint arXiv:2408.00496 (2024)

    Tan, S., Zhang, Z., Cai, Y., Ergu, D., Wu, L., Hu, B., Yu, P., Zhao, Y.: Segstitch: Multidimensional transformer for robust and efficient medical imaging segmenta- tion. arXiv preprint arXiv:2408.00496 (2024)

    2024 arXiv

  26. [34]

    Advances in neural information processing systems30(2017)

    Tarvainen, A., Valpola, H.: Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results. Advances in neural information processing systems30(2017)

    2017

  27. [35]

    Advances in Neural Information Processing Systems36, 8815–8827 (2023)

    Wang, D., Zhang, J., Du, B., Xu, M., Liu, L., Tao, D., Zhang, L.: Samrs: Scaling-up remote sensing segmentation dataset with segment anything model. Advances in Neural Information Processing Systems36, 8815–8827 (2023)

    2023

  28. [36]

    In: Interna- tional Workshop on Machine Learning in Medical Imaging

    Wu, B., Xie, Y., Zhang, Z., Ge, J., Yaxley, K., Bahadir, S., Wu, Q., Liu, Y., To, M.S.: Bhsd: A 3d multi-class brain hemorrhage segmentation dataset. In: Interna- tional Workshop on Machine Learning in Medical Imaging. pp. 147–156. Springer (2023)

    2023

  29. [37]

    arXiv preprint arXiv:2407.19546 (2024) SSS: Semi-Supervised SAM-2 19

    Wu, B., Xie, Y., Zhang, Z., Phan, M.H., Chen, Q., Chen, L., Wu, Q.: Xlip: Cross- modal attention masked modelling for medical language-image pre-training. arXiv preprint arXiv:2407.19546 (2024) SSS: Semi-Supervised SAM-2 19

    2024 arXiv

  30. [38]

    arXiv preprint arXiv:2304.12620 (2023)

    Wu, J., Ji, W., Liu, Y., Fu, H., Xu, M., Xu, Y., Jin, Y.: Medical sam adapter: Adapting segment anything model for medical image segmentation. arXiv preprint arXiv:2304.12620 (2023)

    2023 arXiv

  31. [39]

    Medical image analysis102, 103547 (2025)

    Wu, J., Wang, Z., Hong, M., Ji, W., Fu, H., Xu, Y., Xu, M., Jin, Y.: Medical sam adapter: Adapting segment anything model for medical image segmentation. Medical image analysis102, 103547 (2025)

    2025

  32. [40]

    In: International conference on medical image computing and computer-assisted intervention

    Wu, Y., Wu, Z., Wu, Q., Ge, Z., Cai, J.: Exploring smoothness and class-separation for semi-supervised medical image segmentation. In: International conference on medical image computing and computer-assisted intervention. pp. 34–43. Springer (2022)

    2022

  33. [41]

    Wu, Y., Xu, M., Ge, Z., Cai, J., Zhang, L.: Semi-supervised left atrium segmen- tation with mutual consistency training. In: Medical image computing and com- puter assisted intervention–MICCAI 2021: 24th international conference, Stras- bourg, France, September 27–October 1, 2...

    2021

  34. [42]

    In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision

    Xia, Y., Liu, F., Yang, D., Cai, J., Yu, L., Zhu, Z., Xu, D., Yuille, A., Roth, H.: 3d semi-supervised learning with uncertainty-aware multi-view co-training. In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. pp. 3646–3655 (2020)

    2020

  35. [43]

    In: Proceedings of the AAAI Conference on Artificial Intel- ligence

    Xing, W., Cheng, Y., Yi, H., Gao, X., Wei, X., Guo, X., Zhang, Y., Pang, X.: Lcgc: Learning from consistency gradient conflicting for class-imbalanced semi- supervised debiasing. In: Proceedings of the AAAI Conference on Artificial Intel- ligence. vol. 39, pp. 21697–21706 (2025)

    2025

  36. [44]

    IEEE Transactions on Pattern Analysis and Machine In- telligence (2025)

    Yang, L., Zhao, Z., Zhao, H.: Unimatch v2: Pushing the limit of semi-supervised semantic segmentation. IEEE Transactions on Pattern Analysis and Machine In- telligence (2025)

    2025

  37. [45]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Yang, L., Zhuo, W., Qi, L., Shi, Y., Gao, Y.: St++: Make self-training work bet- ter for semi-supervised semantic segmentation. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 4268–4277 (2022)

    2022

  38. [46]

    arXiv preprint arXiv:2501.12844 (2025)

    Zhang, R., Guo, H., Zhang, Z., Yan, P., Zhao, S.: Gamed-snake: Gradient-aware adaptive momentum evolution deep snake model for multi-organ segmentation. arXiv preprint arXiv:2501.12844 (2025)

