Pith. sign in

REVIEW 2 major objections 6 minor 7 cited by

Towards Robust Foundation Models for Digital Pathology

T0 review · 2 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 Every one of the 20 pathology foundation models evaluated encodes its hospital of origin along with the tissue biology, and the paper shows this contamination directly predicts how badly downstream diagnostic models fail.

desk verdict Worth a look: the PathoROB benchmark is a genuinely useful resource with a solid empirical finding, but the center-as-technical-confounder premise is unverified and the headline correlation is partly circular; the authors themselves acknowledge the key caveat. read the letter →

arxiv 2507.17845 v1 pith:4NIZJOFU submitted 2025-07-22 eess.IV cs.AIcs.CVcs.LGq-bio.QM

Jonah Kömen , Edwin D. de Jong , Julius Hense , Hannah Marienwald , Jonas Dippel , Philip Naumann , Eric Marcus , Lukas Ruff
show 4 more authors
Maximilian Alber Jonas Teuwen Frederick Klauschen Klaus-Robert Müller
This is my paper
classification eess.IVcs.AIcs.CVcs.LGq-bio.QM
keywords pathologyfoundationmodelsrobustnessindexmedicalcenterconfoundersbatcheffectsshortcutlearningmulti-centerbenchmarkself-supervisedrepresentationdigital
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 sets out to establish that every prominent pathology foundation model currently in use encodes the originating medical center — its staining, scanner, and laboratory signatures — alongside the biological tissue signal, and that this contamination is strong enough to cause diagnostic errors in clinical settings. To make that claim measurable, the authors build PathoROB, a benchmark of four multicentre histopathology datasets spanning 28 biological classes and 34 medical centers, with a robustness index $R = \frac{|SO|}{|SO|+|OS|}$ that scores each model from 0 to 1 by asking whether a patch's nearest neighbours in embedding space share its biological class or its medical center. Across all 20 models the index ranges from 0.463 to 0.877: none is fully robust, and the index strongly predicts downstream failure (Spearman $\rho = 0.908$) when training data correlate hospital with diagnosis. The paper also shows that stain normalization, batch correction, and domain-adversarial training reduce but do not eliminate this risk, and concludes that robustness evaluation should be a standard part of validating medical foundation models before clinical adoption.

What carries the argument

The robustness index. For each patch in a balanced dataset with one biological-class axis and one medical-center axis, the paper collects the $k$ nearest neighbours in the normalized embedding space and counts neighbours with the Same biological class and Other center (SO) versus neighbours with Other biological class and Same center (OS); the index is $R = \frac{|SO|}{|SO| + |OS|}$. $R = 1$ means biological features completely determine the local neighbourhood; $R = 0$ means center features do. Two companion metrics complete the PathoROB benchmark: the average performance drop of a linear-probe classifier trained under increasing center–diagnosis correlation, and a clustering score equal to the adjusted Rand index against biological labels minus the adjusted Rand index against center labels. The mechanism that carries the argument is that the largest-variance principal directions of FM embeddings contain both biological and technical signal, which is why removing one risks removing the other and why post-hoc robustification reduces but cannot eliminate center encoding.

What would settle it

One decisive experiment: re-scan and re-stain the same set of slides through different scanners and staining protocols, so the biological content is identical while the technical signature changes, then compute the robustness index with center defined by processing pipeline. If a model that scores low on PathoROB also scores low in this controlled pairing, its center signal is provably technical; if it scores near 1 here while scoring low on real multi-center data, the low scores reflect genuine biological differences between centers rather than artifacts, which would undermine the paper's premise.

Watch

Extended reading notes

Core claim

The central claim is that current self-supervised objectives give pathology foundation models no reason to distinguish biological variation from technical variation, so the models encode medical center identity as a first-class feature — and this is clinically harmful in a measurable way. From 99,392 patches across 34 centers, a linear probe recovers the originating center from any model's representations with 88–98% accuracy, and for many models the t-SNE organization of the embedding space is visibly center-first. The paper quantifies this with a robustness index, defined as the fraction of local nearest neighbours that share the biological class but not the center; scores range from 0.463 (Ciga) to 0.877 (Virchow2, the best of the 20). The clinically consequential result is the shortcut-learning experiment: when downstream training data correlate hospital with diagnosis, which the authors argue is common especially for rare diseases, downstream classifiers exploit the center signature; on the Camelyon tumour-detection task accuracy falls from above 92% to 53–87%, and between 19% and 99% of tumour patches from one of the two centers are missed depending on the foundation model. The robustness index predicts the size of this degradation ($\rho = 0.908$, $p = 0.00004$), which is the evidence that the metric is measuring the right thing.

Load-bearing premise

The entire benchmark treats medical center as a purely technical confounder, so a low robustness index is counted as harm; if centers genuinely differ in patient biology, demographics, or treatment history, part of the measured signal could be legitimate, and the model ranking would be biased.

Editorial extensions

If this is right

  • Robustness should be reported alongside accuracy in pathology foundation-model evaluations: only two of the twenty models (Virchow2, Atlas) are not beaten on both axes by another model, so accuracy-only leaderboards mislead.
  • Under the common real-world condition of hospital–diagnosis correlation in training data, downstream diagnostic models inherit systematic, center-specific errors — on Camelyon, 19–99% of tumor patches from one center were missed, depending on the foundation model.
  • Post-hoc robustification raises the robustness index (up to 0.92 with stain normalization plus ComBat) and measurably reduces errors, but no method eliminates them, so robustness must be engineered into pre-training rather than patched on afterward.
  • Unsupervised applications also degrade: clustering for subtype discovery and diagnostic case retrieval are both corrupted by center signatures, with clustering scores near zero or negative for most models.
  • Scale and language supervision help — robustness correlates with pre-training slide count ($\rho = 0.692$) and vision-language models rank among the most robust — but even the best model is not fully robust, so scale alone is insufficient.

Reading between the lines

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

  • Because the robustness index needs only paired biological-class and center labels, it could serve as a cheap pre-deployment screen in other biomedical domains with known batch structure — radiology, genomics, flow cytometry — before any downstream model is trained.
  • The paper's own PCA analysis shows center and biology share the same principal directions; a natural next step is targeted robustification that nullifies only directions with high center-separability and low biological-separability, rather than ComBat's wholesale correction, which the paper shows can destroy biological signal.
  • The correlation between pre-training slide count and robustness ($\rho = 0.692$) suggests robustness may follow its own scaling law; measuring robustness at intermediate checkpoints during pre-training could separate the effect of scale from the effect of objective and make robustness a trainable objective.
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

2 major / 6 minor

Summary. The paper introduces PathoROB, a benchmark for evaluating the robustness of pathology foundation models (FMs) to non-biological, medical-center-related variation. It proposes three metrics: a k-nearest-neighbor robustness index, a clustering score comparing adjusted Rand indices against biological and center labels, and an average performance drop (APD) measured under training splits with increasing Cramér's V between medical center and biological label. The authors evaluate 20 FMs on 99,392 patches from 28 biological classes and 34 medical centers, report that all FMs encode center information with robustness indices between 0.463 and 0.877, and observe a strong correlation between the robustness index and downstream APD (Spearman rho = 0.908). They further show that post-hoc robustification via Reinhard stain normalization, ComBat, and domain-adversarial training partially mitigates the identified issues. The central claim is that non-robust FM representations can cause major diagnostic downstream errors and prevent safe clinical adoption.

