Pith. sign in

REVIEW 4 major objections 5 minor 32 references

Memory-Augmented Incomplete Multimodal Survival Prediction via Cross-Slide and Gene-Attentive Hypergraph Learning

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A hypergraph-plus-memory framework lifts multimodal cancer survival prediction past prior methods.

desk verdict Solid incremental engineering; the incomplete-modality claim is only tested for test-time missingness, so the headline is somewhat overstated. read the letter →

arxiv 2506.19324 v1 pith:QNDJX4HB submitted 2025-06-24 cs.CV

Mingcheng Qu , Guang Yang , Donglin Di , Yue Gao , Tonghua Su , Yang Song , Lei Fan This is my paper · ORCID
classification cs.CV
keywords survivalpredictionwholeslideimagesgenomicshypergraphlearningincompletemodalitymemorybankimbalancecancerprognosis
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

This paper tries to establish that cancer survival prediction improves when pathology slides and genomic data are fused through hypergraph structures rather than standard cross-attention, and that a memory bank can stand in for a missing modality at prediction time. The proposed model, M²Surv, treats patches from multiple slides as nodes in a hypergraph, capturing spatial structure inside each slide and shared morphology across slides, then links six gene groups densely to the pathology patches they attend to most. On five large public cancer cohorts, the authors report a mean concordance index (C-index, the standard ranking metric for survival models) of 0.701, ahead of prior multimodal methods by more than 2.3%. When either slides or genes are withheld at inference, the memory bank retrieves the closest complete patient pair and reports gains of 3.3% over pathology-only models and 7.9% over gene-only models. The result matters because real clinical records often have one missing assay, and most existing methods require both modalities.

What carries the argument

The load-bearing object is a pair of hypergraphs plus a memory bank. A hypergraph is a graph whose edges can link more than two nodes; here the nodes are pathology patches, plus six gene groups in the second stage. The multi-slide hypergraph builds intra-slide hyperedges from spatial neighborhoods and inter-slide hyperedges from patch-feature similarity, so both slide-specific staining variation and shared morphology are encoded. The gene-attentive hypergraph creates one dense hyperedge per gene group to its top-attending patches, which the paper argues rebalances the pathology-dominated fusion. The memory bank uses a momentum update during training and cosine-similarity retrieval with top-µ aggregation at inference to synthesize the missing modality. These components jointly carry the claim that performance gains come from higher-order slide interactions and balanced cross-modal fusion, not from a single architectural trick.

What would settle it

Train the same pipeline on cohorts where a fixed fraction of patients lacks one modality during training, then evaluate on a complete test set: if the memory-augmented model no longer beats the best unimodal baseline, the paper's central claim about incomplete modalities is falsified. A simpler observable check is that the reported 3.3% and 7.9% gaps should reproduce when the bank is seeded with complete training pairs.

Watch

Extended reading notes

Core claim

The paper's central claim is that M²Surv improves pathology-genomics survival prediction by replacing single-slide, cross-attention fusion with two hypergraph stages: a multi-slide hypergraph that first links spatially neighboring patches inside each slide and then links similar patches across slides, and a gene-attentive hypergraph that connects each of six gene groups to the pathology patches it attends to most. This second stage is the paper's answer to modality imbalance: instead of letting thousands of patches swamp a few hundred genes, each gene group reaches every relevant patch through dense hyperedges. A momentum-updated memory bank stores paired pathology-genomic features from complete training patients; at inference, if one modality is missing, the available modality retrieves the top matching stored pair and uses it to approximate the absent one. On five large cancer cohorts the authors report a mean C-index of 0.701, surpassing prior multimodal methods by more than 2.3%, and in missing-modality settings beating pathology-only models by 3.3% and gene-only models by 7.9%.

Load-bearing premise

The memory bank is built only from patients who have both slide and gene data during training, so the method's value for truly incomplete datasets, where some patients are missing data from the start, remains unshown.

