Pith. sign in

REVIEW 4 major objections 4 minor 118 references

MorphGen argues that aligning image representations with nuclear segmentation masks via supervised contrastive learning makes cancer classifiers generalize across institutions.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

MorphGen uses supervised contrastive learning to align histopathology images with nuclear masks and applies SWA, reporting improved out-of-domain cancer classification accuracy on CAMELYON17, BCSS, and OCELOT.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection A useful extension of SFL with broad evaluation, but the mask-alignment loss is never isolated from SWA, so the central attribution claim is unproven. the 4 major comments →

arxiv 2509.00311 v1 pith:AXSIR7IO submitted 2025-08-30 cs.CV

MorphGen: Morphology-Guided Representation Learning for Robust Single-Domain Generalization in Histopathological Cancer Classification

Hikmat Khan , Syed Farhan Alam Zaidi , Pir Masoom Shah , Kiruthika Balakrishnan , Rabia Khan , Muhammad Waqas , Jia Wu This is my paper
classification cs.CV
keywords single-domain generalizationhistopathologynuclear morphologysupervised contrastive learningdomain shiftcancer classificationstochastic weight averagingwhole slide images
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

MorphGen sets out to show that histopathology models generalize across hospitals, scanners, and stains more reliably when training explicitly ties image representations to nuclear morphology. It does this by aligning latent embeddings of tissue patches with embeddings of their nuclear segmentation masks through supervised contrastive learning, then steering training toward flatter loss minima with stochastic weight averaging. On the paper's own experiments, models trained on one CAMELYON17 center reach about 95% average accuracy on unseen centers, and higher accuracy than all tested baselines on the BCSS and OCELOT datasets, including organs never seen during training. If the claim holds, single-institution training could produce deployable diagnostic models without target-domain data, stain normalization, or synthetic augmentation.

Core claim

The paper's central claim is that morphology-guided representation alignment—rather than stain normalization, augmentation alone, or Euclidean-distance mask regularization—is what makes learned cancer classifiers domain-invariant. Concretely, MorphGen trains a shared ResNet encoder on original patches, aggressively augmented patches, and nuclear segmentation masks; the mask embedding acts as an anchor, positive pairs are the matching patch and its augmentations, and other patches in the batch are repelled in the latent space. The authors report average out-of-domain accuracies of 95.4% (95.6% with augmentation) on CAMELYON17, 77.5% (77.4%) on BCSS, and 72.0% (72.2%) on OCELOT, consistently a

What carries the argument

The load-bearing object is the morphology-guided supervised contrastive loss. Its loss has two terms: an attraction term that drives the embedding of a nuclear segmentation mask toward the embeddings of the corresponding patch and its augmentations, and a repulsion term, with a similarity margin eta, that pushes every other patch in the batch away from the mask anchor. Because the same encoder processes masks and images, the model is forced to find image features that predict mask structure—nuclear size, shape, contour, and spatial arrangement—while discarding staining and scanner-specific appearance. Stochastic weight averaging is then applied to the weights, averaging late-training checkpo

Load-bearing premise

The load-bearing premise is that the nuclear segmentation mask actually contains the diagnostic cues the paper credits it with, including chromatin texture and hyperchromasia; a binary mask stores shape and location, so texture and staining-density cues are not present in the anchor signal, and if those cues are essential to the claimed gains, the alignment mechanism cannot be the source of them.

What would settle it

Train MorphGen exactly as described but replace each nuclear mask with a randomly permuted or random-blob binary mask matched for foreground density. If out-of-domain accuracy on CAMELYON17, BCSS, and OCELOT stays at the reported level, the morphology-specific content of the masks is not what drives generalization; the gains would instead come from contrastive regularization, augmentation, and SWA. Alternatively, keep only nuclear centroid positions and erase nuclear boundaries: if accuracy drops sharply, contour and shape information is the load-bearing cue.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • A model trained on a single institution could be deployed on slides from unseen scanners and staining protocols without collecting target-domain data.
  • Nuclear segmentation masks are only needed during training, so the deployed system is a standard, fast patch classifier.
  • The approach appears to transfer across organs, from lymph node training data to bladder, endometrium, kidney, prostate, and stomach test sets, which would reduce the need for organ-specific retraining.
  • Because the learned features are morphology-based, they are more interpretable to pathologists than color-based or texture-artifact features.
  • The robustness results suggest morphology-guided alignment and flat-minima optimization could be combined with existing augmentation methods for further gains.

Where Pith is reading between the lines

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

  • The paper claims to be the first to directly embed segmentation masks into representation learning, but its own closest baseline, SFL [27], aligns image and mask representations with Euclidean-distance regularization; the actual novelty is contrastive alignment, not mask embedding per se.
  • A binary nuclei mask carries shape, size, and spatial layout, but not chromatin texture or hyperchromasia, which live in image intensities; the paper's assertion that mask alignment teaches chromatin texture overstates what the mask signal can provide.
  • A sharper ablation would train MorphGen with the same contrastive loss and SWA but with masks replaced by random binary noise matched in foreground fraction; if out-of-domain accuracy stays high, the morphological content of the masks is not the driver.
  • The reported gains could be tested prospectively on an external multi-institutional cohort with clinically relevant endpoints such as grading or subtyping rather than binary tumor-versus-normal classification.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper proposes MorphGen, a training objective for single-domain generalization in histopathological cancer classification. MorphGen combines a supervised contrastive alignment between embeddings of histopathology patches (and their augmentations) and embeddings of nuclear segmentation masks, a binary cross-entropy classification loss, and stochastic weight averaging (SWA). The method is evaluated by training on each CAMELYON17 center and testing on the other centers, BCSS, and OCELOT, including organ-specific subsets, plus robustness to corruptions and PGD adversarial perturbations. The central claim is that morphology-guided representation alignment yields domain-invariant features that consistently outperform normalization-, augmentation-, and representation-alignment baselines.