Significance. If the central assumption holds, this is a valuable and timely contribution: it is the first systematic multi-FM robustness benchmark in digital pathology, with precisely defined metrics, experiments across 20 models, confidence intervals, bootstrap estimates, and permutation-based p-values. The distinction between local neighborhood structure (robustness index) and global structure (clustering score) is conceptually useful, and the comparison of post-hoc robustification strategies is practically relevant. The manuscript is also transparent in acknowledging the risk of center-biology entanglement in Section 3 and in Sup. Note G.2. However, the strength of the headline claims is currently limited by two load-bearing issues: the unverified assumption that medical center is a pure technical confounder, and the partly expected correlation between the robustness index and the APD, which both measure the dominance of center information in the representations. These issues do not invalidate the benchmark as a measurement tool, but they prevent the current strong causal and clinical conclusions from being fully established.

major comments (2)
  1. [§2.3.1, §4.5, Fig. 3d] The central claim that all 20 FMs encode non-biological center information to a clinically harmful degree depends on treating "medical center" as a pure technical confounder, conditionally independent of biology given the class label. Section 4.1 states that the subsampling "tried to ensure that systematic differences between medical centers are solely due to biologically irrelevant factors," but no verification of this balance is reported. The Camelyon centers are distinct patient populations, and TCGA tissue source sites are documented to differ in stage, subtype, and demographic composition; the paper itself concedes in Section 3 that "center-specific biological information could be mistaken for technical artifacts." The ComBat experiment in Sup. Note G.2 directly shows that under V-correlated splits batch correction "likely removed important biological signals," demonstrating that center and biology are entangled in the same data used for the APD experiment. If center labels carry biological signal, then the robustness index (Eq. 1 in Section 4.4) penalizes FMs for correctly encoding real morphology or patient biology, and the FM ranking in Fig. 2b, together with the headline rho = 0.908 in Fig. 3d, conflates sensitivity to technical artifacts with sensitivity to biological variation. Please add a covariate-balance table (stage, subtype, treatment, and demographic variables) for the subsampled PathoROB datasets, or a technical-replicate control (identical tissue scanned or stained at different sites), or explicitly re-frame all claims as "sensitivity to center identity" rather than "non-biological robustness."
  2. [§2.3.1, §4.5, Fig. 3d] The reported Spearman correlation rho = 0.908 between the robustness index and the average performance drop is partly expected by construction and should not be presented as an independent validation of the robustness index. The APD defined in Section 4.5 measures the relative accuracy drop of a downstream model as the medical-center/label correlation increases from Cramér's V = 0 to V = 1, which directly quantifies how strongly center identity is encoded in the FM representation. The robustness index defined in Section 4.4 measures the same underlying construct from a different angle, namely the local neighborhood composition in terms of same-biological/other-center versus other-biological/same-center neighbors. Both metrics are, in different forms, monotone functions of the degree to which center information dominates biological information, so a high correlation between them is largely a consistency check rather than a predictive validation. Please re-frame this result accordingly and provide an independent validation of the robustness index, for example by evaluating FMs on paired technical replicates (same slides imaged with different scanners or stain batches) or by reporting a partial correlation that controls for pre-training data size and model family.
minor comments (6)
  1. [§2.2] The phrase "lacking behind top vision-only prediction accuracy" should be "lagging behind."
  2. [Figure 7 caption] The caption contains a typo: "neigbhors" should be "neighbors."
  3. [Sup. Note A.2, Table 4] The abbreviations "CLS" and "MCT" in the caption of Table 4 are not defined; please define them as biological class and medical center.
  4. [Data & code availability] The statement "All data and codes from the PathoROB benchmark will be made available in a public repository soon" is insufficient for a benchmark paper; a reviewer-accessible repository or DOI should be provided at submission time.
  5. [Sup. Note G.3 and Fig. 24] The clustering scores reported in Fig. 5b depend on the silhouette-score-based choice of K, and Sup. Note G.3 shows that this selection is highly suboptimal for several FMs, changing scores from near zero to above 0.8; this sensitivity should be summarized in the main text rather than only in the supplement.
  6. [Throughout] The spelling of "ComBat" is inconsistent in a few places (e.g., "Combat"), and the special characters in "Cramér's V" are inconsistently formatted; please unify these.

Circularity Check

0 steps flagged · score 2.0 of 10

No load-bearing circularity: the robustness index and the downstream drop are distinct operationalizations, and the center-encoding claim is independently demonstrated in Figures 1c-1d; self-citations are auxiliary.

full rationale

The central derivation chain is not circular. The robustness index (Section 4.4, Eq. 1) is defined from kNN neighborhood composition (SO versus OS) on balanced multi-center data, whereas the average performance drop (Section 4.5) is a separate behavioral measure: linear-probe accuracy degradation under artificially increased Cramer's V training splits. Both metrics respond to the same underlying class/center label structure, so their high correlation (rho = 0.908) is partly an internal consistency check rather than a fully independent validation, but there is no algebraic reduction: APD depends on trained downstream behavior, regularization, and held-out test data, none of which enter the robustness index formula. The paper's premise that FMs encode medical-center information is supported in-manuscript by linear probing (88-98% center prediction accuracy, Figure 1d) and t-SNE organization (Figure 1c), so the self-citations [40,41] for this premise are not load-bearing. The acknowledged limitation that center labels may correlate with biological or demographic factors is a construct-validity and correctness risk, explicitly flagged in the Discussion ('center-specific biological information could be mistaken for technical artifacts'), not a circularity. No fitted parameter is renamed a prediction, and no uniqueness claim is imported from the authors' prior work. Score 2 reflects only the minor self-citation presence and the partially self-referential validation logic; neither constitutes circularity by the standards of an algebraic or definitional reduction.

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

The central claim rests mainly on dataset subsampling choices and the equation of medical center with technical confounder, not on new physical entities. The free parameters are methodological (k, K, DANN lambda, normalization reference) rather than fitted constants. No new particles, forces, or conserved quantities are introduced.

free parameters (4)
  • k (number of nearest neighbors) for robustness index = TCGA 2x2: 61; Camelyon: 11; Tolkach ESCA: 46
    k is chosen as the median of per-model values that maximize biological class kNN accuracy for each dataset (Sup. Note B.2). The reported robustness indices depend on this choice, and the metric's absolute values shift with k.
  • K (number of clusters) for clustering score = Selected per dataset and model in [2,30] by silhouette score
    K-means K is selected by maximizing silhouette score (Section 4.6). The clustering score changes with K; Sup. Note G.3 shows silhouette selection can be suboptimal for some FMs.
  • Lambda schedule for DANN = Linear from 0 to 1 over 20 epochs
    Domain-adversarial loss weight is scheduled by hand (Section 4.7). The robustification results are conditional on this schedule.
  • Reinhard normalization reference statistics = Mean and standard deviation of 500 TCGA-LUSC patches
    Stain normalization target is a hand-chosen reference patch set (Section 4.7); different references would change the improved robustness indices.
assumptions (5)
  • domain assumption Medical center identity is a purely technical, non-biological confounder in the evaluated datasets.
    Invoked in Section 4.1 when subsampling such that systematic center differences are solely due to biologically irrelevant factors; the robustness index treats center as the only confounding axis. If center correlates with patient or tumor biology, the metric can punish true biological signal.
  • domain assumption Patch-level representations are the appropriate unit for assessing pathology FM robustness.
    Section 3 states patch encoders are most crucial to evaluate because slide representations are built from patches; this is a reasoned but unproven premise.
  • domain assumption Neighborhood structure in l2-normalized embedding space reflects semantic biological similarity.
    The robustness index and kNN analyses assume cosine distance in FM embedding space is a meaningful similarity measure (Section 4.4, Sup. Note B.1).
  • standard math The provided biological class labels (normal/tumor, tumor types, tissue compartments) are ground truth.
    All metrics compare against these labels from Camelyon, TCGA-UT, and Tolkach ESCA; no label-error analysis is provided.
  • domain assumption Classic batch-correction and stain-normalization methods transfer from molecular and imaging domains to FM feature spaces.
    ComBat is adopted from molecular biology (Section 4.7); its effectiveness and safety depend on this transfer being valid.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards Robust Foundation Models for Digital Pathology." pith.science (2026). https://pith.science/paper/4NIZJOFU

@misc{pith2026250717845,
  author       = {Pith},
  title        = {Pith review of: Towards Robust Foundation Models for Digital Pathology},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4NIZJOFU}},
  note         = {Machine review of arXiv:2507.17845}
}
read the original abstract

Biomedical Foundation Models (FMs) are rapidly transforming AI-enabled healthcare research and entering clinical validation. However, their susceptibility to learning non-biological technical features -- including variations in surgical/endoscopic techniques, laboratory procedures, and scanner hardware -- poses risks for clinical deployment. We present the first systematic investigation of pathology FM robustness to non-biological features. Our work (i) introduces measures to quantify FM robustness, (ii) demonstrates the consequences of limited robustness, and (iii) proposes a framework for FM robustification to mitigate these issues. Specifically, we developed PathoROB, a robustness benchmark with three novel metrics, including the robustness index, and four datasets covering 28 biological classes from 34 medical centers. Our experiments reveal robustness deficits across all 20 evaluated FMs, and substantial robustness differences between them. We found that non-robust FM representations can cause major diagnostic downstream errors and clinical blunders that prevent safe clinical adoption. Using more robust FMs and post-hoc robustification considerably reduced (but did not yet eliminate) the risk of such errors. This work establishes that robustness evaluation is essential for validating pathology FMs before clinical adoption and demonstrates that future FM development must integrate robustness as a core design principle. PathoROB provides a blueprint for assessing robustness across biomedical domains, guiding FM improvement efforts towards more robust, representative, and clinically deployable AI systems that prioritize biological information over technical artifacts.