Editorial extensions

If this is right

  • Multi-slide hypergraph aggregation should replace single-slide pooling in multimodal survival models, because the ablations tie each hyperedge type (intra-slide and inter-slide) to a measurable C-index improvement.
  • The memory bank is a cheap, generalizable remedy for missing modalities: it adds about 4.5% training time and 2.5% inference time and restores near-full performance for the two co-attention baselines it was added to.
  • A mean C-index of 0.701 across five cohorts, exceeding prior multimodal methods by more than 2.3%, implies the design transfers across tumor types rather than overfitting one site.
  • Gene-attentive hyperedges make the model's cross-modal reasoning inspectable, since the attention scores highlight which gene groups target which tissue regions.

Reading between the lines

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

  • The paper does not simulate missing modalities during training; a natural extension is to test whether the memory bank can be updated with partial pairs, since hospitals often lack one assay for some patients.
  • Because the best retrieval uses only the single most similar stored patient (µ=1), the quality of the approximation is bounded by how well the training cohort covers the test patient's biological profile, so rare subtypes are where this design is most likely to fail.
  • The dense gene-to-patch hyperedges could be reinterpreted as a weak form of cross-modal attention with an explicit balance mechanism; comparing against a loss function that directly penalizes modality imbalance would isolate where the improvement truly comes from.
  • If the multi-slide gain is driven mostly by fresh-frozen slides, archives that contain only formalin-fixed slides may see smaller benefits; the paper does not break down its 0.701 mean by slide type, so this boundary is untested.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes M2Surv, a multimodal survival prediction framework that combines a multi-slide hypergraph for pathology features, a gene-attentive hypergraph for cross-modal fusion, and a momentum-updated memory bank that retrieves previously seen paired pathology-genomic features to compensate for missing modalities at inference. The method is evaluated on five TCGA datasets with 5-fold cross-validation and C-Index as the metric, reporting a mean C-Index of 0.701 versus 0.685 for the strongest prior baseline and 3.3% and 7.9% gains in pathology-only and gene-only incomplete-modality scenarios, respectively.

Significance. If the reported results hold, M2Surv is a practically useful contribution to multimodal survival prediction, particularly for combining multiple FFPE and FF slides and for mitigating modality imbalance through gene-attentive hyperedges. The memory-bank design is lightweight and is also shown to be adaptable to other multimodal models, and the authors release code. These are genuine strengths. However, the empirical support for the headline claims is currently incomplete: the main comparison lacks significance testing, and the incomplete-modality evaluation does not cover the clinically motivating scenario of modalities missing during training, which is precisely the condition on which the memory mechanism depends.

major comments (4)
  1. [Table 1, Sec. 3.1] The headline improvement is not statistically established. The absolute mean C-Index gain over the strongest baseline PIBD is 0.016 (from 0.685 to 0.701), i.e., the abstract's "over 2.3%" is a relative rather than absolute improvement, and this is not stated. On most datasets the reported differences from several baselines are within one standard deviation, and no significance tests, confidence intervals, or paired comparisons are provided. In addition, some baselines are quoted from [29] rather than reproduced under the same folds and preprocessing, so the comparison may be confounded. Please provide per-dataset significance testing with all baselines evaluated under an identical protocol, or substantially temper the superiority claim.
  2. [Sec. 2.4, Sec. 3.2, Fig. 2c] The incomplete-modality evaluation only tests missingness at inference after training on complete pathology-genomic pairs. The memory bank stores only paired features from complete training samples, so a patient whose modality is missing during training cannot contribute a cross-modal entry for the retrieval in Eq. (5). This is exactly the scenario motivated in the Introduction (insufficient tissue, sequencing errors), yet the paper never simulates missing-at-training modalities. The Limitations paragraph in Sec. 4 concedes that retrieval efficacy depends on training-data coverage, which is the untested condition. The 3.3% pathology-only and 7.9% gene-only claims are therefore unvalidated for the motivating setting; a missing-at-training experiment, or a clear restriction of the claims, is required.
  3. [Sec. 3.2, Table 2, Fig. 2b] Several key hyperparameters are selected using the same datasets whose test performance is then reported as the final result. The hyperedge construction threshold λ is selected from the ablation table (5, 9, 25), the gene-attentive threshold β is described as "empirically set" to the top 5%, and the retrieval count μ is chosen as the best value in Fig. 2b. Without nested validation or a separate model-selection protocol, this selection on the evaluation data can inflate the reported mean C-Index. Please describe exactly how tuning was performed and provide sensitivity results at the selected hyperparameter values.
  4. [Table 1, Pathology and Genomic rows] The rows labeled "M²Surv (Ours)" in the Pathology and Genomic blocks are not comparable to the unimodal baselines in the same blocks. These rows correspond to a model trained on both modalities with one modality removed at inference and compensated by the memory bank, whereas ABMIL, TransMIL, MLP, and SNN are single-modality models trained from scratch. The comparison therefore conflates the effect of the memory mechanism with the quality of the unimodal representation. Please include a genuinely unimodal M2Surv variant, or label these rows explicitly as modality-dropout experiments rather than unimodal comparisons.
minor comments (5)
  1. [Abstract and Introduction] The phrase "Fresh Froze" should read "Fresh Frozen".
  2. [Eq. (2)] The distance condition is written as "ζpj − ζpk ∥2 ≤ δ"; a norm delimiter is missing in the typeset expression.
  3. [Table 2] The header "F use" appears to mean "Fusion", and row labels such as "HGNNM ulti9" contain spacing artifacts that should be corrected.
  4. [Eq. (1)] The definition of c_i is unconventional (c_i = 0 indicates event occurrence); although the loss terms are consistent with this coding, the meaning of "event status" should be stated explicitly to avoid confusion.
  5. [Fig. 2c and Sec. 3.2] The text uses "MCat" in the memory-bank experiment while Table 1 and the main text use "MCAT"; please standardize the notation.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: M2Surv is an empirical evaluation with explicit equations and external baselines; self-citations are background, not load-bearing.

