Pith. sign in

REVIEW 3 major objections 5 minor 62 references

Constrained domain adaptation for Image segmentation

T0 review · 3 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read Inequality constraints on predicted region sizes adapt segmentation networks across MRI and CT domains, outperforming adversarial domain adaptation and approaching fully supervised target accuracy.

desk verdict A simple, well-executed constrained DA method for medical segmentation, but the headline performance claims are oversold and the robustness study misses the systematic-prior-shift failure mode. read the letter →

arxiv 1908.02996 v2 pith:FFNIYB2O submitted 2019-08-08 cs.CV

Mathilde Bateson , Jose Dolz , Hoel Kervadec , Hervé Lombaert , Ismail Ben Ayed This is my paper · ORCID
classification cs.CV
keywords domainadaptationimagesegmentationconstrainedoptimizationinequalityconstraintsregionsizepriormedicalimagingMRItoCTsemantic
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 tries to establish that domain adaptation for image segmentation can be done by constraining the network's outputs on unlabeled target images with domain-invariant prior knowledge, rather than by training an adversarial domain discriminator. The specific prior used is region size, encoded as lower and upper inequality bounds on the predicted size of each structure, with a quadratic penalty that allows imprecision. This matters because adversarial adaptation for segmentation is unstable, slow, and underperforms; a single constrained loss is simpler and about twice as fast per batch. On two medical imaging tasks—spine MRI across modalities and cardiac MRI to CT—the constrained model outperforms adversarial and curriculum baselines and reaches 86% and 80% of the fully supervised Oracle Dice with learned size priors. The authors also show that rough priors from source statistics, with up to 50% uncertainty, still yield large gains over no adaptation.

What carries the argument

The load-bearing object is the inequality constraint on the spatial sum of a class's softmax output, i.e., the predicted region size, enforced through the penalty $F(\theta)=\sum_c \sum_t [f_c(P_t(\theta))]_+^2$ with $[x]_+=\max(0,x)$. Size is the zeroth-order shape moment, so the same construction can in principle carry higher-order moment constraints. The penalty turns the constrained problem into an unconstrained one that runs in ordinary stochastic gradient descent on a single network, replacing the alternating discriminator of adversarial domain adaptation and allowing uncertainty through lower and upper bounds $a$ and $b$.

What would settle it

Apply the source-median-prior variant to a target cohort with systematically larger structures (for example, adults versus children, or thicker slices) and compare Dice to the random-margin ablation; a drop beyond the reported 5–6% would falsify the claimed robustness to prior imprecision.

Watch

Extended reading notes

Core claim

The central claim is that a segmentation network can be adapted across domains by minimizing the source supervised loss subject to inequality constraints on target predictions, and that this constrained problem can be solved with one differentiable quadratic penalty instead of two-step adversarial training. When the constraints encode region size—known exactly, estimated by an auxiliary regression network, or derived from source statistics—the method implicitly matches prediction statistics between target and source with permitted uncertainty. On spine (Water to In-Phase MRI) and cardiac (MRI to CT) segmentation, this formulation outperforms the adversarial output-space adaptation method and the KL-divergence curriculum method, reaching 72.3% and 71.4% mean Dice respectively, which is 86% and 80% of the fully supervised upper bound. The robustness experiments show that loosening the size bounds to ±25–50% costs at most 5–6% Dice, and that even source-median priors beat no adaptation by a wide margin.

Load-bearing premise

The size prior learned in the source domain or taken from anatomical norms remains valid in the target domain; systematic differences in structure size between source and target would make the constraints pull segmentations toward the wrong size.

Editorial extensions

If this is right

  • With precise target sizes (within ±10%), the constrained model reaches 80.4% Dice on spine and 84.6% on cardiac, which is 95% of the fully supervised Oracle, so good priors nearly close the gap to full supervision.
  • With estimated priors from an auxiliary regression network, the method reaches 72.3% Dice on spine and 71.4% on cardiac, outperforming adversarial adaptation (57.3% and 41.1%) and KL-based curriculum adaptation (68.8% and 70.7%).
  • Loose priors still help: ±25–50% uncertainty costs at most 5–6% Dice, and even source-median priors beat no adaptation by a wide margin (60.7% and 64.2% versus 46.8% and 38.2%).
  • The method is model-agnostic—the same ranking holds with ENet and UNet—and it is about twice as fast per training batch as adversarial adaptation.
  • Removing the image-level tag supervision degrades all methods, but the constrained formulation still outperforms the adversarial and curriculum baselines in that harder setting.

