Aggregation Schemes for Single-Vector WSI Representation Learning in Digital Pathology

Ghazal Alabtah; H.R. Tizhoosh; Saghir Alfasly; Sobhan Hemati

REVIEW 4 major objections 5 minor 28 references

Aggregation Schemes for Single-Vector WSI Representation Learning in Digital Pathology

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Deep Fisher vectors win WSI retrieval benchmark

desk verdict Useful benchmark of WSI aggregation schemes, but missing variance and uneven supervision leave the headline ranking unsupported. read the letter →

arxiv 2501.17822 v2 pith:PVLYQRQS submitted 2025-01-29 eess.IV cs.AIcs.CVcs.IRq-bio.QM

Sobhan Hemati , Ghazal Alabtah , Saghir Alfasly , H.R. Tizhoosh This is my paper · ORCID

classification eess.IVcs.AIcs.CVcs.IRq-bio.QM

keywords HistopathologyWholeSlideImageWSIRepresentationSearchSetLearningSingle-VectorAggregation

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish which aggregation scheme should turn a whole slide image, represented as an unordered bag of patch embeddings, into a single fixed-size vector for WSI search. It benchmarks average and max pooling, Deep Sets, memory networks, focal attention, GMM Fisher vector, and deep sparse and binary Fisher vector on four TCGA cancer datasets (bladder, breast, kidney, colon) using k-NN retrieval. Its central claim is that the deep Fisher vector and its sparse and binary variants achieve the best overall accuracy and F1 scores in most settings, while also offering memory and speed advantages. If true, the practical recommendation for building single-vector WSI search indexes is to use the deep Fisher vector, with the binary version for fastest Hamming-distance search.

What carries the argument

The central object is the Fisher vector, defined as the gradient of a generative model's log-likelihood with respect to its parameters, aggregated over the set of patch embeddings. In the deep version, a VAE is trained with reconstruction, KL-divergence, classification, and gradient-regularization losses; the gradient of the reconstruction loss with respect to VAE parameters is power- and L2-normalized, and the top-M highest-variance dimensions are selected to form a single permutation-invariant WSI vector. Sparse and binary variants are produced by gradient sparsity and quantization regularization controlled by an alpha parameter, which the paper ablates.

What would settle it

A controlled experiment that trains Deep Sets, memory networks, focal attention, and the deep Fisher vector on identical patch embeddings with the same label supervision and the same hyperparameter tuning budget, and then removes the classification loss from the Fisher vector; if the Fisher vector no longer outperforms the others, the paper's central claim is falsified.

Watch

Extended reading notes

Core claim

The paper reports that when a WSI is represented as a set of patch embeddings from a pretrained DenseNet, the deep Fisher vector family outperforms other set-aggregation baselines for k-NN WSI retrieval. Across the four datasets, deep sparse and deep binary Fisher vectors rank first or near-first on accuracy and weighted F1, with the GMM Fisher vector also strong; the median-of-minimums non-aggregating baseline is generally beaten. The binary Fisher vector additionally gives the fastest search because it supports CPU-level XOR-based Hamming distance, while the sparse version reduces storage by producing sparse embeddings.

Load-bearing premise

The load-bearing premise is that all aggregation methods were trained and tuned with equivalent supervision and effort, but only the deep Fisher vector is documented as using subtype labels, so if the others received no matching label signal, the ranking could be an artifact of supervision rather than of aggregation quality.

Editorial extensions

If this is right

If the claim holds, deep Fisher vector aggregation should be the default choice for building single-vector WSI retrieval systems from patch embeddings.
Sparse Fisher vectors reduce storage costs for WSI indexes, and binary Fisher vectors make search dramatically faster through Hamming distance computed with XOR operations.
Non-aggregating patch-to-patch matching methods like the median of minimums are no longer necessary for competitive WSI retrieval, since single-vector aggregation matches or beats them.
The performance advantage is consistent across four different cancer types, suggesting the Fisher vector captures tissue morphology patterns that generalize across primary sites.
The alpha ablation indicates that adding gradient regularization improves embedding quality while imposing sparsity or binarization, pointing to a tunable trade-off between accuracy and efficiency.

Reading between the lines

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

The winning method was trained with cancer-subtype labels via a classification loss, while the paper does not document equivalent label supervision for Deep Sets, memory networks, or focal attention; the ranking could partly reflect supervision rather than aggregation quality.
A natural extension is to apply deep Fisher vector aggregation to patch embeddings from foundation models, which the paper notes are expensive but increasingly common; the same aggregation scheme may transfer without retraining.
The binary Fisher vector could be viewed as a learned hashing scheme, and its retrieval quality at lower dimensionality is worth testing against other learned hash codes.
The ablation results suggest the optimal alpha is dataset-dependent, so a practical deployment would need to tune sparsity or binarization strength per cohort rather than assuming one setting works everywhere.

