Pith. sign in

REVIEW 3 major objections 6 minor 43 references

TABNet: A Triplet Augmentation Self-Recovery Framework with Boundary-Aware Pseudo-Labels for Medical Image Segmentation

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read TABNet's scribble-only training reaches 89.1% and 91.1% average Dice on cardiac MRI datasets, nearly matching fully supervised nnU-Net.

desk verdict A well-ablated weak-supervision paper with a plausible new design, but the headline gains may hinge on hyperparameters tuned on the test split rather than a held-out validation fold. read the letter →

arxiv 2507.02399 v1 pith:NMTT777E submitted 2025-07-03 cs.CV cs.LG

Peilin Zhang , Shaouxan Wua , Jun Feng , Zhuo Jin , Zhizezhang Gao , Jingkun Chen , Yaqiong Xing , Xiao Zhang This is my paper
classification cs.CVcs.LG
keywords scribble-supervisedsegmentationmedicalimagetripletaugmentationself-recoveryboundary-awarepseudo-labelsloss-weightedfusioncardiacMRIweaksupervision
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

Scribble annotations are sparse lines drawn through organs, but this paper claims it can train a cardiac MRI segmentation network to nearly full-supervision quality from them alone. The proposed framework, TABNet, couples a triplet augmentation self-recovery module with a boundary-aware pseudo-label module, and reports 89.1% average Dice on MSCMRseg and 91.1% on ACDC. Those numbers beat every compared weakly supervised method and sit within 0.6–0.8% of the fully supervised nnU-Net on the same splits. If the result holds, the main annotation bottleneck for cardiac segmentation—dense pixel masks—could be relaxed to quick scribbles.

What carries the argument

The load-bearing mechanism is the pairing of two modules. The triplet augmentation self-recovery (TAS) module builds three augmented views $x_i=C(x)$ (cutout masking the scribble-bounding box), $x_j=J(x)$ (jigsaw patch shuffle), and $x_k=I(x)$ (intensity shift), runs them through one shared-weight network, un-shuffles the jigsaw prediction with the inverse operation $J'(\cdot)$, and supervises each branch with cross-entropy only on scribble pixels. The boundary-aware pseudo-label (BAP) module then weights the jigsaw and intensity branch predictions by the reciprocal of their scribble cross-entropy losses, $\omega_j = \mathcal{L}_{ce}(y_j,s)/(\mathcal{L}_{ce}(y_j,s)+\mathcal{L}_{ce}(y_k,s))$ and symmetrically for $\omega_k$, forms the fused pseudo-label by class-wise argmax, and applies Dice loss between the pseudo-label and each branch plus a boundary Dice loss computed from soft-eroded boundary maps $B=\mathrm{ReLU}(y-\mathrm{MinPooling}(y))$. The final loss is $\mathcal{L}=\lambda_1\mathcal{L}_{TAS}+\lambda_2\mathcal{L}_{PL}+\lambda_3\mathcal{L}_{BD}$ with weights 1.0, 0.3, 0.1.

What would settle it

Take a test slice, compute each branch's score on just the scribble pixels and its Dice score against the full ground truth; if the branch with the lower scribble-pixel error is not the branch with the higher full-image Dice in a substantial share of slices, loss-weighted fusion is not selecting the better prediction and the reported 89.1% vs 88.3% gap over average fusion should shrink or vanish.

Watch

Extended reading notes

Core claim

The paper's central claim is that two complementary mechanisms—forcing a shared network to reconstruct full masks from intensity-shifted, cutout, and jigsaw-shuffled inputs, and supervising it with pseudo-labels formed by loss-weighted fusion of the jigsaw and intensity branches plus a soft boundary consistency term—recover most of the information that dense labels provide. On MSCMRseg the model reaches 89.1% average Dice versus 89.7% for nnU-Net; on ACDC it reaches 91.1% versus 91.7%. The authors attribute the gain to the self-recovery objective for semantic completeness and to the boundary-aware pseudo-label supervision for contour precision, and the ablations support this by showing each term adds a step: base 81.9%, plus TAS 83.4%, plus pseudo-label Dice 87.8%, plus boundary loss 89.1%.