    2025 arXiv

  39. [47]

    In: 2024 IEEE International Conference on Bioinformatics and Biomedicine (BIBM)

    Zhang, Y., Yang, J., Liu, Y., Cheng, Y., Qi, Y.: Semisam: Enhancing semi- supervised medical image segmentation via sam-assisted consistency regulariza- tion. In: 2024 IEEE International Conference on Bioinformatics and Biomedicine (BIBM). pp. 3982–3986. IEEE (2024)

    2024

  40. [48]

    Zhang, Z., Ahmed, K.A., Hasan, M.R., Gedeon, T., Hossain, M.Z.: A deep learning approachtodiabetesdiagnosis.In:AsianConferenceonIntelligentInformationand Database Systems. pp. 87–99. Springer (2024)

    2024

  41. [49]

    In: Annual Conference on Medical Image Understanding and Analysis

    Zhang, Z., Qi, X., Chen, M., Li, G., Pham, R., Qassim, A., Berry, E., Liao, Z., Siggs, O., Mclaughlin, R., et al.: Jointvit: Modeling oxygen saturation levels with joint supervision on long-tailed octa. In: Annual Conference on Medical Image Understanding and Analysis. pp. 158...

    2024

  42. [50]

    In: 2024 IEEE International Symposium on Biomedical Imaging (ISBI)

    Zhang, Z., Qi, X., Zhang, B., Wu, B., Le, H., Jeong, B., Liao, Z., Liu, Y., Verjans, J., To, M.S., et al.: Segreg: Segmenting oars by registering mr images and ct an- notations. In: 2024 IEEE International Symposium on Biomedical Imaging (ISBI). pp. 1–5. IEEE (2024)

    2024

  43. [51]

    In: 2024 IEEE International Conference on Bioinformatics and Biomedicine (BIBM)

    Zhang, Z., Yi, N., Tan, S., Cai, Y., Yang, Y., Xu, L., Li, Q., Yi, Z., Ergu, D., Zhao, Y.: Meddet: Generative adversarial distillation for efficient cervical disc her- niation detection. In: 2024 IEEE International Conference on Bioinformatics and Biomedicine (BIBM). pp. 4024–...

    2024

  44. [52]

    OpenReview (2023)

    Zhang, Z., Zhang, B., Hiwase, A., Barras, C., Chen, F., Wu, B., Wells, A.J., Ellis, D.Y., Reddi, B., Burgan, A.W., To, M.S., Reid, I., Hartley, R.: Thin-thick adapter: Segmenting thin scans using thick annotations. OpenReview (2023)

    2023

  45. [53]

    arXiv preprint arXiv:2502.14063 (2025)

    Zhao, R., Zhang, Z., Xu, Y., Yao, Y., Huang, Y., Zhang, W., Song, Z., Chen, X., Zhao, Y.: Peddet: Adaptive spectral optimization for multimodal pedestrian detection. arXiv preprint arXiv:2502.14063 (2025)

    2025 arXiv

  46. [54]

    In: 2024 IEEE International Symposium on Biomedical Imaging (ISBI)

    Zhao, Y., Liao, Z., Liu, Y., Oude Nijhuis, K., Barvelink, B., Prijs, J., Colaris, J., Wijffels, M., Reijman, M., Zhang, Z., et al.: A landmark-based approach for insta- bility prediction in distal radius fractures. In: 2024 IEEE International Symposium on Biomedical Imaging (I...

    2024

  47. [55]

    In: European Conference on Computer Vision

    Zhao, Z., Wang, Z., Wang, L., Yu, D., Yuan, Y., Zhou, L.: Alternate diverse teach- ing for semi-supervised medical image segmentation. In: European Conference on Computer Vision. pp. 227–243. Springer (2024)

    2024

  48. [56]

    In: Proceedings of the 31st ACM International Conference on Multimedia

    Zhao, Z., Zhao, M., Liu, Y., Yin, D., Zhou, L.: Entropy-based optimization on individual and global predictions for semi-supervised learning. In: Proceedings of the 31st ACM International Conference on Multimedia. pp. 8346–8355 (2023)

    2023

  49. [57]

    arXiv preprint arXiv:2502.15885 (2025)

    Zhu, H., Zhang, Z., Pang, G., Wang, X., Wen, S., Bai, Y., Ergu, D., Cai, Y., Zhao, Y.: Doei: Dual optimization of embedding information for attention-enhanced class activation maps. arXiv preprint arXiv:2502.15885 (2025)

    2025 arXiv

  50. [58]

    arXiv preprint arXiv:2408.00874 (2024)

    Zhu, J., Qi, Y., Wu, J.: Medical sam 2: Segment medical images as video via segment anything model 2. arXiv preprint arXiv:2408.00874 (2024)

    2024 arXiv

Pith tools

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