full rationale

The paper's central claim is an empirical performance comparison (Table 1), not a derivation from first principles. The method is specified by explicit equations: intra-slide hyperedges (Eq. 2), inter-slide hyperedges (Eq. 3), gene-attentive hyperedges (Eq. 4), and memory-bank retrieval (Eq. 5). Nothing in these equations defines the output C-index in terms of the inputs; the memory bank stores unlabeled paired features and retrieves them by cosine similarity, without access to survival labels, so it cannot force the reported C-index by construction. The incomplete-modality evaluation drops a modality only at inference after full-pair training; this is a defined evaluation protocol, and the acknowledged limitation in Section 4 ('retrieval efficacy of memory banks depends on the coverage of training data') is a scope restriction, not a circular step. Self-citations to prior hypergraph works ([4,5,13,19,20,24,32]) are used as background or building blocks, not as the load-bearing justification for the reported improvements; the improvements are established by comparison with external baselines (MCAT, MOTCat, SurvPath, PIBD, etc.) using their released code. No fitted parameter is renamed as a prediction, and no uniqueness theorem is imported from the authors' own work. Therefore no circularity step can be exhibited with the required specificity.

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

Free parameters are the explicitly tuned hyperparameters of the architecture and memory bank; all are chosen on the same five TCGA datasets. Axioms are modeling assumptions imported from prior work or introduced by the design; the most fragile is the assumption that the memory bank, populated from complete training pairs, can compensate for missing modalities at inference. No new physical or metaphysical entities are invented, so the invented-entities ledger is empty.

free parameters (5)
  • Hyperedge construction threshold λ = 9 (chosen from {5,9,25}, Table 2)
    Determines the number of neighbor patches in both intra-WSI and inter-WSI hyperedges; λ=9 is the best-performing value in ablations, making the final result partly dependent on this tuned number.
  • Gene-attentive edge threshold β = top 5% (exact value not given)
    Set empirically to select the top 5% of patch-gene hyperedges, §2.3; changing this threshold changes the dense cross-modal graph.
  • Memory retrieval count μ = 1 (evaluated {1,3,5}, Fig.2b)
    The best test performance was achieved at μ=1 and this is the setting reported for incomplete-modality results.
  • Momentum coefficient θ = not reported
    Used in the memory bank update in §2.4; without a specified value the update procedure cannot be exactly reproduced.
  • Patch count per WSI N_k = 4096
    Each WSI is randomly subsampled to 4096 patches following prior work, §2.1; this is a design choice that affects feature distributions.
assumptions (5)
  • domain assumption FF and FFPE slides can be encoded by the same pretrained ResNet50 and aggregated in one hypergraph without explicit stain normalization or slide-type correction.
    The paper motivates slide-type heterogeneity in §1 but the pipeline treats all slides uniformly after patch extraction.
  • domain assumption Six functional gene groups, Tumor Suppression, Oncogenesis, Kinases, Cellular Differentiation, Transcription, and Cytokines, capture the relevant genomic signal for survival.
    Adopted from [3] in §2.1; no evidence is given that these six groups are sufficient.
  • domain assumption The memory bank assumes the training cohort covers the feature space of test patients well enough that cosine-nearest-neighbor retrieval yields useful missing-modality features.
    §2.4; the authors' Limitations section concedes retrieval depends on training-data coverage.
  • domain assumption Hypergraph convolution with the chosen thresholds preserves survival-relevant high-order structure in pathology and gene features.
    §2.2-§2.3; no theoretical guarantee is given.
  • standard math The standard negative log-likelihood survival loss (Eq.1) is appropriate for censored TCGA data with the authors' event/censoring convention.
    Adopted from [27] and used as the central training objective.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Memory-Augmented Incomplete Multimodal Survival Prediction via Cross-Slide and Gene-Attentive Hypergraph Learning." pith.science (2026). https://pith.science/paper/QNDJX4HB