Load-bearing premise

The whole pseudo-label fusion trick rests on assuming that how well a branch matches the few labeled scribble lines tells which branch's full-image prediction is better; a branch can match those lines while misplacing edges elsewhere, and the fused label would then copy that mistake.

Editorial extensions

If this is right

  • Scribble-only training becomes a practical substitute for dense annotation in cardiac MRI: on these benchmarks the gap to the best fully supervised method is under one Dice point.
  • The full triplet of intensity, cutout, and jigsaw augmentations is worth more than any pair: the three-branch TAS reaches 89.1% versus 88.5% for the best pair, and jigsaw alone is the strongest single branch at 86.9%.
  • Loss-weighted fusion is the right way to combine pseudo-label branches: it beats average fusion by 0.8 Dice points (89.1% vs 88.3%) and random fusion by 0.7 points (89.1% vs 88.4%).
  • Pseudo-labels should be built from jigsaw and intensity branches only: adding the cutout branch to the fusion drops performance from 89.1% to 88.5%, so the cutout view helps as a training signal but hurts as a pseudo-label source.
  • Boundary-aware supervision contributes at small weight: adding $\mathcal{L}_{BD}$ with $\lambda_3=0.1$ improves RV and Myo Dice by about two points over the no-boundary setting.

Reading between the lines

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

  • A natural next test is to move the same two modules to other sparse-supervision regimes, such as point-click or bounding-box labels; nothing in the design is cardiac-specific, but the benefit may depend on the scribble stroke pattern.
  • The drop from three-branch to two-branch fusion suggests the cutout branch's predictions are systematically less reliable under occlusion; gating branches by predicted uncertainty instead of scribble loss might recover the extra view's value.
  • The result implies the gap between weakly and fully supervised segmentation may be more about annotation geometry than about the label budget: scribbles happen to run through organ interiors, which is exactly what the fusion loss rewards, so a different sparse label layout could change the ranking of methods.
  • A cheaper practical extension would be to apply the loss-weighted fusion to two checkpoints of the same network instead of two augmented branches, which would test whether view diversity or model diversity is the active ingredient.
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

3 major / 6 minor

Summary. The manuscript presents TABNet, a weakly supervised medical image segmentation framework combining a triplet augmentation self-recovery module (TAS) with boundary-aware pseudo-label supervision (BAP). TAS feeds intensity, cutout, and jigsaw-augmented views into a shared-weight segmentation network and trains with cross-entropy on scribble pixels; BAP fuses branch predictions with loss-based weights and adds a boundary Dice loss. The method is evaluated on MSCMRseg and ACDC with scribble supervision, reporting average Dice of 89.1% and 91.1%, outperforming prior weakly supervised methods and approaching fully supervised nnU-Net (89.7% and 91.7%). The paper includes detailed ablations of each module and hyperparameters.

Significance. If the reported numbers survive closer scrutiny, TABNet is a practically useful contribution: it gives a clear, reproducible framework (code released) with competitive performance on standard benchmarks, and the ablations support the design choices of the TAS and BAP modules. The paper's main weaknesses are methodological reporting ambiguities: the split used for hyperparameter selection is not stated, and no statistical significance tests are reported for the headline claims. These issues are fixable but currently prevent full confidence in the central claim of state-of-the-art scribble-supervised segmentation.