Figures

Figures reproduced from arXiv: 2507.17845 by the authors.

Figure 1
Figure 1. PathoROB benchmark and FM representation space exploration [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Quantification of foundation model robustness [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Downstream model performance under spuriously correlated training data [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (20 more)
Figure 4
Figure 4. Figure 4: Downstream model predictions for the Camelyon tumor detection task [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Clustering and retrieval with non-robust FM representations. a [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]
Figure 6
Figure 6. Figure 6: Robustification of foundation model representations without FM retraining [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]
Figure 7
Figure 7. Figure 7: Visual explanation of the robustness index calculation on TCGA 2x2. a. [PITH_FULL_IMAGE:figures/full_fig_p021_7.png]
Figure 8
Figure 8. Figure 8: Sample patches from the (a) Camelyon, (b) TCGA 4x4, and (c) Tolkach ESCA datasets. [PITH_FULL_IMAGE:figures/full_fig_p037_8.png]
Figure 9
Figure 9. Figure 9: Trading off performance vs robustness. by choosing different values for k, different tradeoffs can be obtained. the TCGA-2k dataset described in [41], which has five biological and confounding classes chosen such that for 20 of the 25 combinations of these, an equal nu…
Figure 10
Figure 10. Figure 10: Robustness index results on TCGA 2x2 for all models [PITH_FULL_IMAGE:figures/full_fig_p042_10.png]
Figure 11
Figure 11. Figure 11: Robustness index results on Camelyon. 43 [PITH_FULL_IMAGE:figures/full_fig_p043_11.png]
Figure 12
Figure 12. Figure 12: Robustness index results on Tolkach ESCA [PITH_FULL_IMAGE:figures/full_fig_p044_12.png]
Figure 13
Figure 13. Figure 13: Robustness index per biological class (left) and per confounding class (right) for UNI2-h [PITH_FULL_IMAGE:figures/full_fig_p048_13.png]
Figure 14
Figure 14. Figure 14: Training and test data compositions, including the number of patches (slides), for the Camelyon [PITH_FULL_IMAGE:figures/full_fig_p053_14.png]
Figure 15
Figure 15. Figure 15: Training and test data compositions, including the number of patches (slides), for the TCGA [PITH_FULL_IMAGE:figures/full_fig_p054_15.png]
Figure 16
Figure 16. Figure 16: Training and test data compositions, including the number of patches, for the Tolkach ESCA [PITH_FULL_IMAGE:figures/full_fig_p055_16.png]
Figure 17
Figure 17. Figure 17: t-SNE visualization of the representation spaces per FM on Camelyon. [PITH_FULL_IMAGE:figures/full_fig_p058_17.png]
Figure 18
Figure 18. Figure 18: t-SNE visualization of the representation spaces after robustification by ComBat (RR) per [PITH_FULL_IMAGE:figures/full_fig_p059_18.png]
Figure 19
Figure 19. Figure 19: Classification accuracy for predicting the medical center and biological class before and after [PITH_FULL_IMAGE:figures/full_fig_p060_19.png]
Figure 20
Figure 20. Figure 20: Biological and medical center separability per PC [PITH_FULL_IMAGE:figures/full_fig_p063_20.png]
Figure 21
Figure 21. Figure 21: Retrieval performances from retrieval databases with heterogeneous data contributions across [PITH_FULL_IMAGE:figures/full_fig_p065_21.png]
Figure 22
Figure 22. Figure 22: Downstream model generalization performance on robustified feature spaces [PITH_FULL_IMAGE:figures/full_fig_p066_22.png]
Figure 23
Figure 23. Figure 23: Downstream model generalization performance on robustified feature spaces [PITH_FULL_IMAGE:figures/full_fig_p067_23.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 7 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. In-Context Multiple Instance Learning

    cs.LG 2026-06 unverdicted novelty 7.0 of 10

    A model pretrained on synthetic bag-structured data performs in-context learning for new MIL tasks from a handful of examples and outperforms task-specific supervised baselines on twelve benchmarks.

  2. Beyond Counts: A Distributional Robustness Margin For Pathology Foundation Models

    cs.CV 2026-07 conditional novelty 6.0 of 10

    CRoMa scores each pathology image embedding by the margin between cross-site biological matches and same-site biological distractors, revealing distributional lower tails that pooled robustness scores hide.

  3. DaX: Learning General Pathology Representations Across Scales

    eess.IV 2026-06 unverdicted novelty 6.0 of 10

    DaX is a pathology vision foundation model that extends DINOv3 with continuous magnification training and cross-scale consistency, achieving top average performance on a benchmark of 161 tasks from 44 datasets coverin...

  4. Symb-xMIL: Symbolic Explanations for Multiple Instance Learning in Digital Pathology

    cs.CV 2026-06 unverdicted novelty 6.0 of 10

    Symb-xMIL is a post-hoc explanation framework that quantifies MIL model alignment with logical decision rules in histopathology to enable rule-based interpretability.

  5. When Are Multimodal Predictions Biologically Supported? A Diagnostic Evaluation Framework

    cs.LG 2026-05 unverdicted novelty 6.0 of 10

    DECAT classifies multimodal representations into four diagnostic scenarios using null-referenced metrics and a rule-based procedure to detect shared biology versus confounders without knowing the confounder identity.

  6. Reliable Modeling of Distribution Shifts via Displacement-Reshaped Optimal Transport

    cs.LG 2026-05 unverdicted novelty 6.0 of 10

    ReshapeOT improves optimal transport reliability for distribution shifts by replacing the Euclidean ground metric with a Mahalanobis distance derived from observed displacement second moments.

  7. Mitigating Clever Hans Strategies in Image Classifiers through Generating Counterexamples

    cs.LG 2025-10 unverdicted novelty 6.0 of 10

    CFKD generates counterfactuals for human-guided correction of Clever Hans predictors in image classifiers, removing the need for confounder labels while scaling to multiple spurious correlations.

Reference graph

Works this paper leans on

111 extracted references · 54 canonical work pages · cited by 7 Pith papers

  1. [1]

    On the opportunities and risks of foundation models

    Rishi Bommasani, Drew A Hudson, Ehsan Adeli, Russ Altman, Simran Arora, Sydney von Arx, Michael S Bernstein, Jeannette Bohg, Antoine Bosselut, Emma Brunskill, et al. On the opportunities and risks of foundation models. arXiv preprint arXiv:2108.07258, 2021

    arXiv 2021

  2. [2]

    Foundational models in medical imaging: A comprehensive survey and future vision

    Bobby Azad, Reza Azad, Sania Eskandari, Afshin Bozorgpour, Amirhossein Kazerouni, Islem Rekik, and Dorit Merhof. Foundational models in medical imaging: A comprehensive survey and future vision. arXiv preprint arXiv:2310.18689, 2023

    arXiv

  3. [3]

    A comprehensive survey of foundation models in medicine

    Wasif Khan, Seowung Leem, Kyle B See, Joshua K Wong, Shaoting Zhang, and Ruogu Fang. A comprehensive survey of foundation models in medicine. IEEE Reviews in Biomedical Engineering, 2025

    2025

  4. [4]

    Foundation models in computational pathology: A review of challenges, opportunities, and impact

    Mohsin Bilal, Manahil Raza, Youssef Altherwy, Anas Alsuhaibani, Abdulrahman Abduljabbar, Fahdah Almarshad, Paul Golding, Nasir Rajpoot, et al. Foundation models in computational pathology: A review of challenges, opportunities, and impact. arXiv preprint arXiv:2502.08333, 2025

    arXiv 2025

  5. [5]