Reading between the lines

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

  • Inference: Because region size is the zeroth-order shape moment, the same single-penalty construction should extend to higher-order moment constraints such as centroids and moments of inertia, and that extension is a natural next test for tasks where size alone is not discriminative.
  • Inference: The reported robustness ablations perturb the prior symmetrically around the true size; a systematic bias in the prior—for example, pediatric versus adult anatomy or a different slice thickness—is not covered and would be the first failure mode to probe before clinical deployment.
  • Inference: The tag-free variant still beats adversarial baselines, which suggests the constraint penalty could be combined with self-training or entropy minimization to push the fully unsupervised setting closer to the tagged results.
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

3 major / 5 minor

Summary. The paper proposes Constrained Domain Adaptation (CDA) for semantic segmentation, replacing two-step adversarial domain adaptation with a single-network penalty that imposes inequality constraints on target-domain region sizes. The constraints are encoded with a ReLU-squared penalty (Eq. 7), and the required size priors are obtained in three ways: exact target sizes with margins (Constaint10/25/50/75), an auxiliary regression network trained with source masks plus target image-level tags (ConstraintAdap), or source-domain median sizes (ConstraintLit). The method is evaluated on IVDM3Seg spine MRI Water-to-In-Phase adaptation and on MMWHS cardiac MRI-to-CT adaptation, with ENet and UNet backbones, and is compared against NoAdap, adversarial adaptation [19], curriculum KL-based adaptation [38], and the PnP-AdaNet benchmark. The central claims are that the constrained formulation improves over adversarial state-of-the-art methods, approaches Oracle accuracy when the size prior is accurate, and is robust to imprecision in the size prior.

Significance. If the results hold, CDA is a practically attractive alternative to adversarial domain adaptation for segmentation: it reduces training to a single segmentation network with one additional scalar penalty, supports weak image-level tags, and is accompanied by public code. The derivation of the penalty is straightforward and the constraints are external priors (source statistics, anatomical knowledge, or an auxiliary regressor), so there is no circularity in the method itself. The oracle-style Constraint10 ablation is clearly labeled as an upper-bound study. The empirical evidence that a noisy size prior can guide adaptation is valuable, and the efficiency comparison in Table 6 is a useful practical contribution. However, the significance is tempered by the absence of a test for systematically biased priors and by some overclaiming in the reported gains.

major comments (3)
  1. [§3.1.5 / Eq. (14), Table 3] The robustness study does not exercise the failure mode that matters most for the central premise that region size is a domain-invariant prior. Equation (14) transfers source-domain median pixel counts directly to target slices, but pixel counts are resolution- and field-of-view-dependent, and the MMWHS source and target have non-overlapping subjects acquired with different protocols. The ablation in Section 3.1.3 varies the margins around the true target size τ_t (Eq. 12), so the interval center is always correct and only the width changes. Under a systematic shift between the source-derived prior and true target sizes, the penalty in Eq. (7) actively penalizes correct target segmentations that fall outside a misplaced interval, so the constraint could harm rather than help adaptation. I request an experiment with a deliberately biased prior center (e.g., τ_S scaled by 0.7, 0.85, 1.15, 1.3) or, at minimum, a report of source versus target size distributions for both datasets, to support the claim that source-derived priors are a reliable guide.
  2. [Abstract, §1.2, Tables 1–2] The repeated claim of a consistent 1–4% Dice gain across architectures and datasets is not supported by the reported numbers. Table 1 shows ConstraintAdap at 73.4% versus KLAdap at 73.3% with UNet on spine (+0.1%), and Table 2 shows 71.4% versus 70.7% on cardiac (+0.7%). Per-structure results in Table 2 are worse for some structures (LA DSC 72.8 vs 77.9; MYO HD95 16.1 vs 11.2). With only 3 validation subjects for spine and 4 for cardiac, these differences are within the reported standard deviations, so the word 'significant' is not justified for the comparison against KLAdap. The stronger and more defensible claims are the gains over NoAdap and over the adversarial baseline [19]; the manuscript should be revised to state the comparison precisely and avoid the 1–4% phrasing.
  3. [§3.2, Tables 1–2] The paper does not provide any statistical significance testing or confidence intervals for the reported mean Dice differences. Given the small validation/test sets (3 spine subjects and 4 cardiac subjects) and the overlapping standard deviations for the main ConstraintAdap versus KLAdap comparison, the central claim of outperforming the curriculum baseline rests on small mean differences. I ask the authors to either add paired significance tests (e.g., Wilcoxon signed-rank or bootstrap confidence intervals on the per-subject Dice values) or explicitly soften the comparative claims to 'consistent improvements in mean Dice' where appropriate.
minor comments (5)
  1. [Eq. (12)] The notation a,b = {0.9τ_t, 1.1τ_t} is overloaded; it should be stated explicitly that a is the lower bound and b is the upper bound, since Eq. (4) uses a as an upper bound and Eq. (5) uses b as a lower bound, which is reversed relative to this shorthand.
  2. [Tables 1 and 2] The column header 'T ags' appears to be a typo for 'Tags'.
  3. [Figure 3 and §3.1.4] The size regressor's accuracy is described qualitatively as 'quite noisy', but no quantitative regression error (e.g., MAE or relative error) is reported. Reporting such numbers would make the robustness analysis in Table 3 more interpretable.
  4. [Eq. (6)] In Eq. (6), the cross-entropy term is written with p(i,θ) rather than p_s(i,θ), unlike Eq. (2); this is a minor notation inconsistency that should be corrected.
  5. [Table 4] The MMWHS KLAdap HD95 value is listed as N/A without a footnote; a brief explanation (e.g., no prediction for at least one structure) would improve the table.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the constrained size priors are external inputs or explicitly labeled upper-bound ablations, and the reported comparisons are against independent baselines.