major comments (3)
  1. [Section 4.4 and Tables 3-5] The ablation studies in Section 4.4 select the loss weights (λ1=1.0, λ2=0.3, λ3=0.1) and the pseudo-label fusion pair PL(y_j,y_k) on the MSCMRseg dataset, but the text does not state that the 5-case validation split described in Section 4.1 was used for this selection. Since Table 5's λ-sweep spans 86.5 to 89.1 Dice and Table 4's branch-pair choices span 88.4 to 89.1 Dice, while the reported margin over QMaxViT-Unet+ is only 1.1 points (89.1 vs 88.0), selection on the 15-case test fold would materially inflate the headline result. Please state explicitly that all model and hyperparameter selection was performed on the validation split and that the test fold was used only for the final evaluation; if this was not the case, the experiments should be re-run with selection on the validation split.
  2. [Abstract and Table 1] The abstract states that TABNet 'significantly outperforms' prior weakly supervised methods, but Table 1 reports no statistical significance tests for the main comparisons. Table 2 introduces a significance marker (p<0.05) without describing the statistical test used or reporting p-values. Please add appropriate case-level paired tests (e.g., Wilcoxon signed-rank or paired t-test) across the test cases and report the resulting p-values for the key comparisons in Table 1, or remove the word 'significantly' from the claims.
  3. [Equation (5) and Table 4] The loss-weighted fusion rule in Eq. (5) weights each branch's full-image prediction using the cross-entropy loss computed only on scribble pixels. Because scribble strokes typically pass through structure interiors, a branch with low scribble loss can still have poor boundary predictions elsewhere; the fused pseudo-label in Eq. (6) can then propagate those boundary errors through the Dice loss in Eq. (8). The only evidence supporting this proxy is the 0.7–0.8 point improvement of loss-weighted fusion over average and random weighting in Table 4. Please provide a direct analysis of the fusion rule's reliability (e.g., ablating with oracle or ground-truth-based weights, or reporting the correlation between loss-based weights and region-level segmentation quality), or explicitly discuss this limitation in the text.
minor comments (6)
  1. [Table 4] The term 'cropping augmentation' used in Table 4 and its surrounding text should be 'cutout augmentation' for consistency with the method description.
  2. [Conclusions] The sentence 'closely methods the performance of fully supervised models' should read 'closely matches the performance of fully supervised models.'
  3. [Abstract] The name 'TAB Net' appears with a space in the abstract, while the rest of the paper uses 'TABNet'; please unify the spelling.
  4. [Equation (5)] The notation 'w_j,w_k= ...' in Eq. (5) is ambiguous; please define each weight separately as w_j and w_k with explicit formulas.
  5. [Section 4.4.4] The claim that the model has 'standard deviations below ±7%' is unclear and inconsistent with the reported standard deviations, which are actually below 0.07 in Dice units; please rephrase as 'standard deviations below 0.07'.
  6. [Section 3.1] The term 'self-recovery' overstates the mechanism, since the cross-entropy loss in Eq. (3) is applied only to scribble pixels; consider clarifying that the model is trained to maintain correct predictions on labeled pixels under heavy augmentation rather than to reconstruct the complete mask.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the headline 89.1%/91.1% Dice results are held-out test evaluations, and the BAP/TAS modules are empirical consistency/pseudo-label mechanisms, not predictions derived from their own outputs.

full rationale

The paper makes no first-principles derivation that could collapse into its inputs. The central claims are empirical: Table 1 reports Dice on the 15-case MSCMRseg and 15-case ACDC test folds fixed in Section 4.1 (25/5/15 and 35/15/15 splits), so the comparison to QMaxViT-Unet+ and nnU-Net is against held-out ground truth. The BAP module (Eqs. 5-8) constructs a loss-weighted pseudo-label from the model's own two branch predictions and then applies Dice and boundary consistency losses; this is a self-referential consistency regularizer standard in pseudo-labeling, but it is not a 'prediction' claimed to be derived from independent first principles, and its contribution is tested by ablations (Table 2) rather than assumed. No load-bearing self-citation, ansatz-smuggling, or renamed known result was found; the few self-citations (e.g., [27], [34]) are related-work context. The only flagged concern is reporting ambiguity: Sections 4.4.3-4.4.4 and Tables 3-5 describe ablations 'on the MSCMRseg dataset' without explicitly stating that the 5-case validation fold was used for hyperparameter/branch selection. If the test fold were used for selection, the headline 89.1% would be a selected rather than predicted number; but that is an evaluation-protocol risk, not a circularity of the derivation, and the paper's stated adherence to the standard split supports the non-circular reading.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The free parameters are the three loss weights (tuned on the reported dataset) plus four unquantified augmentation and boundary-extraction settings. The axioms are the proxy assumptions that make the pseudo-label and boundary modules work, and the assumption that reported baselines are comparable. No new entities are postulated; the pseudo-label is a standard self-training construct, not a newly invented object.