Significance. If validated, the idea of using nuclear segmentation masks as anchors for contrastive representation learning is useful for the histopathology domain-generalization community, and the paper provides a clearly specified loss (Eqs. 1–4), extensive comparisons across three datasets and six organs, public code/data, and interpretability analyses. These are genuine strengths. However, the central causal claim is currently not established because SWA is enabled for MorphGen but not for any baseline, and no ablation isolates the contribution of the morphology-guided contrastive loss. The paper also overstates the information content of binary masks and the consistency of its empirical wins. With the requested ablations and more careful claims, the contribution would be a solid empirical study.

major comments (4)
  1. [§5.4 and Eq. (4)] MorphGen's reported protocol always includes SWA (Section 5.4: 'SWA Enabled'), while the total loss in Eq. (4) is L_align + L_bce. The comparison in Table 3 and Tables 4–5 is therefore MorphGen+SWA against baselines without SWA. There is no ablation that removes L_align while keeping SWA, and no baseline receives SWA. Thus the headline attribution of the OOD gains to morphology-guided contrastive learning is untested; the gains could come entirely from SWA or the aggressive augmentation of Section 5.2. Please provide an ablation matrix: (i) L_bce + augmentation, (ii) +SWA, (iii) L_align + L_bce without SWA, (iv) full method, and also add SWA to the strongest baselines.
  2. [§4.2.1] The text states that mask alignment 'ensures that the model captures features essential to nuclear size, contour, chromatin texture, and local spatial patterns.' A binary nuclear mask encodes localization and shape, but chromatin texture and staining density live in the image intensity, not in the mask. The biological grounding therefore overstates what the mask signal can provide. Since the masks are derived from the same H&E images, the alignment may also be partly circular: image embeddings are pulled toward the outputs of a segmentation model trained on those images. Please temper the mechanism claim, or provide direct evidence (e.g., representation probing) that the learned features encode texture beyond shape and spatial cues.
  3. [§6.3, Figures 6 and 8] The corruption and adversarial robustness claims are supported only by plotted means, with no error bars or significance tests, despite Section 5.4 stating that experiments use three seeds. Given the small average differences in Table 3, the caption 'MorphGen consistently maintains higher accuracy' is not supported; Section 6.3.1 itself says 'five out of the eight distortion types' at severity 3. Please report per-seed variability, confidence intervals, or statistical tests, and reconcile the text with the figure captions.
  4. [§6.1 and Tables 4–5] The abstract and introduction claim that MorphGen 'consistently outperforms' is contradicted by the paper's own tables. On BCSS, Ours-Aug averages 77.4% while SFL-Aug averages 78.8% (Table 3). On bladder, prostate, and stomach, Ours-Aug trails SFL-Aug by 1.4, 0.3, and 0.9 points respectively (Tables 4–5). Please replace 'consistently' with a precise characterization, report paired tests or confidence intervals, and discuss the regimes in which the method wins (e.g., without augmentation) versus loses.
minor comments (4)
  1. [Section 2] The claim 'this is the first method to directly embed segmentation masks into representation learning' is contradicted by the cited SFL [27], which is described in Section 3.3 as 'aligning image and mask representations using Euclidean distance-based regularization.' This novelty claim should be revised to emphasize the contrastive formulation rather than direct mask embedding.
  2. [Eq. (1)] The subscripts/superscripts in L_attract and L_repel are inconsistent between the displayed equation and the surrounding text. Please align the notation.
  3. [Figures 4–6] Figure captions say 'level 1' and 'level 2', while the text refers to 'level 0 (no distortion) and level 3 (severe distortion)' and Figure 6 is said to be at severity level 3. Please make the corruption levels consistent.
  4. [References and typos] The integrated-gradients attribution in Section 6.4 cites [118] (Sundararajan and Najmi, 'The many Shapley values for model explanation'), which is not the standard integrated-gradients reference. Also, the template footer says 'Preprint submitted to Nuclear Physics B', which is clearly the wrong journal template, and there is a typo 'enahances' in Section 4.2.3.