full rationale

Careful walk-through of the derivation chain (Eqs. 1-9 and the experimental protocols) found no step in which a claimed prediction reduces by construction to its own input. The size constraints come from one of three sources: explicit anatomical knowledge, statistics of the source domain (Section 3.1.5, Eq. 14), or an auxiliary regression network trained on source labels plus target image-level tags (Section 2.2, Eqs. 8-9). None of these is derived from the segmentation network being adapted or from the final Dice evaluation. The Constraint10/25/50/75 settings use ground-truth target sizes, but the paper explicitly labels them as an investigation of precise information and as distinct from the main setting: 'First, we investigate the capability of the proposed CDA approach when precise information about the size of the segmentation regions is known' and 'This is different from the main experiments, where the ground truth target size tau_t is unknown.' The Oracle baseline is likewise explicitly an upper bound on achievable performance, not a claimed unsupervised prediction. Self-citations to [7] and [44] are normal prior-work references: the penalty in Eq. (7) is derived in the text, and [44] is identified as a preliminary conference version rather than being used as an external authority. The Discussion does acknowledge a real limitation ('the main limitation of our method relies on obtaining an accurate estimation of region size'), and the robustness ablations vary the margin around the true target size rather than testing a systematically biased prior center. That is a correctness or deployment-robustness concern about the size-invariance assumption, not a circularity of the derivation. The central comparisons against the adversarial method of [19], KLAdap [38], and the benchmark results of [43] are external and independent. Overall, no significant circularity is present.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The method's contributions are empirical; it introduces no new entities. The central burden is the domain-invariance of size priors and the adequacy of the penalty approximation, both acknowledged in the paper. The free parameters are training hyperparameters and prior-margin choices rather than physical constants.

free parameters (3)
  • gamma (penalty weight) = grid-searched, values not reported in text
    The weight of the penalty term in Eq. (6) is chosen by grid search per setting (Section 3.1.7), so the reported results depend on this choice.
  • prior margin percentages = 10, 25, 50, 75 percent
    The allowed uncertainty around the size prior (Section 3.1.3) is set by hand; results vary with this choice, and the main ConstraintAdap uses ±10%.
  • source median size tau_bar_S = not given numerically
    Used as the prior in ConstraintLit and in fake labels for the auxiliary regressor (Eqs. 8, 14); it is a statistic computed from source training data.
assumptions (4)
  • domain assumption Region size is approximately domain-invariant
    The entire method assumes the size of the target structure can be bounded using source-derived or anatomical values; if sizes shift systematically across domains, constraints mislead the network (Section 2.1, Eq. 3).
  • standard math Penalty-based soft constraints substitute adequately for hard constraints
    The paper relies on the standard penalty method from constrained optimization (Bertsekas, cited as [49]) to replace Eq. (1) with Eq. (6); there is no guarantee the constraints are satisfied exactly.
  • domain assumption The auxiliary size regressor transfers across domains sufficiently
    The size regressor R trained on source (and weakly labeled target) must estimate target sizes within the allowed margins; the paper shows this holds empirically but notes it as the main limitation (Section 4).
  • domain assumption Gradient descent can optimize the penalized loss
    Standard deep learning training assumptions; the penalty is differentiable, but no analysis of convergence or local minima is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Constrained domain adaptation for Image segmentation." pith.science (2026). https://pith.science/paper/FFNIYB2O

@misc{pith2026190802996,
  author       = {Pith},
  title        = {Pith review of: Constrained domain adaptation for Image segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FFNIYB2O}},
  note         = {Machine review of arXiv:1908.02996}
}
read the original abstract

We propose to adapt segmentation networks with a constrained formulation, which embeds domain-invariant prior knowledge about the segmentation regions. Such knowledge may take the form of simple anatomical information, e.g., structure size or shape, estimated from source samples or known a priori. Our method imposes domain-invariant inequality constraints on the network outputs of unlabeled target samples. It implicitly matches prediction statistics between target and source domains with permitted uncertainty of prior knowledge. We address our constrained problem with a differentiable penalty, fully suited for standard stochastic gradient descent approaches, removing the need for computationally expensive Lagrangian optimization with dual projections. Unlike current two-step adversarial training, our formulation is based on a single loss in a single network, which simplifies adaptation by avoiding extra adversarial steps, while improving convergence and quality of training. The comparison of our approach with state-of-the-art adversarial methods reveals substantially better performance on the challenging task of adapting spine segmentation across different MRI modalities. Our results also show a robustness to imprecision of size priors, approaching the accuracy of a fully supervised model trained directly in a target domain.Our method can be readily used for various constraints and segmentation problems.