    Self supervised contrastive learning for digital histopathology

    Ozan Ciga, Tony Xu, and Anne Martel. Self supervised contrastive learning for digital histopathology. Machine Learning with Applications, 7, 2022

    2022

  6. [6]

    A clinical benchmark of public self-supervised pathology foundation models

    Gabriele Campanella, Shengjia Chen, Manbir Singh, Ruchika Verma, Silke Muehlstedt, Jennifer Zeng, Aryeh Stock, Matt Croken, Brandon Veremis, Abdulkadir Elmas, et al. A clinical benchmark of public self-supervised pathology foundation models. Nature Communications, 16(1):3640, 2025

    2025

  7. [7]

    Expert-level detection of pathologies from unannotated chest X-ray images via self-supervised learning

    Ekin Tiu, Ellie Talius, Pujan Patel, Curtis P Langlotz, Andrew Y Ng, and Pranav Rajpurkar. Expert-level detection of pathologies from unannotated chest X-ray images via self-supervised learning. Nature Biomedical Engineering, 6(12):1399–1406, 2022

    2022

  8. [8]

    A multimodal biomedical foundation model trained from fifteen million image–text pairs

    Sheng Zhang, Yanbo Xu, Naoto Usuyama, Hanwen Xu, Jaspreet Bagga, Robert Tinn, Sam Preston, Rajesh Rao, Mu Wei, Naveen Valluri, et al. A multimodal biomedical foundation model trained from fifteen million image–text pairs. NEJM AI, 2(1):AIoa2400640, 2025

    2025

Show all 111 references
  1. [9]

    BioBERT: a pre-trained biomedical language representation model for biomedical text mining

    Jinhyuk Lee, Wonjin Yoon, Sungdong Kim, Donghyeon Kim, Sunkyu Kim, Chan Ho So, and Jaewoo Kang. BioBERT: a pre-trained biomedical language representation model for biomedical text mining. Bioinformatics, 36(4):1234–1240, 09 2019

    2019

  2. [10]

    Domain-specific language model pretraining for biomedical natural language processing

    Yu Gu, Robert Tinn, Hao Cheng, Michael Lucas, Naoto Usuyama, Xiaodong Liu, Tristan Naumann, Jianfeng Gao, and Hoifung Poon. Domain-specific language model pretraining for biomedical natural language processing. ACM Transactions on Computing for Healthcare (HEALTH), 3(1):1–23, 2021

    2021

  3. [11]

    Transfer learning enables predictions in network biology

    Christina V Theodoris, Ling Xiao, Anant Chopra, Mark D Chaffin, Zeina R Al Sayed, Matthew C Hill, Helene Mantineo, Elizabeth M Brydon, Zexian Zeng, X Shirley Liu, et al. Transfer learning enables predictions in network biology. Nature, 618(7965):616–624, 2023

    2023

  4. [12]

    A foundation model of transcription across human cell types

    Xi Fu, Shentong Mo, Alejandro Buendia, Anouchka P Laurent, Anqi Shao, Maria del Mar Alvarez-Torres, Tianji Yu, Jimin Tan, Jiayu Su, Romella Sagatelian, et al. A foundation model of transcription across human cell types. Nature, pages 1–9, 2025

    2025

  5. [13]

    Virchow2: Scaling self-supervised mixed magnification models in pathology

    Eric Zimmermann, Eugene V orontsov, Julian Viret, Adam Casson, Michal Zelechowski, George Shaikovski, Neil Tenenholtz, James Hall, David Klimstra, Razik Yousfi, et al. Virchow2: Scaling self-supervised mixed magnification models in pathology. arXiv preprint arXiv:2408.00738, 2024

    2024 arXiv

  6. [14]

    Atlas: A novel pathology foundation model by Mayo Clinic, Charit´e, and Aignostics

    Maximilian Alber, Stephan Tietz, Jonas Dippel, Timo Milbich, Timoth´ee Lesort, Panos Korfiatis, Moritz Kr¨ugener, Beatriz Perez Cancer, Neelay Shah, Alexander M ¨ollers, et al. Atlas: A novel pathology foundation model by Mayo Clinic, Charit´e, and Aignostics. arXiv preprint a...

    2025 arXiv

  7. [15]

    Towards a general-purpose foundation model for computational pathology

    Richard J Chen, Tong Ding, Ming Y Lu, Drew FK Williamson, Guillaume Jaume, Andrew H Song, Bowen Chen, Andrew Zhang, Daniel Shao, Muhammad Shaban, et al. Towards a general-purpose foundation model for computational pathology. Nature Medicine, 30(3):850–862, 2024. 25

    2024

  8. [16]

    A foundation model for clinical-grade computational pathology and rare cancers detection

    Eugene V orontsov, Alican Bozkurt, Adam Casson, George Shaikovski, Michal Zelechowski, Kristen Severson, Eric Zimmermann, James Hall, Neil Tenenholtz, Nicolo Fusi, et al. A foundation model for clinical-grade computational pathology and rare cancers detection. Nature Medicine,...

    2024

  9. [17]

    AI-based anomaly detection for clinical-grade histopathological diagnostics

    Jonas Dippel, Niklas Prenißl, Julius Hense, Philipp Liznerski, Tobias Winterhoff, Simon Schallenberg, Marius Kloft, Oliver Buchstab, David Horst, Maximilian Alber, Lukas Ruff, Klaus-Robert M¨uller, and Frederick Klauschen. AI-based anomaly detection for clinical-grade histopat...

    2024

  10. [18]

    DeepSMILE: Contrastive self-supervised pre-training benefits MSI and HRD classification directly from H&E whole- slide images in colorectal and breast cancer

    Yoni Schirris, Efstratios Gavves, Iris Nederlof, Hugo Mark Horlings, and Jonas Teuwen. DeepSMILE: Contrastive self-supervised pre-training benefits MSI and HRD classification directly from H&E whole- slide images in colorectal and breast cancer. Medical Image Analysis, 79:102464, 2022

    2022

  11. [19]

    Rudolfv: A foundation model by pathologists for pathologists

    Jonas Dippel, Barbara Feulner, Tobias Winterhoff, Timo Milbich, Stephan Tietz, Simon Schallenberg, Gabriel Dernbach, Andreas Kunft, Simon Heinke, Marie-Lisa Eich, Julika Ribbat-Idel, Rosemarie Krupar, Philipp Anders, Niklas Prenißl, Philipp Jurmeister, David Horst, Lukas Ruff,...

    2024 arXiv

  12. [20]

    HEST-1k: A dataset for spatial transcriptomics and histology image analysis

    Guillaume Jaume, Paul Doucet, Andrew Song, Ming Yang Lu, Cristina Almagro P´erez, Sophia Wagner, Anurag Vaidya, Richard Chen, Drew Williamson, Ahrong Kim, et al. HEST-1k: A dataset for spatial transcriptomics and histology image analysis. Advances in Neural Information Process...

    2024

  13. [21]

    Real-world deployment of a fine-tuned pathology foundation model for lung cancer biomarker detection

    Gabriele Campanella, Neeraj Kumar, Swaraj Nanda, Siddharth Singi, Eugene Fluder, Ricky Kwan, Silke Muehlstedt, Nicole Pfarr, Peter J Sch ¨uffler, Ida H¨aggstr¨om, Noora Neittaanm ¨aki, Levent M Aky ¨urek, Alina Basnet, Tamara Jamaspishvili, Michel R Nasr, Matthew M Croken, Fre...