Circularity Check

0 steps flagged

No circularity found: MorphGen is an empirical training-objective paper evaluated on held-out external datasets; the SWA attribution gap and the SFL novelty contradiction are correctness/prior-art concerns, not circular reductions.

full rationale

The paper does not derive a prediction from a fitted parameter or from a self-referential definition. Its central claim—that morphology-guided contrastive alignment improves out-of-domain generalization—is an empirical training result, supported by evaluation on datasets external to training (BCSS and OCELOT), which the paper states are 'exclusively reserved for final evaluation to provide an unbiased measure of generalization across institutions' (Section 5.4). The total loss (Eq. 4) is a composite of a supervised contrastive alignment term and a BCE classification term; neither term is constructed from the test-domain labels, and no OOD accuracy is forced by the equations. The nuclear masks are training-time targets derived from the input patches; using an input-derived auxiliary target is a legitimate self-supervised/auxiliary objective, not a circular prediction. The paper's self-citations (e.g., refs. [112]–[116] for the adversarial attack protocol, and [7], [10], [19] in the introduction) are incidental or methodological citations, not load-bearing justifications of the morphological-invariance premise. Two concerns raised by a skeptical reading are real but are not circularity: (1) SWA is enabled for MorphGen and not reported for baselines, so the attribution of gains to the mask-alignment loss is confounded—this is an experimental-control/attribution issue, not a definitional reduction; and (2) the claim that this is 'the first method to directly embed segmentation masks into representation learning' is contradicted by the paper's own description of SFL [27], which 'proposed aligning image and mask representations using Euclidean distance-based regularization'—this is a novelty/prior-art inconsistency, not a circular derivation. Accordingly, no circular step can be exhibited with the required specificity, and the score is 0.

Axiom & Free-Parameter Ledger

4 free parameters · 3 axioms · 0 invented entities

No new physical or biological entities are postulated. The central assumptions are the domain-invariance of nuclear morphology, the flat-minima benefit of SWA, and mask quality. The free parameters are training hyperparameters whose values matter for the reported gains but are mostly undisclosed.

free parameters (4)
  • repulsion weight lambda = not reported
    Scales the L_repel term in Eq. 1; no value or sensitivity analysis is given.
  • similarity margin eta = not reported
    Caps negative-pair similarity in Eq. 1; absent from Table 2 so a reproducing group must guess it.
  • augmentation ranges = rescaling 0-20%, aspect 0-10%, rotation 0-360deg, brightness 0-50%, hue 0-10%, contrast 0-70%, saturation 0-30%
    Aggressive augmentation is central to the OOD gains but the ranges are hand-chosen with no ablation.
  • SWA start epoch = 25
    Chosen ad hoc; no ablation on the SWA schedule is reported.
axioms (3)
  • domain assumption Nuclear morphology and spatial organization are domain-invariant across staining and scanner conditions.
    This is the biological hypothesis motivating the method, stated in the Introduction and Section 2, and not proven within the paper.
  • domain assumption A flat-minima solution generalizes better out-of-distribution, and SWA achieves flatter minima.
    Borrowed from Izmailov et al. [34] and SWAD [38]; no verification on histopathology is provided.
  • domain assumption The nuclear segmentation masks used in training are of sufficient quality.
    The method depends on HoVerNet or similar masks, but the paper does not describe how masks were generated for CAMELYON17, BCSS, or OCELOT training patches.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of MorphGen: Morphology-Guided Representation Learning for Robust Single-Domain Generalization in Histopathological Cancer Classification." pith.science (2026). https://pith.science/paper/AXSIR7IO

@misc{pith2026250900311,
  author       = {Pith},
  title        = {Pith review of: MorphGen: Morphology-Guided Representation Learning for Robust Single-Domain Generalization in Histopathological Cancer Classification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AXSIR7IO}},
  note         = {Machine review of arXiv:2509.00311}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Domain generalization in computational histopathology is hindered by heterogeneity in whole slide images (WSIs), caused by variations in tissue preparation, staining, and imaging conditions across institutions. Unlike machine learning systems, pathologists rely on domain-invariant morphological cues such as nuclear atypia (enlargement, irregular contours, hyperchromasia, chromatin texture, spatial disorganization), structural atypia (abnormal architecture and gland formation), and overall morphological atypia that remain diagnostic across diverse settings. Motivated by this, we hypothesize that explicitly modeling biologically robust nuclear morphology and spatial organization will enable the learning of cancer representations that are resilient to domain shifts. We propose MorphGen (Morphology-Guided Generalization), a method that integrates histopathology images, augmentations, and nuclear segmentation masks within a supervised contrastive learning framework. By aligning latent representations of images and nuclear masks, MorphGen prioritizes diagnostic features such as nuclear and morphological atypia and spatial organization over staining artifacts and domain-specific features. To further enhance out-of-distribution robustness, we incorporate stochastic weight averaging (SWA), steering optimization toward flatter minima. Attention map analyses revealed that MorphGen primarily relies on nuclear morphology, cellular composition, and spatial cell organization within tumors or normal regions for final classification. Finally, we demonstrate resilience of the learned representations to image corruptions (such as staining artifacts) and adversarial attacks, showcasing not only OOD generalization but also addressing critical vulnerabilities in current deep learning systems for digital pathology. Code, datasets, and trained models are available at: https://github.com/hikmatkhan/MorphGen

Figures

Figures reproduced from arXiv: 2509.00311 by Hikmat Khan, Jia Wu, Kiruthika Balakrishnan, Muhammad Waqas, Pir Masoom Shah, Rabia Khan, Syed Farhan Alam Zaidi.