@misc{pith2026250619324,
  author       = {Pith},
  title        = {Pith review of: Memory-Augmented Incomplete Multimodal Survival Prediction via Cross-Slide and Gene-Attentive Hypergraph Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QNDJX4HB}},
  note         = {Machine review of arXiv:2506.19324}
}
read the original abstract

Multimodal pathology-genomic analysis is critical for cancer survival prediction. However, existing approaches predominantly integrate formalin-fixed paraffin-embedded (FFPE) slides with genomic data, while neglecting the availability of other preservation slides, such as Fresh Froze (FF) slides. Moreover, as the high-resolution spatial nature of pathology data tends to dominate the cross-modality fusion process, it hinders effective multimodal fusion and leads to modality imbalance challenges between pathology and genomics. These methods also typically require complete data modalities, limiting their clinical applicability with incomplete modalities, such as missing either pathology or genomic data. In this paper, we propose a multimodal survival prediction framework that leverages hypergraph learning to effectively integrate multi-WSI information and cross-modality interactions between pathology slides and genomics data while addressing modality imbalance. In addition, we introduce a memory mechanism that stores previously learned paired pathology-genomic features and dynamically compensates for incomplete modalities. Experiments on five TCGA datasets demonstrate that our model outperforms advanced methods by over 2.3% in C-Index. Under incomplete modality scenarios, our approach surpasses pathology-only (3.3%) and gene-only models (7.9%). Code: https://github.com/MCPathology/M2Surv

Figures

Figures reproduced from arXiv: 2506.19324 by the authors.

Figure 2
Figure 2. a), Extra time consumption for memory bank during training and inference. b), Performance with varying retrieval top-µ. c), Performance with incomplete modality using the memory bank across models (M²Surv, MCat,and MOTCat). TCGA-A2-A04V Tumor Suppression Oncogenesis Kinases Visualization Cell Differentiation Transcription Cytokines RECQL4 GATA3 MLH1 NBN WRN FSTL3 HNRNPA2B1 BTG1 PCSK7 COX6C WNK4 PRKCE PINK1 IKBKE FLT… view at source ↗
Figure 3
Figure 3. Visualization for BRCA cases: The heatmap generated from cross￾attention scores, and the top five most influential genes are highlighted by gradient integral. Kaplan-Meier curves (Bottom) shows significant survival stratification (p<0.05 across all datasets) between high/low-risk groups (median split). Memory bank. We evaluated the time consumption, top-µ, and generaliz￾ability of our memory bank across five dataset… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

32 extracted references · 25 canonical work pages

  1. [20]

    arXiv preprint arXiv:2505.11997 (2025)

    Qu, M., Yang, G., Su, T., Gao, Y., Song, Y., Fan, L., et al.: Multimodal can- cer survival analysis via hypergraph learning with cross-modality rebalance. arXiv preprint arXiv:2505.11997 (2025)

    arXiv 2025

  2. [29]

    In: ICLR (2024)

    Zhang, Y., Xu, Y., Chen, J., Xie, F., Chen, H.: Prototypical information bot- tlenecking and disentangling for multimodal cancer survival prediction. In: ICLR (2024)

    work page 2024

  3. [1]