Figures

Figures reproduced from arXiv: 1908.02996 by the authors.

Figure 1
Figure 1. Visualization of severe domain shifts between source and target modalities in two applications. Top: [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. (Left) Pipeline of the proposed CDA framework. The prior knowledge can be learned and predicted with an auxiliary regression network. (Right) The training process of the auxiliary regression network. imposes inequality constraints on the network predictions of unlabeled or weakly labeled target samples, thereby matching implicitly the prediction statistics of the target and source domains, with permitted un￾certaint… view at source ↗
Figure 3
Figure 3. Normalized histograms of the relative size difference between ground truth size and size estimated [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Example of the segmentations achieved by our constrained formulation ( [PITH_FULL_IMAGE:figures/full_fig_p014_4.png]
Figure 5
Figure 5. Figure 5: Examples of the segmentations achieved by our constrained formulation ( [PITH_FULL_IMAGE:figures/full_fig_p015_5.png]
Figure 6
Figure 6. Figure 6: Example of the segmentations achieved on spine images by our constrained formulation with tighter [PITH_FULL_IMAGE:figures/full_fig_p016_6.png]
Figure 7
Figure 7. Figure 7: Examples of the segmentations achieved on cardiac CT images by our constrained formulation [PITH_FULL_IMAGE:figures/full_fig_p017_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

62 extracted references · 60 canonical work pages

  1. [19]

    Learning to adapt structured output space for semantic segmentation,

    Y. Tsai, W. Hung, S. Schulter, K. Sohn, M. Yang, and M. Chandraker, “Learning to adapt structured output space for semantic segmentation,” in Proc. IEEE CVPR , 2018, pp. 7472–7481. 2, 3, 4, 7, 8, 10, 11, 12, 13, 14, 15

    work page 2018

  2. [38]

    A curriculum domain adaptation for semantic segmentation of urban scenes,

    Y. Zhang, P. David, and B. Gong, “A curriculum domain adaptation for semantic segmentation of urban scenes,” IEEE Trans. Pattern Anal. Mach. Intell. , 2019.3, 4, 5, 7, 8, 10, 11, 12, 14, 15

    work page 2019

  3. [1]

    A survey on deep learning in medical image analysis,

    G. Litjens, T. Kooi, B. E. Bejnordi, A. A. A. Setio, F. Ciompi, M. Ghafoorian, J. A. van der Laak, B. van Ginneken, and C. I. S´ anchez, “A survey on deep learning in medical image analysis,” Med. Im. Analysis, vol. 42, pp. 60–88, 2017. 1

    work page 2017

  4. [2]

    3D fully convolutional networks for subcortical segmentation in MRI: A large-scale study,

    J. Dolz, C. Desrosiers, and I. B. Ayed, “3D fully convolutional networks for subcortical segmentation in MRI: A large-scale study,” NeuroImage, vol. 170, pp. 456–470, 2018. 1, 15 16 Figure 7: Examples of the segmentations achieved on cardiac CT images by our constrained formulation with tighter to looser constraints (Constraint10 being the tightest), i.e....

    work page 2018

  5. [3]

    Boxsup: Exploiting bounding boxes to supervise convolutional networks for semantic segmentation,

    J. Dai, K. He, and J. Sun, “Boxsup: Exploiting bounding boxes to supervise convolutional networks for semantic segmentation,” Proc. IEEE ICCV , pp. 1635–1643, 2015. 1, 16

    work page 2015

  6. [4]

    From image-level to pixel-level labeling with convolutional net- works,

    P. H. O. Pinheiro and R. Collobert, “From image-level to pixel-level labeling with convolutional net- works,” Proc. IEEE CVPR , pp. 1713–1721, 2014. 1

    work page 2014

  7. [5]

    Stc: A simple to complex framework for weakly-supervised semantic segmentation,

    Y. Wei, X. Liang, Y. Chen, X. Shen, M.-M. Cheng, Y. Zhao, and S. Yan, “Stc: A simple to complex framework for weakly-supervised semantic segmentation,” IEEE Trans. Pattern Anal. Mach. Intell. , vol. 39, pp. 2314–2320, 2016. 1

    work page 2016

  8. [6]