Figure 1
Figure 1. Figure 1: Illustration of Morphology-guided Representation Alignment for Cancer Classification, utilizing su [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Architecture of MorphGen (Morphology-Guided generalization in histopathological cancer classifica￾tion), illustrating the use of a shared ResNet-based encoder to process histopathology patches and nuclear masks, with a composite loss function combining Morphology-Guided Contrastive Loss and Binary Cross-Entropy (BCE) Loss for robust feature extraction and classification. 4.1. Shared Encoder MorphGen’s enco… view at source ↗
Figure 3
Figure 3. Figure 3: Illustration of stochastic weight averaging (SWA). [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The first row shows the original (uncorrupted) histopathological patches. The second row displays [PITH_FULL_IMAGE:figures/full_fig_p020_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: The first row shows the original (uncorrupted) histopathological patches. The second row displays [PITH_FULL_IMAGE:figures/full_fig_p020_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Robustness to common image corruptions on the CAMELYON17 dataset, evaluated using the cor [PITH_FULL_IMAGE:figures/full_fig_p021_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The first row shows the original histopathological images. The second, third, and fourth rows display [PITH_FULL_IMAGE:figures/full_fig_p022_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Robustness to image perturbations (e.g., staining artifacts) as described in [111, 27]. Results are [PITH_FULL_IMAGE:figures/full_fig_p023_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: The input patches belong to slide-level negative (no-tumor) whole-slide images (WSIs) from the CAME [PITH_FULL_IMAGE:figures/full_fig_p024_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: The input patches belong to slide-level positive (with-tumor) whole-slide images (WSIs) from the [PITH_FULL_IMAGE:figures/full_fig_p025_10.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

118 extracted references · 65 canonical work pages · 4 internal anchors

  1. [1]

    Aggarwal, S

    A. Aggarwal, S. Bharadwaj, G. Corredor, T. Pathak, S. Badve, A. Madabhushi, Artifi- cial intelligence in digital pathology—time for a reality check, Nature Reviews Clinical Oncology (2025) 1–9

    2025

  2. [2]

    McGenity, E

    C. McGenity, E. L. Clarke, C. Jennings, G. Matthews, C. Cartlidge, H. Freduah- Agyemang, D. D. Stocken, D. Treanor, Artificial intelligence in digital pathology: a sys- tematic review and meta-analysis of diagnostic test accuracy, npj Digital Medicine 7 (1) (2024) 114

    2024

  3. [3]

    Datwani, H

    S. Datwani, H. Khan, M. K. K. Niazi, A. V. Parwani, Z. Li, Artificial intelligence in breast pathology: overview and recent updates, Human Pathology (2025) 105819

    2025

  4. [4]

    Tsai, T.-H

    P.-C. Tsai, T.-H. Lee, K.-C. Kuo, F.-Y. Su, T.-L. M. Lee, E. Marostica, T. Ugai, M. Zhao, M. C. Lau, J. P. Väyrynen, et al., Histopathology images predict multi-omics aberrations and prognoses in colorectal cancer patients, Nature communications 14 (1) (2023) 1–13. 23

    2023

  5. [5]

    Javanmard, S

    Z. Javanmard, S. Z. Shahraki, K. Safari, A. Omidi, S. Raoufi, M. Rajabi, M. E. Ak- bari, M. Aria, Artificial intelligence in breast cancer survival prediction: a comprehensive systematic review and meta-analysis, Frontiers in Oncology 14 (2025) 1420328

    2025

  6. [6]

    Piedimonte, M

    S. Piedimonte, M. Mohamed, G. Rosa, B. Gerstl, D. Vicus, Predicting response to treat- ment and survival in advanced ovarian cancer using machine learning and radiomics: A systematic review, Cancers 17 (3) (2025) 336

    2025

  7. [7]

    H. Khan, P. M. Shah, M. A. Shah, S. ul Islam, J. J. Rodrigues, Cascading handcrafted features and convolutional neural network for iot-enabled brain tumor segmentation, Com- puter communications 153 (2020) 196–207

    2020

  8. [8]

    M. M. Jahani, P. Mashayekhi, M. D. Omrani, A. A. Meibody, Efficacy of liquid biopsy for genetic mutations determination in non-small cell lung cancer: a systematic review on literatures, BMC cancer 25 (1) (2025) 433

    2025

  9. [9]

    C. M. Jung, B. Park, H. Son, L. I.-Y. Chung, Y. K. Chae, H&e whole-slide image (wsi) based artificial intelligence (ai) model to detect egfr and alk mutation in non-small cell lung cancer (nsclc), Cancer Research 85 (8_Supplement_1) (2025) 2451–2451

    2025

  10. [10]

    H. Khan, Z. Su, H. Zhang, Y. Wang, B. Ning, S. Wei, H. Guo, Z. Li, M. K. K. Niazi, Predicting neoadjuvant chemotherapy response in triple-negative breast cancer using pre- treatment histopathologic images, Cancers 17 (15) (2025) 2423

    2025

  11. [11]

    Koziarski, B

    M. Koziarski, B. Cyganek, P. Niedziela, B. Olborski, Z. Antosz, M. Żydak, B. Kwolek, P. Wąsowicz, A. Bukała, J. Swadźba, et al., Diagset: a dataset for prostate cancer histopathological image classification, Scientific Reports 14 (1) (2024) 6780

    2024

  12. [12]

    Loménie, C

    N. Loménie, C. Bertrand, R. H. Fick, S. B. Hadj, B. Tayart, C. Tilmant, I. Farré, S. Z. Azdad, S. Dahmani, G. Dequen, et al., Can ai predict epithelial lesion categories via automated analysis of cervical biopsies: The tissuenet challenge?, Journal of Pathology Informatics 13 (2022) 100149

    2022

  13. [13]