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.

Desk Editor's Note

read the letter

Dear colleague,

Here's my read of arXiv:2501.17822.

The useful core is the head-to-head benchmark. It's the first comparison I know of that puts GMM Fisher vector, deep sparse/binary Fisher vector, Deep Sets, memory networks, focal attention, pooling, and Yottixel's median-of-minima on the same four TCGA sites for WSI retrieval. That's a legitimate service to the field, and the search-time table for sparse vs binary vs Yottixel is practically informative. The ablation on the α regularization for the deep Fisher vector variants is also a nice touch.

The problem is that the paper's main recommendation doesn't survive contact with its own tables. Section 3.2 says averages and standard deviations are computed over five splits, but Tables 1–4 report only point estimates. Winning margins are frequently one or two points: bladder 0.859 vs 0.857, breast 0.695 vs 0.692, kidney 0.802 vs 0.799. On colon, Deep Sets (Sum) beats every Fisher variant (0.843 vs 0.825). Without variance or significance tests, "in most cases deep Fisher vector variations achieve the best" is not supported. This is not a side issue; the ranking is the paper's only quantitative evidence.

The supervision asymmetry compounds it. The deep Fisher vector is trained with subtype labels via a classification loss, while the baselines' training protocols are not described. If the others were trained without labels, the comparison is partly supervised vs unsupervised. That alone could explain much of the gap. The paper needs to state what supervision each baseline received, or run a version where all methods get the same labels.

There are minor inconsistencies too: the conclusion mentions "two different deep models" but the experiments use DenseNet only, and the ablation text says skin/lung while the tables show breast/colon.

I'd give credit for the honest hedge in the conclusion ("may achieve the best overall performance") and for citing the authors' own prior method without hiding it. But the citation pattern is fine; the missing variance is the real blocker.

This paper deserves a serious referee because the benchmark is useful and the flaws are fixable, but it should not be accepted as is. Major revision: add per-fold results or std devs, significance tests, equal-supervision controls, and release code/data.

Best, [You]

Referee Report

4 major / 5 minor

Summary. The manuscript benchmarks ten aggregation schemes for converting a set of patch-level deep embeddings into a single whole-slide-image (WSI) vector, evaluating them with k-NN retrieval on four TCGA datasets (bladder, breast, kidney, colon). The methods include simple pooling, Deep Sets, memory networks, focal attention, GMM-based Fisher Vector, and the authors' previously proposed deep sparse and binary Fisher Vector. The paper also compares against Yottixel's median-of-minimum distance as a non-aggregating baseline. The central claim is that the deep Fisher Vector family achieves the best overall retrieval performance in most settings, with sparse and binary variants additionally providing storage and speed benefits. The conclusion suggests that deep sparse and binary Fisher Vectors may be the preferred aggregation choice for WSI search.

Significance. If the ranking were robust, this would be a practically useful comparison for the computational pathology community, which frequently needs to choose an aggregation strategy for WSI-level retrieval and classification. The paper has clear strengths: it covers a broad set of aggregation families, includes a non-aggregation retrieval baseline, uses four public datasets, and reports a timing comparison for the binary representation. It also makes a specific, falsifiable recommendation. However, the empirical support for the headline claim is currently incomplete: the promised variance information is absent, the supervision protocol is asymmetric across methods, and the per-dataset rankings are not fully consistent with the stated conclusion. These issues are fixable but require additional analysis and reporting before the central claim can be accepted.

major comments (4)