    Deepcut: Object segmentation from bounding box annotations using convolutional neural networks,

    M. Rajchl, M. C. H. Lee, O. Oktay, K. Kamnitsas, J. Passerat-Palmbach, W. Bai, B. Kainz, and D. Rueckert, “Deepcut: Object segmentation from bounding box annotations using convolutional neural networks,” IEEE Trans. Med. Imag. , vol. 36, pp. 674–683, 2016. 1, 16

    work page 2016

Show all 62 references
  1. [7]

    Constrained-CNN losses for weakly supervised segmentation,

    H. Kervadec, J. Dolz, M. Tang, E. Granger, Y. Boykov, and I. B. Ayed, “Constrained-CNN losses for weakly supervised segmentation,” Med. Im. Analysis , vol. 54, pp. 88–99, 2019. 1, 3, 6

    2019

  2. [8]

    Constrained deep weak supervision for histopathology image segmentation,

    Z. Jia, X. Huang, E. I.-C. Chang, and Y. Xu, “Constrained deep weak supervision for histopathology image segmentation,” IEEE Trans. Med. Imag. , vol. 36, pp. 2376–2388, 2017. 1, 3, 5, 6

    2017

  3. [9]

    Semi-supervised learning for network-based cardiac mr image segmentation,

    W. Bai, O. Oktay, M. Sinclair, H. Suzuki, M. Rajchl, G. Tarroni, B. Glocker, A. King, P. M. Matthews, and D. Rueckert, “Semi-supervised learning for network-based cardiac mr image segmentation,” inProc. Int. Conf. MICCAI. Cham: Springer, 2017, pp. 253–260. 1

    2017

  4. [10]

    Semi-supervised 3d abdominal multi-organ segmentation via deep multi-planar co-training,

    Y. Zhou, Y. Wang, P. Tang, S. Bai, W. Shen, E. Fishman, and A. Yuille, “Semi-supervised 3d abdominal multi-organ segmentation via deep multi-planar co-training,” in Proc. IEEE WACV, 2019, pp. 121–140. 1

    2019

  5. [11]

    Normalized cut loss for weakly- supervised cnn segmentation,

    M. Tang, A. Djelouah, F. Perazzi, Y. Boykov, and C. Schroers, “Normalized cut loss for weakly- supervised cnn segmentation,” Proc. IEEE CVPR , pp. 1818–1827, 2018. 1, 3 17

    2018

  6. [12]

    Deep domain confusion: Maximizing for domain invariance,

    E. Tzeng, J. Hoffman, N. Zhang, K. Saenko, and T. Darrell, “Deep domain confusion: Maximizing for domain invariance,” ArXiv, vol. abs/1412.3474, 2014. 2

    2014 arXiv

  7. [13]

    Generative adversarial nets,

    I. J. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y. Bengio, “Generative adversarial nets,” in Proc. Int. Conf. NIPS . MIT Press, 2014, pp. 2672–2680. 2

    2014

  8. [14]

    CyCADA: Cycle-consistent adversarial domain adaptation,

    J. Hoffman, E. Tzeng, T. Park, J.-Y. Zhu, P. Isola, K. Saenko, A. Efros, and T. Darrell, “CyCADA: Cycle-consistent adversarial domain adaptation,” in Proc. Int. Conf. on Machine Learning , vol. 80. PMLR, 2018, pp. 1989–1998. 2, 3

    2018

  9. [15]

    Unsupervised pixel-level domain adaptation with generative adversarial networks,

    K. Bousmalis, N. Silberman, D. Dohan, D. Erhan, and D. Krishnan, “Unsupervised pixel-level domain adaptation with generative adversarial networks,” in Proc. IEEE CVPR , 2017, pp. 95–104. 2, 3

    2017

  10. [16]

    From source to target and back: Symmetric bi-directional adaptive gan,

    P. Russo, F. M. Carlucci, T. Tommasi, and B. Caputo, “From source to target and back: Symmetric bi-directional adaptive gan,” in Proc. IEEE CVPR , 2018, pp. 8099–8108. 2, 3

    2018

  11. [17]

    Generate to adapt: Aligning domains using generative adversarial networks,

    S. Sankaranarayanan, Y. Balaji, C. D. Castillo, and R. Chellappa, “Generate to adapt: Aligning domains using generative adversarial networks,” in Proc. IEEE CVPR , 2018, pp. 8503–8512. 2, 3

    2018

  12. [18]

    Semantic-aware generative adversarial nets for unsuper- vised domain adaptation in chest x-ray segmentation,

    C. Chen, Q. Dou, H. Chen, and P.-A. Heng, “Semantic-aware generative adversarial nets for unsuper- vised domain adaptation in chest x-ray segmentation,” in Machine Learning in Med. Imag. Cham: Springer, 2018, pp. 143–151. 2, 3