    In: ICCV

    Alonso, I., Sabater, A., Ferstl, D., Montesano, L., Murillo, A.C.: Semi-supervised semantic segmentation with pixel-level contrastive learning from a class-wise mem- ory bank. In: ICCV. pp. 8219–8228 (2021)

    work page 2021

  4. [2]

    Cancer Cell40(8), 865–878 (2022)

    Chen, R.J., Lu, M.Y., Williamson, D.F., et al.: Pan-cancer integrative histology- genomic analysis via multimodal deep learning. Cancer Cell40(8), 865–878 (2022)

    work page 2022

  5. [3]

    In: ICCV

    Chen, R.J., Lu, M.Y., et al.: Multimodal co-attention transformer for survival prediction in gigapixel whole slide images. In: ICCV. pp. 3995–4005 (2021).https: //doi.org/10.1109/ICCV48922.2021.00398

    arXiv 2021

  6. [4]

    IEEE Transactions on Image Processing31, 1149–1160 (2022)

    Di, D., Zhang, J., Lei, F., Tian, Q., Gao, Y.: Big-hypergraph factorization neural network for survival prediction from whole slide image. IEEE Transactions on Image Processing31, 1149–1160 (2022)

    work page 2022

  7. [5]

    IEEE Trans- actions on Pattern Analysis and Machine Intelligence45(5), 5800–5815 (2022)

    Di, D., Zou, C., Feng, Y., et al.: Generating hypergraph-based high-order represen- tations of whole-slide histopathological images for survival prediction. IEEE Trans- actions on Pattern Analysis and Machine Intelligence45(5), 5800–5815 (2022)

    work page 2022

  8. [6]

    In: MICCAI (2019),https://api.semanticscholar.org/CorpusID:198897112

    Dorent, R., Joutard, S., Modat, M., Ourselin, S., Vercauteren, T.K.M.: Hetero- modalvariationalencoder-decoderforjointmodalitycompletionandsegmentation. In: MICCAI (2019),https://api.semanticscholar.org/CorpusID:198897112

    work page 2019

Show all 32 references
  1. [7]

    IEEE Transactions on Medical Imaging42(5), 1401–1412 (2022)

    Fan,L.,Sowmya,A.,Meijering,E.,Song,Y.:Cancersurvivalpredictionfromwhole slide images with self-supervised learning and slide consistency. IEEE Transactions on Medical Imaging42(5), 1401–1412 (2022)

    2022

  2. [8]

    In: AAAI

    Feng, Y., You, H., Zhang, Z., Ji, R., Gao, Y.: Hypergraph neural networks. In: AAAI. vol. 33, pp. 3558–3565 (2019)

    2019

  3. [9]

    Prentice Hall PTR (1998)

    Haykin, S.: Neural networks: a comprehensive foundation. Prentice Hall PTR (1998)

    1998

  4. [10]

    In: CVPR

    He, K., Fan, H., Wu, Y., Xie, S., Girshick, R.: Momentum contrast for unsupervised visual representation learning. In: CVPR. pp. 9729–9738 (2020)

    2020

  5. [11]

    In: ICML

    Ilse,M.,Tomczak,J.,Welling,M.:Attention-baseddeepmultipleinstancelearning. In: ICML. pp. 2127–2136. PMLR (2018)

    2018

  6. [12]

    In: CVPR

    Jaume, G., Vaidya, A., Chen, R.J., Williamson, D.F., Liang, P.P., Mahmood, F.: Modeling dense multimodal interactions between biological pathways and histology for survival prediction. In: CVPR. pp. 11579–11590 (2024) 10 Mingcheng Qu et al

    2024

  7. [13]

    Pattern Recognition165, 111544 (2025)

    Jing, W., Wang, J., Di, D., et al.: Multi-modal hypergraph contrastive learning for medical image segmentation. Pattern Recognition165, 111544 (2025)

    2025

  8. [14]

    arXiv preprint arXiv:1609.02907 (2016)

    Kipf, T.N., Welling, M.: Semi-supervised classification with graph convolutional networks. arXiv preprint arXiv:1609.02907 (2016)

    2016 arXiv

  9. [15]

    NeurIPS30(2017)