[Section 3.2 and Tables 1-4] Section 3.2 states: "We compute the average and standard deviation from the five splits and use these as performance indicators." Yet Tables 1-4 report only point estimates, with no standard deviations and no per-fold values. This is a load-bearing omission because the ranking that drives the paper's conclusion rests on small margins: on Bladder, Deep Sparse Fisher Vector is 0.859 versus 0.857 for Fisher Vector; on Breast, Deep Binary Fisher Vector is 0.695 versus 0.692 for Fisher Vector; on Kidney, Deep Binary Fisher Vector is 0.802 versus 0.799 for Fisher Vector. Without variance estimates or significance tests, the reader cannot determine whether these differences exceed fold-to-fold noise.
[Section 2.5 and Section 3.2] The supervision is not balanced across methods. Section 2.5 and Fig. 3 show that the deep Fisher Vector VAE is trained with a classification loss using WSI subtype labels, injecting label information into the embedding. Section 3.2 does not specify whether Deep Sets, Memory Network, Focal Attention, and GMM Fisher Vector were trained with equivalent label supervision, the same amount of hyperparameter tuning, or the same number of training epochs. If only the deep Fisher Vector receives subtype labels, the observed advantage could reflect the presence of supervision rather than the aggregation scheme. The revision should document the full training protocol for every baseline, or add an unsupervised/self-supervised condition for the deep Fisher Vector.
[Section 4 and Section 5, Tables 1-4] The cross-dataset generalization claim in Section 5 is not fully supported by Tables 1-4. On Colon, Deep Sets (Sum) achieves accuracy 0.843, which is higher than all Fisher Vector variants (Fisher Vector 0.825, Deep Sparse 0.807, Deep Binary 0.809); Deep Sets (Max) at 0.839 also beats the Fisher variants. On Breast, the advantage of Deep Binary Fisher Vector over Fisher Vector is 0.003, and on Bladder the advantage of Deep Sparse Fisher Vector over Fisher Vector is 0.002. The conclusion that the Fisher Vector family has "robust ability" and "versatility" requires an explicit aggregation of results across datasets and metrics, such as average ranks, plus a significance test; the current narrative relies on "in most cases" without quantifying how often or by how much.
[Section 3.2] The value of k in the k-NN search is never specified. Accuracy and F1 scores in a k-NN retrieval evaluation depend heavily on k, and without this number the results in Tables 1-4 cannot be reproduced or properly interpreted. Please report the k value used for each table, and preferably show sensitivity to k.

minor comments (5)

[Section 3.1] The bladder dataset is described as "The skin data contains 457 cases," which appears to be a typo for "The bladder data." Please correct it.
[Section 5] The conclusion states that the methods were evaluated "for different datasets and two different deep models," but Section 3.1 and the results only describe DenseNet as the feature extractor. Please either add the second model or correct the sentence.
[Section 4.1] The ablation text says the study was performed "on skin and lung datasets," but Tables 6 and 7 report results for breast and colon. Please align the text with the actual datasets.
[Table 5] The timing comparison in Table 5 does not report hardware, number of runs, or whether the times are averaged; moreover, Yottixel is evaluated only at dimension 3,000 while the Fisher Vector methods are evaluated at 30,000, 3,000, and 300. Please add the experimental conditions so the speed claim is interpretable.
[Throughout] There are several typographical and grammatical issues, e.g., "a set of pacthes" in the Introduction, "the search speed will increased" in the Introduction, and inconsistent capitalization of "Colon" in dataset names. A careful proofreading pass is needed.

Circularity Check

1 steps flagged · score 6.0 of 10

Deep Fisher Vector's 'best aggregation' result is partly forced by label supervision: its embeddings are trained with a classifier on the same subtype labels used to measure retrieval accuracy, unlike the reported baselines.

fitted input called prediction [Section 2.5 (Deep Fisher Vector and its variations); evaluation in Section 3.2]
"To obtain the deep sparse, and binary Fisher Vectors, one can employ a simple VAE where its latent space is also connected to a dense layer with a softmax activation function that outputs a probability distribution representing the predicted WSI class. This classifier injects class information (i.e., diagnosis) into the final WSI embeddings."

Section 3.2 evaluates k-NN search with 'accuracy, Macro F1 score, and weighted F1 score' over the same subtype/diagnosis classes. Only the deep Fisher Vector family is described as being trained with a classification loss over those classes; no equivalent supervised training is reported for the pooling, Deep Sets, memory, focal-attention, or GMM baselines. The conclusion that 'deep sparse and binary Fisher Vector may achieve the best overall performance' (Section 5) therefore conflates the aggregation scheme with the label supervision: the embedding has been fitted to predict the very labels later used as retrieval ground truth.

full rationale

The paper is an empirical benchmark, not a mathematical derivation, so most circularity patterns do not apply. The authors' prior work [10] introduces the deep sparse and binary Fisher Vectors, but the present paper provides new experiments rather than relying only on that citation, so the self-citation is not by itself circular. The main circular component is the supervision asymmetry: the deep Fisher Vector's VAE is explicitly trained with a classifier that injects subtype/diagnosis information into the embedding, while the retrieval evaluation measures accuracy and F1 over those same subtype labels. This means the reported advantage of the deep Fisher Vector family over other aggregators is not a clean comparison of aggregation schemes; it is partly a consequence of fitting the embedding to the evaluation taxonomy. The paper also promises standard deviations from five folds but reports only point estimates, which is a serious correctness/reproducibility risk but not itself a circularity issue. Overall, the central 'best aggregation' claim partially reduces to the label-fitting input, warranting a score of 6 rather than a higher score because held-out folds and distance-based retrieval still provide some independent content.