    2018

  13. [20]

    Adversarial discriminative domain adaptation,

    E. Tzeng, J. Hoffman, K. Saenko, and T. Darrell, “Adversarial discriminative domain adaptation,” in Proc. IEEE CVPR , 2017, pp. 2962–2971. 2, 3, 8, 11, 14

    2017

  14. [21]

    Not-so-supervised: A survey of semi-supervised, multi-instance, and transfer learning in medical image analysis,

    V. Cheplygina, M. de Bruijne, and J. P. W. Pluim, “Not-so-supervised: A survey of semi-supervised, multi-instance, and transfer learning in medical image analysis,”Med. Im. Analysis, vol. 54, pp. 280–296, 2019.2

    2019

  15. [22]

    Unsupervised domain adaptation in brain lesion segmentation with adversarial networks,

    K. Kamnitsas, C. F. Baumgartner, C. Ledig, V. F. J. Newcombe, J. P. Simpson, A. D. Kane, D. K. Menon, A. V. Nori, A. Criminisi, D. Rueckert, and B. Glocker, “Unsupervised domain adaptation in brain lesion segmentation with adversarial networks,” in Proc. Int. Conf. IPMI , vol....

    2017

  16. [23]

    Adversarial domain adaptation for classification of prostate histopathology whole-slide images,

    J. Ren, I. Hacihaliloglu, E. A. Singer, D. J. Foran, and X. Qi, “Adversarial domain adaptation for classification of prostate histopathology whole-slide images,” in Proc. Int. Conf. MICCAI. , vol. 11071. Springer, 2018, pp. 201–209. 2

    2018

  17. [24]

    Task driven generative modeling for unsupervised domain adaptation: Application to x-ray image segmentation,

    Y. Zhang, S. Miao, T. Mansi, and R. Liao, “Task driven generative modeling for unsupervised domain adaptation: Application to x-ray image segmentation,” in Proc. Int. Conf. MICCAI. , 2018, pp. 599–607. 2

    2018

  18. [25]

    A dirtt-t approach to unsupervised domain adaptation,

    R. Shu, H. H. Bui, H. Narui, and S. Ermon, “A dirtt-t approach to unsupervised domain adaptation,” in Int. Conf. ICLR , 2018.2, 3

    2018

  19. [26]

    Deep reconstruction-classification net- works for unsupervised domain adaptation,

    M. Ghifary, W. B. Kleijn, M. Zhang, D. Balduzzi, and W. Li, “Deep reconstruction-classification net- works for unsupervised domain adaptation,” in Proc. ECCV. Cham: Springer, 2016, pp. 597–613. 3

    2016

  20. [27]

    Learning transferable features with deep adaptation networks,

    M. Long, Y. Cao, J. Wang, and M. I. Jordan, “Learning transferable features with deep adaptation networks,” in Proc. Int. Conf. on Machine Learning . JMLR.org, 2015, pp. 97–105. 3

    2015

  21. [28]

    Unsupervised domain adaptation with residual transfer networks,

    M. Long, H. Zhu, J. Wang, and M. I. Jordan, “Unsupervised domain adaptation with residual transfer networks,” in Proc. Int. Conf. NIPS , 2016, pp. 136–144. 3 18

    2016

  22. [29]

    Unsupervised detection of lesions in brain MRI using constrained adver- sarial auto-encoders,

    X. Chen and E. Konukoglu, “Unsupervised detection of lesions in brain MRI using constrained adver- sarial auto-encoders,” in MIDL, 2018.3

    2018

  23. [30]

    Transfer learning improves supervised image segmentation across imaging protocols,

    A. Opbroek, M. Ikram, M. Vernooij, and M. de Bruijne, “Transfer learning improves supervised image segmentation across imaging protocols,” IEEE Trans. Med. Imag. , vol. 34, pp. 1018–1030, 2014. 3

    2014

  24. [31]

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

    N. Tajbakhsh, L. Jeyaseelan, Q. Li, J. N. Chiang, Z. Wu, and X. Ding, “Embracing imperfect datasets: A review of deep learning solutions for medical image segmentation,” Medical Image Analysis , vol. 63, p. 101693, 2020.3

    2020

  25. [32]

    A novel domain adaptation framework for medical image segmentation,

    A. Gholami, S. Subramanian, V. Shenoy, N. Himthani, X. Yue, S. Zhao, P. Jin, G. Biros, and K. Keutzer, “A novel domain adaptation framework for medical image segmentation,” in Proc. Int. Conf. MICCAI. Brainlesion Workshop, vol. 11384, 2018, pp. 289–298. 3

    2018

  26. [33]

    Domain adaptation for biomedical image segmentation using adver- sarial training,