    2025

  14. [22]

    A benchmarking crisis in biomedical machine learning

    Faisal Mahmood. A benchmarking crisis in biomedical machine learning. Nature Medicine, 31(4): 1060–1060, 2025

    2025

  15. [23]

    eva: Evaluation framework for pathology foundation models

    kaiko.ai, Ioannis Gatopoulos, Nicolas K ¨anzig, Roman Moser, and Sebastian Ot ´alora. eva: Evaluation framework for pathology foundation models. In Medical Imaging with Deep Learning, 2024. Presented at: Medical Imaging with Deep Learning

    2024

  16. [24]

    Accelerating data processing and benchmarking of AI models for pathology

    Andrew Zhang, Guillaume Jaume, Anurag Vaidya, Tong Ding, and Faisal Mahmood. Accelerating data processing and benchmarking of AI models for pathology. arXiv preprint arXiv:2502.06750, 2025

    2025 arXiv

  17. [25]

    Molecular-driven foundation model for oncologic pathology

    Anurag Vaidya, Andrew Zhang, Guillaume Jaume, Andrew H Song, Tong Ding, Sophia J Wagner, Ming Y Lu, Paul Doucet, Harry Robertson, Cristina Almagro-Perez, et al. Molecular-driven foundation model for oncologic pathology. arXiv preprint arXiv:2501.16652, 2025

    2025 arXiv

  18. [26]

    Pathbench: A comprehensive comparison benchmark for pathology foundation models towards precision oncology

    Jiabo Ma, Yingxue Xu, Fengtao Zhou, Yihui Wang, Cheng Jin, Zhengrui Guo, Jianfeng Wu, On Ki Tang, Huajun Zhou, Xi Wang, et al. Pathbench: A comprehensive comparison benchmark for pathology foundation models towards precision oncology. arXiv preprint arXiv:2505.20202, 2025

    2025 arXiv

  19. [27]

    Benchmarking pathology foundation models: Adaptation strategies and scenarios

    Jaeung Lee, Jeewoo Lim, Keunho Byeon, and Jin Tae Kwak. Benchmarking pathology foundation models: Adaptation strategies and scenarios. Computers in Biology and Medicine, 190:110031, 2025

    2025

  20. [28]

    Foundation models – a panacea for artificial intelligence in pathology? arXiv preprint arXiv:2502.21264, 2025

    Nita Mulliqi, Anders Blilie, Xiaoyi Ji, Kelvin Szolnoky, Henrik Olsson, Sol Erika Boman, Matteo Titus, Geraldine Martinez Gonzalez, Julia Anna Mielcarz, Masi Valkonen, et al. Foundation models – a panacea for artificial intelligence in pathology? arXiv preprint arXiv:2502.21264, 2025

    2025

  21. [29]

    Evaluating vision and pathology foundation models for computational pathology: A comprehensive benchmark study

    Rohan Bareja, Francisco Carrillo-Perez, Yuanning Zheng, Marija Pizurica, Tarak Nath Nandi, Jeanne Shen, Ravi K Madduri, and Olivier Gevaert. Evaluating vision and pathology foundation models for computational pathology: A comprehensive benchmark study. medRxiv, pages 2025–05, 2025. 26

    2025

  22. [30]

    The impact of site-specific digital histology signatures on deep learning model accuracy and bias

    Frederick M Howard, James Dolezal, Sara Kochanny, Jefree Schulte, Heather Chen, Lara Heij, Dezheng Huo, Rita Nanda, Olufunmilayo I Olopade, Jakob N Kather, et al. The impact of site-specific digital histology signatures on deep learning model accuracy and bias. Nature Communic...

    2021

  23. [31]

    Toward explainable artificial intelligence for precision pathology

    Frederick Klauschen, Jonas Dippel, Philipp Keyl, Philipp Jurmeister, Michael Bockmayr, Andreas Mock, Oliver Buchstab, Maximilian Alber, Lukas Ruff, Gr´egoire Montavon, and Klaus-Robert M¨uller. Toward explainable artificial intelligence for precision pathology. Annual Review o...

    2024

  24. [32]

    Tackling the widespread and critical impact of batch effects in high-throughput data

    Jeffrey T Leek, Robert B Scharpf, H´ector Corrada Bravo, David Simcha, Benjamin Langmead, W Evan Johnson, Donald Geman, Keith Baggerly, and Rafael A Irizarry. Tackling the widespread and critical impact of batch effects in high-throughput data. Nature Reviews Genetics, 11(10):...

    2010

  25. [33]

    Why batch effects matter in omics data, and how to avoid them

    Wilson Wen Bin Goh, Wei Wang, and Limsoon Wong. Why batch effects matter in omics data, and how to avoid them. Trends in Biotechnology, 35(6):498–507, 2017

    2017

  26. [34]

    Are batch effects still relevant in the age of big data? Trends in Biotechnology, 40(9):1029–1040, 2022

    Wilson Wen Bin Goh, Chern Han Yong, and Limsoon Wong. Are batch effects still relevant in the age of big data? Trends in Biotechnology, 40(9):1029–1040, 2022

    2022

  27. [35]

    AI for radiographic COVID-19 detection selects shortcuts over signal

    Alex J DeGrave, Joseph D Janizek, and Su-In Lee. AI for radiographic COVID-19 detection selects shortcuts over signal. Nature Machine Intelligence, 3(7):610–619, 2021

    2021

  28. [36]

    Explainable AI reveals Clever Hans effects in unsupervised learning models

    Jacob Kauffmann, Jonas Dippel, Lukas Ruff, Wojciech Samek, Klaus-Robert M ¨uller, and Gr ´egoire Montavon. Explainable AI reveals Clever Hans effects in unsupervised learning models. Nature Machine Intelligence, 7:412—-422, 2025

    2025

  29. [37]

    Standardizing flow cytometry immunophenotyping analysis from the human immunophenotyping consortium

    Greg Finak, Marc Langweiler, Maria Jaimes, Mehrnoush Malek, Jafar Taghiyar, Yael Korin, Khadir Raddassi, Lesley Devine, Gerlinde Obermoser, Marcin L Pekalski, et al. Standardizing flow cytometry immunophenotyping analysis from the human immunophenotyping consortium. Scientific...

    2016

  30. [38]

    Diagnostics and correction of batch effects in large-scale proteomic studies: a tutorial

    Jelena ˇCuklina, Chloe H Lee, Evan G Williams, Tatjana Sajic, Ben C Collins, Mar´ıa Rodr´ıguez Mart´ınez, Varun S Sharma, Fabian Wendt, Sandra Goetze, Gregory R Keele, et al. Diagnostics and correction of batch effects in large-scale proteomic studies: a tutorial. Molecular Sy...

    2021

  31. [39]

    Simulating ComBat: how batch correction can lead to the systematic introduction of false positive results in DNA methylation microarray studies

    Tristan Zindler, Helge Frieling, Alexandra Neyazi, Stefan Bleich, and Eva Friedel. Simulating ComBat: how batch correction can lead to the systematic introduction of false positive results in DNA methylation microarray studies. BMC Bioinformatics, 21:1–15, 2020

    2020

  32. [40]

    Do histopathological foundation models eliminate batch effects? A comparative study

    Jonah K¨omen, Hannah Marienwald, Jonas Dippel, and Julius Hense. Do histopathological foundation models eliminate batch effects? A comparative study. 2024. Presented at: NeurIPS 2024 Workshop on Advancements In Medical Foundation Models: Explainability, Robustness, Security, a...

    2024

  33. [41]

    Current pathology foundation models are unrobust to medical center differences

    Edwin D de Jong, Eric Marcus, and Jonas Teuwen. Current pathology foundation models are unrobust to medical center differences. arXiv preprint arXiv:2501.18055, 2025

    2025 arXiv

  34. [42]

    Distilling foundation models for robust and efficient models in digital pathology

    Alexandre Filiot, Nicolas Dop, Oussama Tchita, Auriane Riou, R ´emy Dubois, Thomas Peeters, Daria Valter, Marin Scalbert, Charlie Saillard, Genevi`eve Robin, et al. Distilling foundation models for robust and efficient models in digital pathology. arXiv preprint arXiv:2501.16239, 2025

    2025 arXiv

  35. [43]

    Evaluating computational pathology foundation models for prostate cancer grading under distribution shifts

    Fredrik K Gustafsson and Mattias Rantalainen. Evaluating computational pathology foundation models for prostate cancer grading under distribution shifts. arXiv preprint arXiv:2410.06723, 2024

    2024 arXiv

  36. [44]

    Physical color calibration of digital pathology scanners for robust artificial intelligence–assisted cancer diagnosis.Modern Pathology, 38(5):100715, 2025

    Xiaoyi Ji, Richard Salmon, Nita Mulliqi, Umair Khan, Yinxi Wang, Anders Blilie, Henrik Olsson, Bodil Ginnerup Pedersen, Karina Dalsgaard Sørensen, Benedicte Parm Ulhøi, et al. Physical color calibration of digital pathology scanners for robust artificial intelligence–assisted ...