    Klambauer, G., Unterthiner, T., Mayr, A., Hochreiter, S.: Self-normalizing neural networks. NeurIPS30(2017)

    2017

  10. [16]

    Nature biomedical engineer- ing5(6), 555–570 (2021)

    Lu, M.Y., Williamson, D.F., Chen, T.Y., et al.: Data-efficient and weakly super- vised computational pathology on whole-slide images. Nature biomedical engineer- ing5(6), 555–570 (2021)

    2021

  11. [17]

    Cancer Chemother Rep50(3), 163–170 (1966)

    Mantel, N., et al.: Evaluation of survival data and two new rank order statistics arising in its consideration. Cancer Chemother Rep50(3), 163–170 (1966)

    1966

  12. [18]

    Nature633(8028), 137–146 (2024)

    Nunes, L., Li, F., Wu, M., et al.: Prognostic genome and transcriptome signatures in colorectal cancers. Nature633(8028), 137–146 (2024)

    2024

  13. [19]

    In: BIBM

    Qu, M., Wu, Y., Di, D., Su, A., Su, T., Song, Y., Fan, L.: Boundary-guided learning for gene expression prediction in spatial transcriptomics. In: BIBM. pp. 445–450. IEEE (2024)

    2024

  14. [21]

    Science309(5743), 2010–2013 (2005)

    Raser, J.M., O’shea, E.K.: Noise in gene expression: origins, consequences, and control. Science309(5743), 2010–2013 (2005)

    2005

  15. [22]

    NeurIPS34, 2136–2147 (2021)

    Shao, Z., Bian, H., Chen, Y., Wang, Y., Zhang, J., Ji, X., et al.: Transmil: Trans- former based correlated multiple instance learning for whole slide image classifica- tion. NeurIPS34, 2136–2147 (2021)

    2021

  16. [23]

    In: Chung, A.C.S., Gee, J.C., Yushkevich, P.A., Bao, S

    Shen, Y., Gao, M.: Brain tumor segmentation on mri with missing modalities. In: Chung, A.C.S., Gee, J.C., Yushkevich, P.A., Bao, S. (eds.) Information Processing in Medical Imaging. pp. 417–428. Springer International Publishing, Cham (2019)

    2019

  17. [24]

    In: CVPR

    Tang, Q., Fan, L., Pagnucco, M., Song, Y.: Prototype-based image prompting for weakly supervised histopathological image segmentation. In: CVPR. pp. 30271– 30280 (2025)

    2025

  18. [25]

    stat 1050(20), 10–48550 (2017)

    Velickovic, P., Cucurull, G., Casanova, A., et al.: Graph attention networks. stat 1050(20), 10–48550 (2017)

    2017

  19. [26]

    In: ICCV

    Xu, Y., Chen, H.: Multimodal optimal transport-based co-attention transformer with global structure consistency for survival prediction. In: ICCV. pp. 21241– 21251 (2023)

    2023

  20. [27]

    Medical Image Analysis 65, 101789 (2020)

    Yao, J., Zhu, X., et al.: Whole slide images based cancer survival prediction using attention guided deep multiple instance learning networks. Medical Image Analysis 65, 101789 (2020)

    2020

  21. [28]

    In: Topological, Algebraic and Geometric Learning Workshops 2022

    Yi, K., Chen, J., Wang, Y.G., et al.: Approximate equivariance so (3) needlet convolution. In: Topological, Algebraic and Geometric Learning Workshops 2022. pp. 189–198. PMLR (2022)

    2022

  22. [30]

    Nature Machine Intelligence1(5), 236– 245 (2019)

    Zhang, Z., Chen, P., McGough, M., et al.: Pathologist-level interpretable whole- slide cancer diagnosis with deep learning. Nature Machine Intelligence1(5), 236– 245 (2019)

    2019

  23. [31]

    In: ICCV

    Zhou, F., Chen, H.: Cross-modal translation and alignment for survival analysis. In: ICCV. pp. 21485–21494 (2023)

    2023

  24. [32]

    In: CVPR

    Zhu, Z., Fan, L., Pagnucco, M., Song, Y.: Interpretable image classification via non-parametric part prototype learning. In: CVPR. pp. 9762–9771 (2025)

    2025

Pith tools

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