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

This is an empirical benchmark, so the ledger is dominated by unstated implementation choices rather than mathematical axioms. The free parameters are hyperparameters and training details that are not reported; without them the comparison cannot be reproduced. The key domain assumptions are that TCGA subtype labels are a valid similarity ground truth and that ImageNet-pretrained DenseNet features are a reasonable morphological basis. No new entities are introduced.

free parameters (4)

alpha (regularization weight for sparse/binary Fisher Vector gradient loss) = 0.0 to 1e-5 in ablation; no single selected value reported
Controls sparsity and quantization gradient loss; the paper tests a range of values but never states the value used for the main tables or whether it was tuned per dataset.
Top M Fisher vector dimensions = 30,000, 3,000, and 300 in timing experiment; main tables not specified
The number of selected parameters with highest gradient variance affects dimensionality, speed, and likely accuracy; the main results do not report which M was used.
k in k-NN search = not reported
k-NN accuracy depends on k, and the paper does not report k or how it was selected, making the metric values hard to interpret or reproduce.
Hyperparameters for baseline aggregators = not reported
Implementation choices for GMM components, Deep Sets architecture, memory units, focal factor, and training schedules are not given, so comparisons may reflect unstated tuning differences.

assumptions (4)

domain assumption TCGA subtype labels are a valid ground truth for WSI retrieval similarity.
Retrieval accuracy and F1 are computed against subtype classes; if subtypes are noisy or not the clinically relevant grouping, the ranking may not generalize (Section 3.1).
domain assumption ImageNet-pretrained DenseNet patch embeddings capture morphology relevant for WSI similarity.
All aggregation methods operate on the same fixed patch embeddings, so their quality bounds the entire comparison (Section 3.1).
standard math The deep Fisher vector gradient aggregation and normalization formulas from prior work are correct.
The paper relies on Ref. [10] for the deep sparse and binary Fisher vector construction without re-deriving it (Section 2.5).
ad hoc to paper All methods were trained and tuned comparably.
This is not stated anywhere; the paper presents results without training details, so this assumption is necessary for the comparison to be valid (Section 3.2).

how reviews work

0 comments

Cite this review

Pith. "Pith review of Aggregation Schemes for Single-Vector WSI Representation Learning in Digital Pathology." pith.science (2026). https://pith.science/paper/PVLYQRQS

@misc{pith2026250117822,
  author       = {Pith},
  title        = {Pith review of: Aggregation Schemes for Single-Vector WSI Representation Learning in Digital Pathology},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PVLYQRQS}},
  note         = {Machine review of arXiv:2501.17822}
}

read the original abstract

A crucial step to efficiently integrate Whole Slide Images (WSIs) in computational pathology is assigning a single high-quality feature vector, i.e., one embedding, to each WSI. With the existence of many pre-trained deep neural networks and the emergence of foundation models, extracting embeddings for sub-images (i.e., tiles or patches) is straightforward. However, for WSIs, given their high resolution and gigapixel nature, inputting them into existing GPUs as a single image is not feasible. As a result, WSIs are usually split into many patches. Feeding each patch to a pre-trained model, each WSI can then be represented by a set of patches, hence, a set of embeddings. Hence, in such a setup, WSI representation learning reduces to set representation learning where for each WSI we have access to a set of patch embeddings. To obtain a single embedding from a set of patch embeddings for each WSI, multiple set-based learning schemes have been proposed in the literature. In this paper, we evaluate the WSI search performance of multiple recently developed aggregation techniques (mainly set representation learning techniques) including simple average or max pooling operations, Deep Sets, Memory networks, Focal attention, Gaussian Mixture Model (GMM) Fisher Vector, and deep sparse and binary Fisher Vector on four different primary sites including bladder, breast, kidney, and Colon from TCGA. Further, we benchmark the search performance of these methods against the median of minimum distances of patch embeddings, a non-aggregating approach used for WSI retrieval.

Figures

Figures reproduced from arXiv: 2501.17822 by the authors.