    2025

  37. [45]

    Medi: Metadata-guided diffusion models for mitigating biases in tumor classification

    David Jacob Drexlin, Jonas Dippel, Julius Hense, Niklas Prenißl, Gr´egoire Montavon, Frederick Klauschen, and Klaus-Robert M ¨uller. Medi: Metadata-guided diffusion models for mitigating biases in tumor classification. arXiv preprint arXiv:2506.17140, 2025

    2025 arXiv

  38. [46]

    Color transfer between images

    Erik Reinhard, Michael Adhikhmin, Bruce Gooch, and Peter Shirley. Color transfer between images. IEEE Computer Graphics and Applications, 21(5):34–41, 2001

    2001

  39. [47]

    Adjusting batch effects in microarray expression data using empirical bayes methods

    W Evan Johnson, Cheng Li, and Ariel Rabinovic. Adjusting batch effects in microarray expression data using empirical bayes methods. Biostatistics, 8(1):118–127, 2007

    2007

  40. [48]

    pyComBat, a Python tool for batch effects correction in high-throughput molecular data using empirical Bayes methods

    Abdelkader Behdenna, Maximilien Colange, Julien Haziza, Aryo Gema, Guillaume App´e, Chlo´e-Agathe Azencott, and Akp´eli Nordor. pyComBat, a Python tool for batch effects correction in high-throughput molecular data using empirical Bayes methods. BMC Bioinformatics, 24(1):459, 2023

    2023

  41. [49]

    Deep feature batch correction using ComBat for machine learning applications in computational pathology

    Pierre Murchan, Pilib ´O Broin, Anne-Marie Baird, Orla Sheils, and Stephen P Finn. Deep feature batch correction using ComBat for machine learning applications in computational pathology. Journal of Pathology Informatics, 15:100396, 2024

    2024

  42. [50]

    Diagnostic assessment of deep learning algorithms for detection of lymph node metastases in women with breast cancer

    Babak Ehteshami Bejnordi, Mitko Veta, Paul Johannes Van Diest, Bram Van Ginneken, Nico Karssemeijer, Geert Litjens, Jeroen AWM Van Der Laak, Meyke Hermsen, Quirine F Manson, Maschenka Balkenhol, et al. Diagnostic assessment of deep learning algorithms for detection of lymph no...

    2017

  43. [51]

    From detection of individual metastases to classification of lymph node status at the patient level: The CAMELYON17 challenge

    P´eter B´andi, Oscar Geessink, Quirine Manson, Marcory Van Dijk, Maschenka Balkenhol, Meyke Hermsen, Babak Ehteshami Bejnordi, Byungjae Lee, Kyunghyun Paeng, Aoxiao Zhong, et al. From detection of individual metastases to classification of lymph node status at the patient leve...

    2019

  44. [52]

    Universal encoding of pan-cancer histology by deep texture representations

    Daisuke Komura, Akihiro Kawabe, Keisuke Fukuta, Kyohei Sano, Toshikazu Umezaki, Hirotomo Koda, Ryohei Suzuki, Ken Tominaga, Mieko Ochi, Hiroki Konishi, Fumiya Masakado, Noriyuki Saito, Yasuyoshi Sato, Takumi Onoyama, Shu Nishida, Genta Furuya, Hiroto Katoh, Hiroharu Yamashita,...

    2022

  45. [53]

    Artificial intelligence for tumour tissue detection and histological regression grading in oesophageal adenocarcinomas: a retrospective algorithm development and validation study

    Yuri Tolkach, Lisa Marie Wolgast, Alexander Damanakis, Alexey Pryalukhin, Simon Schallenberg, Wolfgang Hulla, Marie-Lisa Eich, Wolfgang Schroeder, Anirban Mukhopadhyay, Moritz Fuchs, et al. Artificial intelligence for tumour tissue detection and histological regression grading...

    2023

  46. [54]

    Unmasking Clever Hans predictors and assessing what machines really learn.Nature Communications, 10(1):1096, 2019

    Sebastian Lapuschkin, Stephan W¨aldchen, Alexander Binder, Gr´egoire Montavon, Wojciech Samek, and Klaus-Robert M¨uller. Unmasking Clever Hans predictors and assessing what machines really learn.Nature Communications, 10(1):1096, 2019

    2019

  47. [55]

    Shortcut learning in deep neural networks

    Robert Geirhos, J¨orn-Henrik Jacobsen, Claudio Michaelis, Richard Zemel, Wieland Brendel, Matthias Bethge, and Felix A Wichmann. Shortcut learning in deep neural networks. Nature Machine Intelligence, 2(11):665–673, 2020

    2020

  48. [56]

    Katherine Hermann, Hossein Mobahi, Thomas Fel, and Michael C. Mozer. On the foundations of shortcut learning. 2024. Presented at: International Conference on Learning Representations (ICLR)

    2024

  49. [57]

    Demographic bias in misdiagnosis by computational pathology models

    Anurag Vaidya, Richard J Chen, Drew FK Williamson, Andrew H Song, Guillaume Jaume, Yuzhe Yang, Thomas Hartvigsen, Emma C Dyer, Ming Y Lu, Jana Lipkova, et al. Demographic bias in misdiagnosis by computational pathology models. Nature Medicine, 30(4):1174–1190, 2024

    2024

  50. [58]

    Detecting shortcut learning for fair medical AI using shortcut testing

    Alexander Brown, Nenad Tomasev, Jan Freyberg, Yuan Liu, Alan Karthikesalingam, and Jessica Schrouff. Detecting shortcut learning for fair medical AI using shortcut testing. Nature Communications, 14(1): 4314, 2023. 28

    2023

  51. [59]

    Durbin, Xiao-Cheng Wu, Antoinette Stroup, Jennifer Doherty, Stephen Schwartz, Charles Wiggins, Mark Damesyn, Linda Coyle, Lynne Penberthy, Georgia D

    Kevin De Angeli, Shang Gao, Ioana Danciu, Eric B. Durbin, Xiao-Cheng Wu, Antoinette Stroup, Jennifer Doherty, Stephen Schwartz, Charles Wiggins, Mark Damesyn, Linda Coyle, Lynne Penberthy, Georgia D. Tourassi, and Hong-Jun Yoon. Class imbalance in out-of-distribution datasets:...

    2022

  52. [60]

    A data augmentation methodology to reduce the class imbalance in histopathology images

    Rodrigo Escobar D´ıaz Guerrero, Lina Carvalho, Thomas Bocklitz, Juergen Popp, and Jos´e Luis Oliveira. A data augmentation methodology to reduce the class imbalance in histopathology images. Journal of Imaging Informatics in Medicine, 37(4):1767–1782, 2024

    2024

  53. [61]

    Tizhoosh and Liron Pantanowitz

    H.R. Tizhoosh and Liron Pantanowitz. On image search in histopathology. Journal of Pathology Informatics, 15:100375, 2024

    2024

  54. [62]

    Validation of histopathology foundation models through whole slide image retrieval

    Saghir Alfasly, Ghazal Alabtah, Sobhan Hemati, Krishna Rani Kalari, Joaquin J Garcia, and HR Tizhoosh. Validation of histopathology foundation models through whole slide image retrieval. Scientific Reports, 15(1):3990, 2025

    2025

  55. [63]

    Histopathology images-based deep learning prediction of prognosis and therapeutic response in small cell lung cancer

    Yibo Zhang, Zijian Yang, Ruanqi Chen, Yanli Zhu, Li Liu, Jiyan Dong, Zicheng Zhang, Xujie Sun, Jianming Ying, Dongmei Lin, et al. Histopathology images-based deep learning prediction of prognosis and therapeutic response in small cell lung cancer. npj Digital Medicine, 7(1):15, 2024

    2024

  56. [64]

    Depicter: Deep represen- tation clustering for histology annotation

    Eduard Chelebian, Chirstophe Avenel, Francesco Ciompi, and Carolina W¨ahlby. Depicter: Deep represen- tation clustering for histology annotation. Computers in Biology and Medicine, 170:108026, 2024

    2024

  57. [65]

    Lempitsky

    Yaroslav Ganin, Evgeniya Ustinova, Hana Ajakan, Pascal Germain, Hugo Larochelle, Franc ¸ois Laviolette, Mario Marchand, and Victor S. Lempitsky. Domain-adversarial training of neural networks. J. Mach. Learn. Res., 17:59:1–59:35, 2016

    2016

  58. [66]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. 2021. Presented at: International Conference on Mac...