    E. D. Shulman, E. M. Campagnolo, R. Lodha, A. Stemmer, T. Cantore, B. Ru, A. Wang, T. Hu, M. Nasrallah, D.-T. Hoang, et al., Path2space: an ai approach for cancer biomarker discovery via histopathology inferred spatial transcriptomics, Cancer Research 85 (8_Sup- plement_1) (2025) 6353–6353

    2025

  14. [14]

    P. M. Shah, H. Zhu, Z. Lu, K. Wang, J. Tang, M. Li, Deepdtagen: a multitask deep learning framework for drug-target affinity prediction and target-aware drugs generation, Nature Communications 16 (1) (2025) 5021

    2025

  15. [15]

    Campanella, N

    G. Campanella, N. Kumar, S. Nanda, S. Singi, E. Fluder, R. Kwan, S. Muehlstedt, N. Pfarr, P. J. Schüffler, I. Häggström, et al., Real-world deployment of a fine-tuned pathology foundation model for lung cancer biomarker detection, Nature Medicine (2025) 1–9

    2025

  16. [16]

    W. Shao, Y. Shi, D. Zhang, J. Zhou, P. Wan, Tumor micro-environment interactions guided graph learning for survival analysis of human cancers from whole-slide pathological images, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 11694–11703. 24

    2024

  17. [17]

    T. Xie, A. Huang, H. Yan, X. Ju, L. Xiang, J. Yuan, Artificial intelligence: illuminating the depths of the tumor microenvironment, Journal of Translational Medicine 22 (1) (2024) 799

    2024

  18. [18]

    Muneer, M

    A. Muneer, M. Waqas, M. B. Saad, E. Showkatian, R. Bandyopadhyay, H. Xu, W. Li, J. Y. Chang, Z. Liao, C. Haymaker, et al., From classical machine learning to emerging foundation models: Review on multimodal data integration for cancer research, arXiv preprint arXiv:2507.09028 (2025)

    Pith/arXiv arXiv 2025

  19. [19]

    Balakrishnan, D

    K. Balakrishnan, D. Velusamy, H. E. Hinkle, Z. Li, K. Ramasamy, H. Khan, S. Ra- maswamy, P. M. Shah, Artificial intelligence in rural healthcare delivery: Bridging gaps and enhancing equity through innovation, arXiv preprint arXiv:2508.11738 (2025)

    Pith/arXiv arXiv 2025

  20. [20]

    Waqas, S

    M. Waqas, S. U. Ahmed, M. A. Tahir, J. Wu, R. Qureshi, Exploring multiple instance learning (mil): A brief survey, Expert Systems with Applications 250 (2024) 123893

    2024

  21. [21]

    S. S. Hussain, X. Degang, P. M. Shah, H. Khan, A. Zeb, Alzformer: Multi-modal frame- work for alzheimer’s classification using mri and graph-embedded demographics guided by adaptive attention gating, Computerized Medical Imaging and Graphics (2025) 102638

    2025

  22. [22]

    K. Zhou, Z. Liu, Y. Qiao, T. Xiang, C. C. Loy, Domain generalization: A survey, IEEE Transactions on Pattern Analysis and Machine Intelligence 45 (4) (2023) 4396–4415.doi: 10.1109/TPAMI.2022.3195549

    arXiv 2023

  23. [23]

    M.Jahanifar, M.Raza, K.Xu, T.T.L.Vuong, R.Jewsbury, A.Shephard, N.Zamanitajed- din, J. T. Kwak, S. E. A. Raza, F. Minhas, et al., Domain generalization in computational pathology: survey and guidelines, ACM Computing Surveys 57 (11) (2025) 1–37

    2025

  24. [24]

    Poceviči¯ ut˙ e, G

    M. Poceviči¯ ut˙ e, G. Eilertsen, S. Garvin, C. Lundström, Detecting domain shift in multiple instance learning for digital pathology using fréchet domain distance, in: International con- ference on medical image computing and computer-assisted intervention, Springer, 2023, pp. 157–167

    2023

  25. [25]

    M. Ochi, D. Komura, T. Onoyama, K. Shinbo, H. Endo, H. Odaka, M. Kakiuchi, H. Ka- toh, T. Ushiku, S. Ishikawa, Registered multi-device/staining histology image dataset for domain-agnostic machine learning models, Scientific Data 11 (1) (2024) 330

    2024

  26. [26]

    Enhancing Whole Slide Image Classification through Supervised Contrastive Domain Adaptation

    I. Carretero, P. Meseguer, R. del Amor, V. Naranjo, Enhancing whole slide image classifi- cation through supervised contrastive domain adaptation, arXiv preprint arXiv:2412.04260 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  27. [27]

    Tomar, A

    D. Tomar, A. Binder, A. Kleppe, Are nuclear masks all you need for improved out-of- domain generalisation? a closer look at cancer classification in histopathology, Advances in Neural Information Processing Systems 37 (2024) 43499–43532

    2024

  28. [28]

    Macenko, M

    M. Macenko, M. Niethammer, J. S. Marron, D. Borland, J. T. Woosley, X. Guan, C. Schmitt, N. E. Thomas, A method for normalizing histology slides for quantitative analysis, in: 2009 IEEE international symposium on biomedical imaging: from nano to macro, IEEE, 2009, pp. 1107–1110. 25

    2009

  29. [29]