free parameters (5)
  • λ1, λ2, λ3 loss weights = λ1=1.0, λ2=0.3, λ3=0.1
    Chosen by sweeping each λ on MSCMRseg (Table 5) and keeping the combination that maximized Dice; the same MSCMRseg test split is used for the headline 89.1% result, making the reported number dependent on test-set hyperparameter selection.
  • Jigsaw augmentation configuration = not reported
    The number and size of shuffled patches and the shuffle randomness are not specified; they control the difficulty of the self-recovery task and the quality of the jigsaw branch pseudo-labels.
  • Cutout mask size and position = not reported
    The cutout masks the maximum bounding box inferred from scribbles, but the fraction of the box removed and the mask shape are unspecified; this directly controls how much recovery the model must perform.
  • Intensity shift and contrast range = not reported
    The brightness and contrast perturbation ranges are not quantified; they determine the difficulty of the intensity branch.
  • Min-pooling kernel for boundary extraction = not reported
    Eq. 9 uses MinPooling over an unspecified kernel size; the kernel determines the thickness of the extracted boundary maps and therefore the behavior of the boundary loss.
assumptions (4)
  • domain assumption The cross-entropy loss computed over scribble pixels is a reliable proxy for full-image prediction quality of each branch, so weighting the pseudo-label by the inverse of these losses improves supervision.
    Eq. 5 computes fusion weights solely from L_ce on the sparse scribble set Ω_s; the entire BAP contribution (Table 2, +4.4% Dice) depends on this proxy holding for unlabeled and boundary pixels.
  • domain assumption Scribble annotations correctly mark foreground pixels of the target structures and provide enough signal to train all three augmented branches.
    Scribble labels are the only ground-truth signal; the TAS loss (Eq. 4) trains the intensity, cutout, and jigsaw branches only on Ω_s, and the cutout masks the region inferred from these same scribbles.
  • domain assumption Soft erosion by min-pooling (B = ReLU(y - MinPooling(y))) yields boundary maps whose Dice consistency corresponds to anatomically meaningful contour refinement.
    Eq. 9 defines the boundary maps used by the boundary-aware loss in Eq. 10; there is no external boundary ground truth, so the boundary supervision is a self-consistency constraint on the model's own predictions.
  • domain assumption Previously published results for baseline methods (Table 1) are directly comparable when the same public data splits are used.
    All weakly supervised baselines in Table 1 are taken from their papers; the comparison assumes matched backbones, training schedules, and augmentation settings, which the authors do not re-run.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TABNet: A Triplet Augmentation Self-Recovery Framework with Boundary-Aware Pseudo-Labels for Medical Image Segmentation." pith.science (2026). https://pith.science/paper/NMTT777E

@misc{pith2026250702399,
  author       = {Pith},
  title        = {Pith review of: TABNet: A Triplet Augmentation Self-Recovery Framework with Boundary-Aware Pseudo-Labels for Medical Image Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NMTT777E}},
  note         = {Machine review of arXiv:2507.02399}
}
read the original abstract

Background and objective: Medical image segmentation is a core task in various clinical applications. However, acquiring large-scale, fully annotated medical image datasets is both time-consuming and costly. Scribble annotations, as a form of sparse labeling, provide an efficient and cost-effective alternative for medical image segmentation. However, the sparsity of scribble annotations limits the feature learning of the target region and lacks sufficient boundary supervision, which poses significant challenges for training segmentation networks. Methods: We propose TAB Net, a novel weakly-supervised medical image segmentation framework, consisting of two key components: the triplet augmentation self-recovery (TAS) module and the boundary-aware pseudo-label supervision (BAP) module. The TAS module enhances feature learning through three complementary augmentation strategies: intensity transformation improves the model's sensitivity to texture and contrast variations, cutout forces the network to capture local anatomical structures by masking key regions, and jigsaw augmentation strengthens the modeling of global anatomical layout by disrupting spatial continuity. By guiding the network to recover complete masks from diverse augmented inputs, TAS promotes a deeper semantic understanding of medical images under sparse supervision. The BAP module enhances pseudo-supervision accuracy and boundary modeling by fusing dual-branch predictions into a loss-weighted pseudo-label and introducing a boundary-aware loss for fine-grained contour refinement. Results: Experimental evaluations on two public datasets, ACDC and MSCMR seg, demonstrate that TAB Net significantly outperforms state-of-the-art methods for scribble-based weakly supervised segmentation. Moreover, it achieves performance comparable to that of fully supervised methods.