    2021

  59. [67]

    Scaling up visual and vision-language representation learning with noisy text supervision

    Chao Jia, Yinfei Yang, Ye Xia, Yi-Ting Chen, Zarana Parekh, Hieu Pham, Quoc Le, Yun-Hsuan Sung, Zhen Li, and Tom Duerig. Scaling up visual and vision-language representation learning with noisy text supervision. 2021. Presented at: International Conference on Machine Learning (ICML)

    2021

  60. [68]

    Gpt-4 technical report

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. Gpt-4 technical report. arXiv preprint arXiv:2303.08774, 2023

    2023 arXiv

  61. [69]

    A comprehensive survey on pretrained foundation models: A history from bert to chatgpt

    Ce Zhou, Qian Li, Chen Li, Jun Yu, Yixin Liu, Guangjing Wang, Kai Zhang, Cheng Ji, Qiben Yan, Lifang He, et al. A comprehensive survey on pretrained foundation models: A history from bert to chatgpt. International Journal of Machine Learning and Cybernetics, pages 1–65, 2024

    2024

  62. [70]

    Large language models encode clinical knowledge

    Karan Singhal, Shekoofeh Azizi, Tao Tu, S Sara Mahdavi, Jason Wei, Hyung Won Chung, Nathan Scales, Ajay Tanwani, Heather Cole-Lewis, Stephen Pfohl, et al. Large language models encode clinical knowledge. Nature, 620(7972):172–180, 2023

    2023

  63. [71]

    Molecular simulations with a pretrained neural network and universal pairwise force fields

    Adil Kabylda, J Thorben Frank, Sergio Suarez Dou, Almaz Khabibrakhmanov, Leonardo Medrano Sandonas, Oliver T Unke, Stefan Chmiela, Klaus-Robert M¨uller, and Alexandre Tkatchenko. Molecular simulations with a pretrained neural network and universal pairwise force fields. ChemRx...

    2025 doi

  64. [72]

    Evaluation of the mace force field architecture: From medicinal chemistry to materials science

    D´avid P´eter Kov´acs, Ilyes Batatia, Eszter S´ara Arany, and G´abor Cs´anyi. Evaluation of the mace force field architecture: From medicinal chemistry to materials science. The Journal of Chemical Physics, 159 (4), 2023. 29

    2023

  65. [73]

    Foundation models defining a new era in vision: A survey and outlook

    Muhammad Awais, Muzammal Naseer, Salman Khan, Rao Muhammad Anwer, Hisham Cholakkal, Mubarak Shah, Ming-Hsuan Yang, and Fahad Shahbaz Khan. Foundation models defining a new era in vision: A survey and outlook. IEEE Transactions on Pattern Analysis and Machine Intelligence, 47(4...

    2025

  66. [74]

    A foundation model for spatial proteomics

    Muhammad Shaban, Yuzhou Chang, Huaying Qiu, Yao Yu Yeo, Andrew H Song, Guillaume Jaume, Yuchen Wang, Luca L Weishaupt, Tong Ding, Anurag Vaidya, et al. A foundation model for spatial proteomics. arXiv preprint arXiv:2506.03373, 2025

    2025 arXiv

  67. [75]

    Aligning machine and human visual representations across abstraction levels

    Lukas Muttenthaler, Klaus Greff, Frieda Born, Bernhard Spitzer, Simon Kornblith, Michael C Mozer, Klaus-Robert M¨uller, Thomas Unterthiner, and Andrew K Lampinen. Aligning machine and human visual representations across abstraction levels. arXiv preprint arXiv:2409.06509, 2024

    2024 arXiv

  68. [76]

    An intentional approach to managing bias in general purpose embedding models

    Wei-Hung Weng, Andrew Sellergen, Atilla P Kiraly, Alexander D’Amour, Jungyeon Park, Rory Pilgrim, Stephen Pfohl, Charles Lau, Vivek Natarajan, Shekoofeh Azizi, et al. An intentional approach to managing bias in general purpose embedding models. The Lancet Digital Health, 6(2):...

    2024

  69. [77]

    Attention-based deep multiple instance learning

    Maximilian Ilse, Jakub Tomczak, and Max Welling. Attention-based deep multiple instance learning. In International Conference on Machine Learning, 2018. Presented at: International Conference on Machine Learning (ICML)

    2018

  70. [78]

    Clinical-grade computational pathology using weakly supervised deep learning on whole slide images

    Gabriele Campanella, Matthew G Hanna, Luke Geneslaw, Allen Miraflor, Vitor Werneck Krauss Silva, Klaus J Busam, Edi Brogi, Victor E Reuter, David S Klimstra, and Thomas J Fuchs. Clinical-grade computational pathology using weakly supervised deep learning on whole slide images....

    2019

  71. [79]

    Transformer-based biomarker prediction from colorectal cancer histology: A large-scale multicentric study

    Sophia J Wagner, Daniel Reisenb ¨uchler, Nicholas P West, Jan Moritz Niehues, Jiefu Zhu, Sebastian Foersch, Gregory Patrick Veldhuizen, Philip Quirke, Heike I Grabsch, Piet A van den Brandt, et al. Transformer-based biomarker prediction from colorectal cancer histology: A larg...

    2023

  72. [80]

    A whole-slide foundation model for digital pathology from real-world data

    Hanwen Xu, Naoto Usuyama, Jaspreet Bagga, Sheng Zhang, Rajesh Rao, Tristan Naumann, Cliff Wong, Zelalem Gero, Javier Gonz´alez, Yu Gu, et al. A whole-slide foundation model for digital pathology from real-world data. Nature, 630(8015), 2024

    2024

  73. [81]

    A pathology foundation model for cancer diagnosis and prognosis prediction

    Xiyue Wang, Junhan Zhao, Eliana Marostica, Wei Yuan, Jietian Jin, Jiayu Zhang, Ruijiang Li, Hongping Tang, Kanran Wang, Yu Li, et al. A pathology foundation model for cancer diagnosis and prognosis prediction. Nature, 634(8035):970–978, 2024

    2024

  74. [82]

    Multimodal whole slide foundation model for pathology

    Tong Ding, Sophia J Wagner, Andrew H Song, Richard J Chen, Ming Y Lu, Andrew Zhang, Anurag J Vaidya, Guillaume Jaume, Muhammad Shaban, Ahrong Kim, et al. Multimodal whole slide foundation model for pathology. arXiv preprint arXiv:2411.19666, 2024

    2024 arXiv

  75. [83]

    Training language models to follow instructions with human feedback

    Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, et al. Training language models to follow instructions with human feedback. Advances in neural information processing systems, 35:2773...

    2022

  76. [84]

    Constitutional AI: Harmlessness from AI feedback

    Yuntao Bai, Saurav Kadavath, Sandipan Kundu, Amanda Askell, Jackson Kernion, Andy Jones, Anna Chen, Anna Goldie, Azalia Mirhoseini, Cameron McKinnon, et al. Constitutional AI: Harmlessness from AI feedback. arXiv preprint arXiv:2212.08073, 2022

    2022 arXiv

  77. [85]

    Chen, Chengkuan Chen, Yicong Li, Tiffany Y

    Richard J. Chen, Chengkuan Chen, Yicong Li, Tiffany Y . Chen, Andrew D. Trister, Rahul G. Krishnan, and Faisal Mahmood. Scaling vision transformers to gigapixel images via hierarchical self-supervised learning

  78. [86]

    Retccl: Clustering-guided contrastive learning for whole-slide image retrieval

    Xiyue Wang, Yuexi Du, Sen Yang, Jun Zhang, Minghui Wang, Jing Zhang, Wei Yang, Junzhou Huang, and Xiao Han. Retccl: Clustering-guided contrastive learning for whole-slide image retrieval. Medical Image Analysis, 83:102645, 2023. 30

    2023

  79. [87]

    Transformer-based unsupervised contrastive learning for histopathological image classification