    M. Javanmardi and T. Tasdizen, “Domain adaptation for biomedical image segmentation using adver- sarial training,” in Proc. IEEE ISBI , 2018, pp. 554–558. 3

    2018

  27. [34]

    Supervised segmentation of un-annotated retinal fundus images by synthesis,

    H. Zhao et al. , “Supervised segmentation of un-annotated retinal fundus images by synthesis,” IEEE Trans. Med. Imag. , vol. 38, no. 1, pp. 46–56, 2019. 3

    2019

  28. [35]

    Generative adver- sarial network for segmentation of motion affected neonatal brain mri,

    N. Khalili, E. Turk, M. Zreik, M. A. Viergever, M. J. N. L. Benders, and I. Iˇ sgum, “Generative adver- sarial network for segmentation of motion affected neonatal brain mri,” in Proc. Int. Conf. MICCAI. Cham:Springer, 2019, pp. 320–328. 3

    2019

  29. [36]

    Road: Reality oriented adaptation for semantic segmentation of urban scenes,

    Y. Chen, W. Li, and L. Van Gool, “Road: Reality oriented adaptation for semantic segmentation of urban scenes,” in Proc. IEEE CVPR , 2018, pp. 7892–7901. 3

    2018

  30. [37]

    Conditional generative adversarial network for structured domain adaptation,

    W. Hong, Z. Wang, M. Yang, and J. Yuan, “Conditional generative adversarial network for structured domain adaptation,” in Proc. IEEE CVPR , 2018, pp. 1335–1344. 3

    2018

  31. [39]

    Unsupervised domain adaptation for semantic segmentation via class-balanced self-training,

    Y. Zou, Z. Yu, B. V. K. V. Kumar, and J. Wang, “Unsupervised domain adaptation for semantic segmentation via class-balanced self-training,” in Proc. ECCV., 2018, pp. 289–305. 3

    2018

  32. [40]

    Curriculum semi-supervised segmentation,

    H. Kervadec, J. Dolz, ´E. Granger, and I. Ben Ayed, “Curriculum semi-supervised segmentation,” in Proc. Int. Conf. MICCAI. Cham: Springer, 2019, pp. 568–576. 3

    2019

  33. [41]

    On regularized losses for weakly-supervised CNN segmentation,

    M. Tang, F. Perazzi, A. Djelouah, I. Ben Ayed, C. Schroers, and Y. Boykov, “On regularized losses for weakly-supervised CNN segmentation,” in Proc. ECCV., vol. 11220, 2018, pp. 524–540. 3

    2018

  34. [42]

    Constrained convolutional neural networks for weakly supervised segmentation,

    D. Pathak, P. Kr¨ ahenb¨ uhl, and T. Darrell, “Constrained convolutional neural networks for weakly supervised segmentation,” in Proc. IEEE ICCV , 2015, pp. 1796–1804. 3, 6

    2015

  35. [43]

    Pnp-adanet: Plug-and- play adversarial domain adaptation network at unpaired cross-modality cardiac segmentation,

    Q. Dou, C. Ouyang, C. Chen, H. Chen, B. Glocker, X. Zhuang, and P. Heng, “Pnp-adanet: Plug-and- play adversarial domain adaptation network at unpaired cross-modality cardiac segmentation,” IEEE Access, vol. 7, pp. 99 065–99 076, 2019.4, 7, 8, 11

    2019

  36. [44]

    Constrained domain adaptation for segmentation,

    M. Bateson, J. Dolz, H. Kervadec, H. Lombaert, and I. B. Ayed, “Constrained domain adaptation for segmentation,” in Proc. Int. Conf. MICCAI. , vol. 11765, 2019, pp. 326–334. 4, 7

    2019

  37. [45]

    Volumetric bias in segmentation and reconstruction: Secrets and solutions,

    Y. Boykov, H. Isack, C. Olsson, and I. Ben Ayed, “Volumetric bias in segmentation and reconstruction: Secrets and solutions,” in Proc. IEEE ICCV , 2015, pp. 1769–1777. 5

    2015

  38. [46]

    A convex framework for image segmentation with moment constraints,

    M. Klodt and D. Cremers, “A convex framework for image segmentation with moment constraints,” in Proc. IEEE ICCV , 2011, pp. 2236–2243. 5, 16

    2011

  39. [47]

    Kernel density estimation and intrinsic alignment for shape priors in level set segmentation,

    D. Cremers, S. J. Osher, and S. Soatto, “Kernel density estimation and intrinsic alignment for shape priors in level set segmentation,” Int. Journ. of Comp. Vis. (IJCV) , vol. 69, no. 3, pp. 335–351, 2006. 5 19

    2006

  40. [48]