Figures

Figures reproduced from arXiv: 2507.02399 by the authors.

Figure 1
Figure 1. Overview of the proposed TABNet framework. The model combines triplet augmentation and self-recovery to produce multi-scale predictions, and improves PL quality and boundary accuracy via loss-weighted and boundary constraints. In contrast to prior methods that generate pseudo-labels via random weighting or averaging, we propose a loss￾guided pseudo-label fusion strategy that dynamically weighs predictions based on t… view at source ↗
Figure 2
Figure 2. Boundary-aware pseudo-label supervision enhances segmentation by combining loss-weighted fusion with soft boundary extraction. weights are computed based on the cross-entropy loss with respect to the scribble annotations: 𝑤𝑗 , 𝑤𝑘= 𝑐𝑒(𝑦𝑘 ,𝑠) 𝑐𝑒(𝑦𝑗 ,𝑠)+𝑐𝑒(𝑦𝑘 ,𝑠) , 𝑐𝑒(𝑦𝑗 ,𝑠) 𝑐𝑒(𝑦𝑗 ,𝑠)+𝑐𝑒(𝑦𝑘 ,𝑠) , (5) where 𝑠 denotes the scribble annotations and 𝑐𝑒 is the cross-entropy loss. This weighting strategy suppresses the… view at source ↗
Figure 3
Figure 3. Visualization on four typical cases from the MSCMRseg dataset for illustration and comparison. Blue dashed circles highlight regions with ambiguous boundaries, and yellow dashed circles indicate areas of over- or under-segmentation. 4.2. Experimental Details All experiments were conducted using the PyTorch framework and accelerated on an NVIDIA RTX 4090 GPU (24GB). During preprocessing, each image was standardized t… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Visualization on four typical cases from the ACDC dataset for illustration and comparison. Blue dashed circles highlight regions with ambiguous boundaries, and yellow dashed circles indicate areas of over- or under-segmentation. that of the state-of-the-art fully super…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

43 extracted references · 37 canonical work pages

  1. [1]

    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.,

    work page

  2. [2]

    Semi-supervisedmedicalimagesegmentationvialearning consistencyundertransformations,in:MedicalImageComputingand Computer Assisted Intervention, Springer

    Bortsova, G., Dubost, F., Hogeweg, L., Katramados, I., De Bruijne, M.,2019. Semi-supervisedmedicalimagesegmentationvialearning consistencyundertransformations,in:MedicalImageComputingand Computer Assisted Intervention, Springer. pp. 810–818

    work page 2019

  3. [3]

    Chen, J., Chen, C., Huang, W., Zhang, J., Debattista, K., Han, J.,

    work page

  4. [4]

    Ad- dressinginconsistentlabelingwithcrossimagematchingforscribble- based medical image segmentation

    Chen, J., Huang, W., Zhang, J., Debattista, K., Han, J., 2025a. Ad- dressinginconsistentlabelingwithcrossimagematchingforscribble- based medical image segmentation. IEEE Transactions on Image Processing

    work page

  5. [5]

    A survey on deep learning in medical imageregistration:Newtechnologies,uncertainty,evaluationmetrics, and beyond

    Chen, J., Liu, Y., Wei, S., Bian, Z., Subramanian, S., Carass, A., Prince, J.L., Du, Y., 2025b. A survey on deep learning in medical imageregistration:Newtechnologies,uncertainty,evaluationmetrics, and beyond. Medical Image Analysis 100, 103385

    work page

  6. [6]

    Improvedregularizationofconvolu- tional neural networks with cutout

    DeVries,T.,Taylor,G.W.,2017. Improvedregularizationofconvolu- tional neural networks with cutout. arXiv preprint arXiv:1708.04552

    arXiv 2017

  7. [7]