    Xiyue Wang, Sen Yang, Jun Zhang, Minghui Wang, Jing Zhang, Wei Yang, Junzhou Huang, and Xiao Han. Transformer-based unsupervised contrastive learning for histopathological image classification. Medical Image Analysis, 81:102559, 2022

    2022

  80. [88]

    Benchmarking self- supervised learning on diverse pathology datasets

    Mingu Kang, Heon Song, Seonwook Park, Donggeun Yoo, and S ´ergio Pereira. Benchmarking self- supervised learning on diverse pathology datasets. 2023. Presented at: IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    2023

  81. [89]

    Towards large-scale training of pathology foundation models

    Nanne Aben, Edwin D de Jong, Ioannis Gatopoulos, Nicolas K ¨anzig, Mikhail Karasikov, Axel Lagr´e, Roman Moser, Joost van Doorn, Fei Tang, et al. Towards large-scale training of pathology foundation models. arXiv preprint arXiv:2404.15217, 2024

    2024 arXiv

  82. [90]

    Scaling self-supervised learning for histopathology with masked image modeling

    Alexandre Filiot, Ridouane Ghermi, Antoine Olivier, Paul Jacob, Lucas Fidon, Alice Kain, Charlie Saillard, and Jean-Baptiste Schiratti. Scaling self-supervised learning for histopathology with masked image modeling. medRxiv preprint medRxiv:2023.07.21.23292757v2, 2023

    2023

  83. [91]

    Phikon-v2, a large and public feature extractor for biomarker prediction

    Alexandre Filiot, Paul Jacob, Alice Mac Kain, and Charlie Saillard. Phikon-v2, a large and public feature extractor for biomarker prediction. arXiv preprint arXiv:2409.09173, 2024

    2024 arXiv

  84. [92]

    A visual-language foundation model for computational pathology

    Ming Y Lu, Bowen Chen, Drew FK Williamson, Richard J Chen, Ivy Liang, Tong Ding, Guillaume Jaume, Igor Odintsov, Long Phi Le, Georg Gerber, et al. A visual-language foundation model for computational pathology. Nature Medicine, 30(3):863–874, 2024

    2024

  85. [93]

    A vision–language foundation model for precision oncology

    Jinxi Xiang, Xiyue Wang, Xiaoming Zhang, Yinghua Xi, Feyisope Eweje, Yijiang Chen, Yuchen Li, Colin Bergstrom, Matthew Gopaulchan, Ted Kim, Kun-Hsing Yu, Sierra Willens, Francesca Maria Olguin, Jeffrey J Nirschl, Joel Neal, Maximilian Diehn, Sen Yang, and Ruijiang Li. A vision...

    2025

  86. [94]

    M. E. J. Masson and G. R. Loftus. Using confidence intervals for graphically based data interpretation. Canadian Journal of Experimental Psychology / Revue canadienne de psychologie exp´erimentale, 57(3): 203–220, 2003

    2003

  87. [95]

    Fix and J.L

    E. Fix and J.L. Hodges. Discriminatory Analysis: Nonparametric Discrimination: Consistency Properties. USAF School of Aviation Medicine, 1951

    1951

  88. [96]

    Cover and P

    T. Cover and P. Hart. Nearest neighbor pattern classification. IEEE Transactions on Information Theory, 13(1):21–27, 1967

    1967

  89. [97]

    Silhouettes: a graphical aid to the interpretation and validation of cluster analysis

    Peter J Rousseeuw. Silhouettes: a graphical aid to the interpretation and validation of cluster analysis. Journal of computational and applied mathematics, 20:53–65, 1987

    1987

  90. [98]

    Comparing partitions

    Lawrence Hubert and Phipps Arabie. Comparing partitions. Journal of classification, 2:193–218, 1985

    1985

  91. [99]

    Domain generalization in computational pathology: survey and guidelines

    Mostafa Jahanifar, Manahil Raza, Kesi Xu, Trinh Thi Le Vuong, Robert Jewsbury, Adam Shephard, Neda Zamanitajeddin, Jin Tae Kwak, Shan E Ahmed Raza, Fayyaz Minhas, et al. Domain generalization in computational pathology: survey and guidelines. ACM Computing Surveys, 57(11):1–37, 2025

    2025

  92. [100]

    Marron, David Borland, John Woosley, Xiaojun Guan, Charles Schmitt, and Nancy Thomas

    Marc Macenko, Marc Niethammer, J. Marron, David Borland, John Woosley, Xiaojun Guan, Charles Schmitt, and Nancy Thomas. A method for normalizing histology slides for quantitative analysis. 2009. Presented at: IEEE International Symposium on Biomedical Imaging: From Nano to Mac...

    2009

  93. [101]

    A threshold selection method from gray-level histograms

    Nobuyuki Otsu. A threshold selection method from gray-level histograms. IEEE Transactions on Systems, Man, and Cybernetics (SMC), 9(1):62–66, 1979

    1979

  94. [102]

    Bootstrap methods: another look at the jackknife

    Bradley Efron. Bootstrap methods: another look at the jackknife. In Breakthroughs in statistics: Method- ology and distribution, pages 569–593. Springer, 1992

    1992

  95. [103]

    Toy models of superposition

    Nelson Elhage, Tristan Hume, Catherine Olsson, Nicholas Schiefer, Tom Henighan, Shauna Kravec, Zac Hatfield-Dodds, Robert Lasenby, Dawn Drain, Carol Chen, Roger Grosse, Sam McCandlish, Jared Kaplan, Dario Amodei, Martin Wattenberg, and Christopher Olah. Toy models of superposi...

    2022

  96. [104]

    Imagenet-trained cnns are biased towards texture; increasing shape bias improves accuracy and robustness

    Robert Geirhos, Patricia Rubisch, Claudio Michaelis, Matthias Bethge, Felix A Wichmann, and Wieland Brendel. Imagenet-trained cnns are biased towards texture; increasing shape bias improves accuracy and robustness. 2019. Presented at: International conference on learning repre...

    2019

  97. [105]

    A Simple Framework for Contrastive Learning of Visual Representations

    Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A Simple Framework for Contrastive Learning of Visual Representations. 2020. Presented at: International Conference on Machine Learning (ICML)

    2020

  98. [106]

    Kunz, Matthew C.H

    Eugene V orontsov, Alican Bozkurt, Adam Casson, George Shaikovski, Michal Zelechowski, Siqi Liu, Philippe Mathieu, Alexander van Eck, Donghun Lee, Julian Viret, Eric Robert, Yi Kan Wang, Jeremy D. Kunz, Matthew C.H. Lee, Jan Bernhard, Ran A. Godrich, Gerard Oakley, Ewan Millar...

    2023 arXiv

  99. [107]

    From attribution maps to human-understandable explanations through concept relevance propagation

    Reduan Achtibat, Maximilian Dreyer, Ilona Eisenbraun, Sebastian Bosse, Thomas Wiegand, Wojciech Samek, and Sebastian Lapuschkin. From attribution maps to human-understandable explanations through concept relevance propagation. Nature Machine Intelligence, 5(9):1006–1019, 2023

    2023

  100. [108]

    Disentangled explana- tions of neural network predictions by finding relevant subspaces

    Pattarawat Chormai, Jan Herrmann, Klaus-Robert M¨uller, and Gr´egoire Montavon. Disentangled explana- tions of neural network predictions by finding relevant subspaces. IEEE Transactions on Pattern Analysis and Machine Intelligence, 46(11):7283–7299, 2024

    2024

  101. [109]

    Wasserstein distances made explainable: Insights into dataset shifts and transport phenomena

    Philip Naumann, Jacob Kauffmann, and Gr´egoire Montavon. Wasserstein distances made explainable: Insights into dataset shifts and transport phenomena. arXiv preprint arXiv:2505.06123, 2025. 32 Supplementary Notes These supplementary notes provide additional data, experiments, ...

    2025

  102. [111]

    tumor” and those fully outside as “normal

    on slide thumbnails and applied a patch-level minimum standard deviation of 8. Each patch was labeled with the medical center from which its slide originated. We additionally labeled those patches fully inside the metastasis annotations as “tumor” and those fully outside as “n...

  103. [2022]

    Presented at: IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

Pith tools

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