    Incorporating prior knowledge in medical image segmentation: a survey,

    M. S. Nosrati and G. Hamarneh, “Incorporating prior knowledge in medical image segmentation: a survey,” ArXiv, vol. abs/1607.01092, 2016. 5

    2016 arXiv

  41. [49]

    D. P. Bertsekas, Nonlinear Programming. Athena Scientific, Belmont, MA, 1995. 6

    1995

  42. [50]

    Imposing Hard Constraints on Deep Networks: Promises and Limitations,

    P. M´ arquez-Neilaet al., “Imposing Hard Constraints on Deep Networks: Promises and Limitations,” in Proc. IEEE CVPR Workshop on Negative Results , 2017.6

    2017

  43. [51]

    Learning to play in a day: Faster deep reinforcement learning by optimality tightening,

    F. S. He, Y. Liu, A. G. Schwing, and J. Peng, “Learning to play in a day: Faster deep reinforcement learning by optimality tightening,” in Int. Conf. ICLR , 2017.6

    2017

  44. [52]

    A morphometric study of human lumbar and selected thoracic vertebrae,

    J. L. Berry, J. M. Moran, W. S. Berg, and A. D. Steffee, “A morphometric study of human lumbar and selected thoracic vertebrae,” Spine, vol. 12, no. 4, pp. 362–367, 1987. 6

    1987

  45. [53]

    Evaluation of algorithms for multi-modality whole heart segmentation: An open-access grand challenge,

    X. Zhuang, L. Li, C. Payer, D. ˇStern, M. Urschler, M. P. Heinrich, J. Oster, C. Wang, ¨Orjan Smedby, C. Bian, X. Yang, P.-A. Heng, A. Mortazi, U. Bagci, G. Yang, C. Sun, G. Galisot, J.-Y. Ramel, T. Brouard, Q. Tong, W. Si, X. Liao, G. Zeng, Z. Shi, G. Zheng, C. Wang, T. MacGi...

    2019

  46. [54]

    Multi-adversarial domain adaptation,

    Z. Pei, Z. Cao, M. Long, and J. Wang, “Multi-adversarial domain adaptation,” in AAAI, 2018.8

    2018

  47. [55]

    No more discrimination: Cross city adaptation of road scene segmenters,

    Y.-H. Chen, W.-Y. Chen, Y.-T. Chen, B.-C. Tsai, Y.-C. F. Wang, and M. Sun, “No more discrimination: Cross city adaptation of road scene segmenters,” Proc. IEEE ICCV , pp. 2011–2020, 2017. 8

    2011

  48. [56]

    Domain-adversarial training of neural networks,

    Y. Ganin, E. Ustinova, H. Ajakan, P. Germain, H. Larochelle, F. Laviolette, M. Marchand, and V. Lem- pitsky, “Domain-adversarial training of neural networks,” J. Mach. Learn. Res. , vol. 17, no. 1, p. 2096–2030, Jan. 2016.8, 11, 14

    2016

  49. [57]

    Unpaired image-to-image translation using cycle-consistent adversarial networks,

    J. Zhu, T. Park, P. Isola, and A. A. Efros, “Unpaired image-to-image translation using cycle-consistent adversarial networks,” in 2017 IEEE International Conference on Computer Vision (ICCV) , 2017, pp. 2242–2251.8, 11, 14

    2017

  50. [58]

    Enet: A deep neural network architecture for real-time semantic segmentation,

    A. Paszke, A. Chaurasia, S. Kim, and E. Culurciello, “Enet: A deep neural network architecture for real-time semantic segmentation,” arXiv preprint:1606.02147, 2016.10

    2016 arXiv

  51. [59]

    Generalised dice overlap as a deep learning loss function for highly unbalanced segmentations,

    C. H. Sudre, W. Li, T. Vercauteren, S. Ourselin, and M. Jorge Cardoso, “Generalised dice overlap as a deep learning loss function for highly unbalanced segmentations,” Lecture Notes in Computer Science , p. 240–248, 2017.10

    2017

  52. [60]

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

    F. Milletari, N. Navab, and S. Ahmadi, “V-net: Fully convolutional neural networks for volumetric medical image segmentation,” in 2016 Fourth International Conference on 3D Vision (3DV) , 2016, pp. 565–571.10

    2016

  53. [61]

    Aggregated residual transformations for deep neural networks,

    S. Xie, R. Girshick, P. Dollar, Z. Tu, and K. He, “Aggregated residual transformations for deep neural networks,” in Proc. IEEE CVPR , 2017, pp. 5987–5995. 10

    2017

  54. [62]

    Oct skin image enhancement through attenuation compen- sation,

    A. Hojjatoleslami and M. R. N. Avanaki, “Oct skin image enhancement through attenuation compen- sation,” Appl. Opt. , vol. 51, no. 21, pp. 4927–4935, 2012. 14 20

    2012

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.