    Han, M., Luo, X., Liao, W., Zhang, S., Zhang, S., Wang, G., 2023. Scribble-based3Dmultipleabdominalorgansegmentationviatriple- branch multi-dilated network with pixel-and class-wise consistency, in: International Conference on Medical Image Computing and Computer-Assisted Intervention, Springer. pp. 33–42

    work page 2023

  8. [8]

    DMSPS: Dynamically mixed soft pseudo- label supervision for scribble-supervised medical image segmenta- tion

    Han, M., Luo, X., Xie, X., Liao, W., Zhang, S., Song, T., Wang, G., Zhang, S., 2024. DMSPS: Dynamically mixed soft pseudo- label supervision for scribble-supervised medical image segmenta- tion. Medical Image Analysis 97, 103274

    work page 2024

Show all 43 references
  1. [9]

    Medical Image Analysis 91, 102984

    He,W.,Zhang,C.,Dai,J.,Liu,L.,Wang,T.,Liu,X.,Jiang,Y.,Li,N., Xiong,J.,Wang,L.,Xie,Y.,Liang,X.,2024.Astatisticaldeformation model-baseddataaugmentationmethodforvolumetricmedicalimage segmentation. Medical Image Analysis 91, 102984

    2024

  2. [10]

    Generative feature style aug- mentation for domain generalization in medical image segmentation

    Huang, Y., Han, L., Dou, H., 2025. Generative feature style aug- mentation for domain generalization in medical image segmentation. Pattern Recognition 162, 111416

    2025

  3. [11]

    Isensee, F., Jaeger, P.F., Kohl, S.A., Petersen, J., Maier-Hein, K.H.,

  4. [12]

    Co-mixup: Saliency guided joint mixup with supermodular diversity

    Kim, J.H., Choo, W., Jeong, H., Song, H.O., 2021. Co-mixup: Saliency guided joint mixup with supermodular diversity. arXiv preprint arXiv:2102.03065

    2021 arXiv

  5. [13]

    Puzzle mix: Exploiting saliency and local statistics for optimal mixup, in: International Con- ference on Machine Learning, PMLR

    Kim, J.H., Choo, W., Song, H.O., 2020. Puzzle mix: Exploiting saliency and local statistics for optimal mixup, in: International Con- ference on Machine Learning, PMLR. pp. 5275–5285

    2020

  6. [14]

    Data augmentationtechniquesformedicalimagesegmentation–areview, in:2024InternationalConferenceonComputerandApplications,pp

    Kozah, N., Dornaika, F., Charafeddine, J., El Jaam, J., 2024. Data augmentationtechniquesformedicalimagesegmentation–areview, in:2024InternationalConferenceonComputerandApplications,pp. 1–8

    2024

  7. [15]

    Lei, Y., Luo, H., Wang, L., Zhang, Z., Zhang, L., 2024. PCLMix: Weakly supervised medical image segmentation via pixel-level con- trastive learning and dynamic mix augmentation, in: International Conference on Intelligent Computing, Springer. pp. 62–73

    2024

  8. [16]

    Transformation-consistentself-ensemblingmodelforsemisupervised medicalimagesegmentation

    Li, X., Yu, L., Chen, H., Fu, C.W., Xing, L., Heng, P.A., 2020. Transformation-consistentself-ensemblingmodelforsemisupervised medicalimagesegmentation. IEEETransactionsonNeuralNetworks and Learning Systems 32, 523–534

    2020

  9. [17]

    ScribbleVC: Scribble-supervised medical image segmentation with vision-class embedding, in: Proceedings of the 31st ACM International Confer- ence on Multimedia, p

    Li, Z., Zheng, Y., Luo, X., Shan, D., Hong, Q., 2023. ScribbleVC: Scribble-supervised medical image segmentation with vision-class embedding, in: Proceedings of the 31st ACM International Confer- ence on Multimedia, p. 3384–3393

    2023

  10. [18]

    Scribformer: Transformer makes CNN work better for scribble-based medical image segmentation