    Y. Shen, Y. Luo, D. Shen, J. Ke, Randstainna: Learning stain-agnostic features from histology slides by bridging stain augmentation and normalization, in: International Con- ference on Medical Image Computing and Computer-Assisted Intervention, Springer, 2022, pp. 212–221

    2022

  30. [30]

    Z. Wang, Y. Luo, R. Qiu, Z. Huang, M. Baktashmotlagh, Learning to diversify for single domain generalization, in: Proceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 834–843

    2021

  31. [31]

    Huang, H

    Z. Huang, H. Wang, E. P. Xing, D. Huang, Self-challenging improves cross-domain gen- eralization, in: Computer vision–ECCV 2020: 16th European conference, Glasgow, UK, August 23–28, 2020, proceedings, part II 16, Springer, 2020, pp. 124–140

    2020

  32. [32]

    Volpi, H

    R. Volpi, H. Namkoong, O. Sener, J. C. Duchi, V. Murino, S. Savarese, Generalizing to unseen domains via adversarial data augmentation, Advances in neural information processing systems 31 (2018)

    2018

  33. [33]

    Khosla, P

    P. Khosla, P. Teterwak, C. Wang, A. Sarna, Y. Tian, P. Isola, A. Maschinot, C. Liu, D. Krishnan, Supervised contrastive learning, Advances in neural information processing systems 33 (2020) 18661–18673

    2020

  34. [34]

    Izmailov, D

    P. Izmailov, D. Podoprikhin, T. Garipov, D. Vetrov, A. G. Wilson, Averaging weights leads to wider optima and better generalization, arXiv preprint arXiv:1803.05407 (2018)

    Pith/arXiv arXiv 2018

  35. [35]

    Stoica, D

    G. Stoica, D. Bolya, J. Bjorner, P. Ramesh, T. Hearn, J. Hoffman, Zipit! merging models from different tasks without training, arXiv preprint arXiv:2305.03053 (2023)

    Pith/arXiv arXiv 2023

  36. [36]

    Wortsman, G

    M. Wortsman, G. Ilharco, S. Y. Gadre, R. Roelofs, R. Gontijo-Lopes, A. S. Morcos, H. Namkoong, A. Farhadi, Y. Carmon, S. Kornblith, et al., Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference time, in: International conference on machine learning, PMLR, 2022, pp. 23965–23998

    2022

  37. [37]

    Arpit, H

    D. Arpit, H. Wang, Y. Zhou, C. Xiong, Ensemble of averages: Improving model selec- tion and boosting performance in domain generalization, Advances in Neural Information Processing Systems 35 (2022) 8265–8277

    2022

  38. [38]

    J. Cha, S. Chun, K. Lee, H.-C. Cho, S. Park, Y. Lee, S. Park, Swad: Domain generalization by seeking flat minima, Advances in Neural Information Processing Systems 34 (2021) 22405–22418

    2021

  39. [39]

    A. Rame, M. Kirchmeyer, T. Rahier, A. Rakotomamonjy, P. Gallinari, M. Cord, Diverse weight averaging for out-of-distribution generalization, Advances in Neural Information Processing Systems 35 (2022) 10821–10836

    2022

  40. [40]

    S. Jain, S. Addepalli, P. K. Sahu, P. Dey, R. V. Babu, Dart: Diversify-aggregate-repeat training improves generalization of neural networks, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 16048–16059. 26

    2023

  41. [41]

    Z. Li, K. Ren, X. JIANG, Y. Shen, H. Zhang, D. Li, SIMPLE: Specialized model-sample matching for domain generalization, in: The Eleventh International Conference on Learn- ing Representations, 2023, pp. 1–3467. URL https://openreview.net/forum?id=BqrPeZ_e5P

    2023

  42. [42]

    Y. Shu, X. Guo, J. Wu, X. Wang, J. Wang, M. Long, Clipood: Generalizing clip to out-of-distributions, in: International Conference on Machine Learning, PMLR, 2023, pp. 31716–31731

    2023

  43. [43]

    P. Wang, Z. Zhang, Z. Lei, L. Zhang, Sharpness-aware gradient matching for domain generalization, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 3769–3778

    2023

  44. [44]

    B. Shen, A. Saito, A. Ueda, K. Fujita, Y. Nagamatsu, M. Hashimoto, M. Kobayashi, A. H. Mirza, H. P. Graf, E. Cosatto, et al., Development of multiple ai pipelines that predict neoadjuvant chemotherapy response of breast cancer using h&e-stained tissues, The Journal of Pathology: Clinical Research 9 (3) (2023) 182–194

    2023

  45. [45]

    D. W. Dodington, A. Lagree, S. Tabbarah, M. Mohebpour, A. Sadeghi-Naini, W. T. Tran, F.-I. Lu, Analysis of tumor nuclear features using artificial intelligence to predict response to neoadjuvant chemotherapy in high-risk breast cancer patients, Breast Cancer Research and Treatment 186 (2021) 379–389

    work page 2021

  46. [46]

    X. Wang, W. Yuan, Nuclei-level prior knowledge constrained multiple instance learning for breast histopathology whole slide image classification, Iscience 27 (6) (2024)

    work page 2024

  47. [47]

    N. M. Carleton, G. Lee, A. Madabhushi, R. W. Veltri, Advances in the computational and molecular understanding of the prostate cancer cell nucleus, Journal of cellular biochem- istry 119 (9) (2018) 7127–7142

    work page 2018

  48. [48]

    Chen, A.-P

    J.-M. Chen, A.-P. Qu, L.-W. Wang, J.-P. Yuan, F. Yang, Q.-M. Xiang, N. Maskey, G.-F. Yang, J. Liu, Y. Li, New breast cancer prognostic factors identified by computer-aided image analysis of he stained histopathology images, Scientific reports 5 (1) (2015) 10690

    work page 2015

  49. [49]

    Nawaz, Y

    S. Nawaz, Y. Yuan, Computational pathology: Exploring the spatial dimension of tumor ecology, Cancer letters 380 (1) (2016) 296–303

    work page 2016

  50. [50]

    E. G. Fischer, Nuclear morphology and the biology of cancer cells, Acta cytologica 64 (6) (2020) 511–519

    work page 2020

  51. [51]

    M. M. Melssen, N. D. Sheybani, K. M. Leick, C. L. Slingluff Jr, Barriers to immune cell infiltration in tumors, Journal for immunotherapy of cancer 11 (4) (2023) e006401

    work page 2023

  52. [52]

    E. R. Sherwood, T. Toliver-Kinsky, Mechanisms of the inflammatory response, Best prac- tice & research Clinical anaesthesiology 18 (3) (2004) 385–405

    work page 2004

  53. [53]

    W. Park, S. Wei, B.-S. Kim, B. Kim, S.-J. Bae, Y. C. Chae, D. Ryu, K.-T. Ha, Diversity and complexity of cell death: a historical review, Experimental & molecular medicine 55 (8) (2023) 1573–1594. 27

    work page 2023

  54. [54]

    Gamper, N

    J. Gamper, N. A. Koohbanani, K. Benes, S. Graham, M. Jahanifar, S. A. Khurram, A. Azam, K. Hewitt, N. Rajpoot, Pannuke dataset extension, insights and baselines, arXiv preprint arXiv:2003.10778 (2020)

    Pith/arXiv arXiv 2003

  55. [55]

    Graham, H

    S. Graham, H. Chen, J. Gamper, Q. Dou, P.-A. Heng, D. Snead, Y. W. Tsang, N. Rajpoot, Mild-net: Minimal information loss dilated network for gland instance segmentation in colon histology images, Medical image analysis 52 (2019) 199–211

    work page 2019

  56. [56]

    Graham, M

    S. Graham, M. Jahanifar, A. Azam, M. Nimir, Y.-W. Tsang, K. Dodd, E. Hero, H. Sahota, A. Tank, K. Benes, et al., Lizard: a large-scale dataset for colonic nuclear instance seg- mentation and classification, in: Proceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 684–693

    work page 2021

  57. [57]

    Graham, Q

    S. Graham, Q. D. Vu, S. E. A. Raza, A. Azam, Y. W. Tsang, J. T. Kwak, N. Rajpoot, Hover-net: Simultaneous segmentation and classification of nuclei in multi-tissue histology images, Medical image analysis 58 (2019) 101563

    work page 2019

  58. [58]

    R. L. Grossman, A. P. Heath, V. Ferretti, H. E. Varmus, D. R. Lowy, W. A. Kibbe, L. M. Staudt, Toward a shared vision for cancer genomic data, New England Journal of Medicine 375 (12) (2016) 1109–1112

    work page 2016

  59. [59]

    Kumar, R

    N. Kumar, R. Verma, S. Sharma, S. Bhargava, A. Vahadane, A. Sethi, A dataset and a technique for generalized nuclear segmentation for computational pathology, IEEE trans- actions on medical imaging 36 (7) (2017) 1550–1560

    work page 2017

  60. [60]

    A.Mahbod, G.Schaefer, B.Bancher, C.Löw, G.Dorffner, R.Ecker, I.Ellinger, Cryonuseg: A dataset for nuclei instance segmentation of cryosectioned h&e-stained histological im- ages, Computers in biology and medicine 132 (2021) 104349

    work page 2021

  61. [61]

    Naylor, M

    P. Naylor, M. Laé, F. Reyal, T. Walter, Nuclei segmentation in histopathology images using deep neural networks, in: 2017 IEEE 14th international symposium on biomedical imaging (ISBI 2017), IEEE, 2017, pp. 933–936

    work page 2017

  62. [62]

    Naylor, M

    P. Naylor, M. Laé, F. Reyal, T. Walter, Segmentation of nuclei in histopathology images by deep regression of the distance map, IEEE transactions on medical imaging 38 (2) (2018) 448–459

    work page 2018

  63. [63]

    Sirinukunwattana, J

    K. Sirinukunwattana, J. P. Pluim, H. Chen, X. Qi, P.-A. Heng, Y. B. Guo, L. Y. Wang, B. J. Matuszewski, E. Bruni, U. Sanchez, et al., Gland segmentation in colon histology images: The glas challenge contest, Medical image analysis 35 (2017) 489–502

    work page 2017

  64. [64]

    Verma, N

    R. Verma, N. Kumar, A. Patil, N. C. Kurian, S. Rane, S. Graham, Q. D. Vu, M. Zwager, S. E. A. Raza, N. Rajpoot, et al., Monusac2020: A multi-organ nuclei segmentation and classification challenge, IEEE Transactions on Medical Imaging 40 (12) (2021) 3413–3423

    work page 2021

  65. [65]