    Li, Z., Zheng, Y., Shan, D., Yang, S., Li, Q., Wang, B., Zhang, Y., Hong, Q., Shen, D., 2024. Scribformer: Transformer makes CNN work better for scribble-based medical image segmentation. IEEE Transactions on Medical Imaging 43, 2254–2265

    2024

  11. [19]

    SSFam: Scribble supervised salient object detection family

    Liu, Z., Deng, S., Wang, X., Wang, L., Fang, X., Tang, B., 2025. SSFam: Scribble supervised salient object detection family. IEEE Transactions on Multimedia 27, 1988–2000

    2025

  12. [20]

    Luo, X., Hu, M., Liao, W., Zhai, S., Song, T., Wang, G., Zhang, S., 2022. Scribble-supervised medical image segmentation via dual-branch network and dynamically mixed pseudo labels supervi- sion, in: International Conference on Medical Image Computing and Computer-Assisted Int...

    2022

  13. [21]

    QMaxViT-Unet+: A query-based MaxViT-Unet with edge enhancement for scribble- supervised segmentation of medical images

    Nguyen-Tat, T.B., Vo, H.A., Dang, P.S., 2025. QMaxViT-Unet+: A query-based MaxViT-Unet with edge enhancement for scribble- supervised segmentation of medical images. Computers in Biology and Medicine 187, 109762

    2025

  14. [22]

    U-Net: Convolutional networks for biomedical image segmentation, in: Medical Image Computing and Computer-Assisted Intervention, Springer

    Ronneberger, O., Fischer, P., Brox, T., 2015. U-Net: Convolutional networks for biomedical image segmentation, in: Medical Image Computing and Computer-Assisted Intervention, Springer. pp. 234– 241

    2015

  15. [23]

    Mutuallearningwithre- liablepseudolabelforsemi-supervisedmedicalimagesegmentation

    Su,J.,Luo,Z.,Lian,S.,Lin,D.,Li,S.,2024. Mutuallearningwithre- liablepseudolabelforsemi-supervisedmedicalimagesegmentation. Medical Image Analysis 94, 103111

    2024

  16. [24]

    Embracing imperfect datasets: A review of deep learning solutions for medical image segmentation

    Tajbakhsh, N., Jeyaseelan, L., Li, Q., Chiang, J.N., Wu, Z., Ding, X., 2020. Embracing imperfect datasets: A review of deep learning solutions for medical image segmentation. Medical Image Analysis 63, 101693

    2020

  17. [25]

    Deeplearningonmedicalimage analysis

    Wang,J.,Wang,S.,Zhang,Y.,2025. Deeplearningonmedicalimage analysis. CAAI Transactions on Intelligence Technology 10, 1–35

    2025

  18. [26]

    Competetowin:Enhanc- ingpseudolabelsforbarely-supervisedmedicalimagesegmentation

    Wu,H.,Li,X.,Lin,Y.,Cheng,K.T.,2023. Competetowin:Enhanc- ingpseudolabelsforbarely-supervisedmedicalimagesegmentation. IEEE Transactions on Medical Imaging 42, 3244–3255

    2023

  19. [27]

    Gaze- directed vision GNN for mitigating shortcut learning in medical image,in:InternationalConferenceonMedicalImageComputingand Computer-Assisted Intervention, Springer

    Wu, S., Zhang, X., Wang, B., Jin, Z., Li, H., Feng, J., 2024. Gaze- directed vision GNN for mitigating shortcut learning in medical image,in:InternationalConferenceonMedicalImageComputingand Computer-Assisted Intervention, Springer. pp. 514–524

    2024

  20. [28]

    Xu, Z., Tang, J., Qi, C., Yao, D., Liu, C., Zhan, Y., Lukasiewicz, T.,

  21. [29]

    Non-iterative scribble- supervised learning with pacing pseudo-masks for medical image segmentation

    Yang, Z., Lin, D., Ni, D., Wang, Y., 2024. Non-iterative scribble- supervised learning with pacing pseudo-masks for medical image segmentation. Expert Systems with Applications 238, 122024

    2024

  22. [30]