    Khurram, J.Kalpathy-Cramer, T.Zhao, etal., Methodsforsegmentationandclassification of digital microscopy tissue images, Frontiers in bioengineering and biotechnology 7 (2019) 53

    Q.D.Vu, S.Graham, T.Kurc, M.N.N.To, M.Shaban, T.Qaiser, N.A.Koohbanani, S.A. Khurram, J.Kalpathy-Cramer, T.Zhao, etal., Methodsforsegmentationandclassification of digital microscopy tissue images, Frontiers in bioengineering and biotechnology 7 (2019) 53. 28

    work page 2019

  66. [66]

    Y. Han, Y. Lei, V. Shkolnikov, D. Xin, A. Auduong, S. Barcelo, J. Allebach, E. J. Delp, An ensemble method with edge awareness for abnormally shaped nuclei segmentation, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2023, pp. 4315–4325

    work page 2023

  67. [67]

    Hollandi, N

    R. Hollandi, N. Moshkov, L. Paavolainen, E. Tasnadi, F. Piccinini, P. Horvath, Nucleus segmentation: towards automated solutions, Trends in Cell Biology 32 (4) (2022) 295–310

    work page 2022

  68. [68]

    Hörst, M

    F. Hörst, M. Rempe, L. Heine, C. Seibold, J. Keyl, G. Baldini, S. Ugurel, J. Siveke, B. Grünwald, J. Egger, et al., Cellvit: Vision transformers for precise cell segmentation and classification, Medical Image Analysis 94 (2024) 103143

    work page 2024

  69. [69]

    J. W. Johnson, Automatic nucleus segmentation with mask-rcnn, in: Advances in Com- puter Vision: Proceedings of the 2019 Computer Vision Conference (CVC), Volume 2 1, Springer, 2020, pp. 399–407

    work page 2019

  70. [70]

    Schmidt, M

    U. Schmidt, M. Weigert, C. Broaddus, G. Myers, Cell detection with star-convex polygons, in: Medical image computing and computer assisted intervention–MICCAI 2018: 21st international conference, Granada, Spain, September 16-20, 2018, proceedings, part II 11, Springer, 2018, pp. 265–273

    work page 2018

  71. [71]

    Stringer, T

    C. Stringer, T. Wang, M. Michaelos, M. Pachitariu, Cellpose: a generalist algorithm for cellular segmentation, Nature methods 18 (1) (2021) 100–106

    work page 2021

  72. [72]

    Hörst, M

    F. Hörst, M. Rempe, H. Becker, L. Heine, J. Keyl, J. Kleesiek, Cellvit++: Energy-efficient and adaptive cell segmentation and classification using foundation models, arXiv preprint arXiv:2501.05269 (2025)

    Pith/arXiv arXiv 2025

  73. [73]

    R. Yuan, W. Zhang, X. Dong, W. Zhang, Crns: Clip-driven referring nuclei segmentation, The Journal of Supercomputing 81 (1) (2025) 174

    work page 2025

  74. [74]

    Ignatov, J

    A. Ignatov, J. Yates, V. Boeva, Histopathological image classification with cell morphology aware deep neural networks, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 6913–6925

    work page 2024

  75. [75]

    X.Wang, S.Yang, J.Zhang, M.Wang, J.Zhang, W.Yang, J.Huang, X.Han, Transformer- based unsupervised contrastive learning for histopathological image classification, Medical image analysis 81 (2022) 102559

    work page 2022

  76. [76]

    Schaeffer, C

    R.K.Pai, I.Banerjee, S.Shivji, S.Jain, D.Hartman, D.D.Buchanan, M.A.Jenkins, D.F. Schaeffer, C. Rosty, J. Como, et al., Quantitative pathologic analysis of digitized images of colorectal carcinoma improves prediction of recurrence-free survival, Gastroenterology 163 (6) (2022) 1531–1546

    work page 2022

  77. [77]

    H. Ueno, Y. Hashiguchi, H. Shimazaki, E. Shinto, Y. Kajiwara, K. Nakanishi, K. Kato, K. Maekawa, K. Miyai, T. Nakamura, et al., Objective criteria for crohn-like lymphoid reaction in colorectal cancer, American journal of clinical pathology 139 (4) (2013) 434– 441. 29

    work page 2013

  78. [78]

    G. E. Idos, J. Kwok, N. Bonthala, L. Kysh, S. B. Gruber, C. Qu, The prognostic impli- cations of tumor infiltrating lymphocytes in colorectal cancer: a systematic review and meta-analysis, Scientific reports 10 (1) (2020) 3360

    work page 2020

  79. [79]

    H. Lee, D. Sha, N. Foster, Q. Shi, S. Alberts, T. Smyrk, F. Sinicrope, Analysis of tumor microenvironmental features to refine prognosis by t, n risk group in patients with stage iii colon cancer (ncctg n0147)(alliance), Annals of Oncology 31 (4) (2020) 487–494

    work page 2020

  80. [80]

    Vahadane, T

    A. Vahadane, T. Peng, A. Sethi, S. Albarqouni, L. Wang, M. Baust, K. Steiger, A. M. Schlitter, I. Esposito, N. Navab, Structure-preserving color normalization and sparse stain separation for histological images, IEEE transactions on medical imaging 35 (8) (2016) 1962–1971

    work page 2016

Showing first 80 references.

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.