    Cutmix: Regularization strategy to train strong classifiers with localizable features, in: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp

    Yun,S.,Han,D.,Oh,S.J.,Chun,S.,Choe,J.,Yoo,Y.,2019. Cutmix: Regularization strategy to train strong classifiers with localizable features, in: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 6023–6032

    2019

  23. [31]

    mixup: Beyond empirical risk minimization

    Zhang, H., Cisse, M., Dauphin, Y.N., Lopez-Paz, D., 2017. mixup: Beyond empirical risk minimization. arXiv preprint arXiv:1710.09412

    2017 arXiv

  24. [32]

    Computers in Biology and Medicine 168, 107744

    Cross-domain attention-guided generative data augmentation for medical image analysis with limited data. Computers in Biology and Medicine 168, 107744

  25. [33]

    Zhang, K., Zhuang, X., 2022b. ShapePU: A new PU learning framework regularized by global consistency for scribble supervised cardiacsegmentation,in:InternationalConferenceonMedicalImage Computing and Computer-Assisted Intervention, Springer. pp. 162– 172

  26. [34]

    An anatomy-and topology-preserving framework for coronary artery segmentation

    Zhang, X., Sun, K., Wu, D., Xiong, X., Liu, J., Yao, L., Li, S., Wang, Y., Feng, J., Shen, D., 2023. An anatomy-and topology-preserving framework for coronary artery segmentation. IEEE Transactions on Peilin Zhang et al.:Preprint submitted to Elsevier Page 10 of 11 Medical Ima...

    2023

  27. [35]

    Scribblehidesclass: Promoting scribble-based weakly-supervised semantic segmentation with its class label, in: Proceedings of the AAAI Conference on Artificial Intelligence, pp

    Zhang,X.,Zhu,L.,He,H.,Jin,L.,Lu,Y.,2024. Scribblehidesclass: Promoting scribble-based weakly-supervised semantic segmentation with its class label, in: Proceedings of the AAAI Conference on Artificial Intelligence, pp. 7332–7340

    2024

  28. [36]

    CycleMix: A holistic strategy for medical image segmentation from scribble supervision, in: Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Zhang, K., Zhuang, X., 2022a. CycleMix: A holistic strategy for medical image segmentation from scribble supervision, in: Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 11656–11665

  29. [37]

    UNet++: A nested U-Net architecture for medical image segmenta- tion, in: Deep learning in Medical Image Analysis and Multimodal Learning For Clinical Decision Support, Springer

    Zhou, Z., Rahman Siddiquee, M.M., Tajbakhsh, N., Liang, J., 2018. UNet++: A nested U-Net architecture for medical image segmenta- tion, in: Deep learning in Medical Image Analysis and Multimodal Learning For Clinical Decision Support, Springer. pp. 3–11

    2018

  30. [38]

    Multivariate mixture model for myocardial seg- mentation combining multi-source images

    Zhuang, X., 2018. Multivariate mixture model for myocardial seg- mentation combining multi-source images. IEEE Transactions on Pattern Analysis and Machine Intelligence 41, 2933–2946. Peilin Zhang et al.:Preprint submitted to Elsevier Page 11 of 11

    2018

  31. [40]

    Zhou, S., Tian, S., Yu, L., Wu, W., Zhang, D., Peng, Z., Zhou, Z.,

  32. [41]

    EngineeringAppli- cations of Artificial Intelligence 130, 107777

    Growththresholdforpseudolabelingandpseudolabeldropout forsemi-supervisedmedicalimageclassification. EngineeringAppli- cations of Artificial Intelligence 130, 107777

  33. [2018]

    Deep learning techniques for automatic MRI cardiac multi- structures segmentation and diagnosis: Is the problem solved? IEEE Transactions on Medical Imaging 37, 2514–2525

  34. [2021]

    Nature Methods 18, 203–211

    nnU-Net: A self-configuring method for deep learning-based biomedical image segmentation. Nature Methods 18, 203–211

  35. [2024]

    Pattern Recognition 145, 109881

    Dynamic contrastive learning guided by class confidence and confusion degree for medical image segmentation. Pattern Recognition 145, 109881

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.