**Figure 2.** Aggregation of patch embeddings can provide a single-vector WSI repre [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** VAE Training architecture [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: Aggregation to generate a single-vector WSI embedding. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

28 extracted references · 19 canonical work pages

[1]

On image search in histopathology

Hamid R Tizhoosh and Liron Pantanowitz. On image search in histopathology. Journal of Pathology Informatics , page 100375, 2024

work page 2024
[2]

Shah, Joaquin J

Isaiah Lahr, Saghir Alfasly, Peyman Nejat, Jibran Khan, Luke Kottom, Vaishnavi Kumbhar, Areej Alsaafin, Abubakr Shafique, Sobhan Hemati, Ghazal Alabtah, Nneka Comfere, Dennis Murphree, Aaron Mangold, Saba Yasir, Chady Meroueh, Lisa Boardman, Vijay H. Shah, Joaquin J. Garcia, and H.R. Tizhoosh. Analy- sis and validation of image search engines in histopath...

work page 2024
[3]

Yottixel–an image search engine for large archives of histopathology whole slide images

Shivam Kalra, Hamid R Tizhoosh, Charles Choi, Sultaan Shah, Phedias Diaman- dis, Clinton JV Campbell, and Liron Pantanowitz. Yottixel–an image search engine for large archives of histopathology whole slide images. Medical Image Analysis , 65:101757, 2020

work page 2020
[4]

Deep sets

Manzil Zaheer, Satwik Kottur, Siamak Ravanbakhsh, Barnabas Poczos, Russ R Salakhutdinov, and Alexander J Smola. Deep sets. Advances in neural information processing systems, 30, 2017

2017
[5]

Learn- ing permutation invariant representations using memory networks

Shivam Kalra, Mohammed Adnan, Graham Taylor, and Hamid R Tizhoosh. Learn- ing permutation invariant representations using memory networks. In Computer α Breast (Sparse) Breast (Binary) Accuracy Macro ¯F 1 Weighted ¯F 1 Accuracy Macro ¯F 1 Weighted ¯F 1 0.0 0.690 0.151 0.651 0.685 0.126 0.640 0.1 0.693 0.149 0.650 0.703 0.131 0.653 0.01 0.698 0.140 0.651 0...

work page 2020
[6]

Pay attention with focus: A novel learning scheme for classification of whole slide images

Shivam Kalra, Mohammed Adnan, Sobhan Hemati, Taher Dehkharghanian, Shahryar Rahnamayan, and Hamid R Tizhoosh. Pay attention with focus: A novel learning scheme for classification of whole slide images. In Medical Image Computing and Computer Assisted Intervention–MICCAI 2021: 24th International Conference, Strasbourg, France, September 27–October 1, 2021,...

work page 2021
[7]

Representation learn- ing of histopathology images using graph neural networks

Mohammed Adnan, Shivam Kalra, and Hamid R Tizhoosh. Representation learn- ing of histopathology images using graph neural networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops , pages 988–989, 2020

work page 2020
[8]

Exploiting generative models in discrim- inative classifiers

Tommi Jaakkola and David Haussler. Exploiting generative models in discrim- inative classifiers. In Advances in neural information processing systems , pages 487–493, 1999

work page 1999

Show all 28 references

[9]

Fisher kernels on visual vocabularies for image categorization

Florent Perronnin and Christopher Dance. Fisher kernels on visual vocabularies for image categorization. In 2007 IEEE conference on computer vision and pattern recognition, pages 1–8. IEEE, 2007

2007
[10]

Learning bi- nary and sparse permutation-invariant representations for fast and memory effi- cient whole slide image search

Sobhan Hemati, Shivam Kalra, Morteza Babaie, and HR Tizhoosh. Learning bi- nary and sparse permutation-invariant representations for fast and memory effi- cient whole slide image search. Computers in Biology and Medicine , 162:107026, 2023

2023
[11]

Visual categorization with bags of keypoints

Gabriella Csurka, Christopher Dance, Lixin Fan, Jutta Willamowski, and C´ edric Bray. Visual categorization with bags of keypoints. In Workshop on statistical learning in computer vision, ECCV , volume 1, pages 1–2. Prague, 2004

2004
[12]

Cnn and deep sets for end-to-end whole slide image repre- sentation learning

Sobhan Hemati, Shivam Kalra, Cameron Meaney, Morteza Babaie, Ali Ghodsi, and Hamid Tizhoosh. Cnn and deep sets for end-to-end whole slide image repre- sentation learning. In Medical Imaging with Deep Learning, pages 301–311. PMLR, 2021

2021
[13]

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 , pages 2127–
[14]

Deep quantization: Encoding convolutional activations with deep generative model

Zhaofan Qiu, Ting Yao, and Tao Mei. Deep quantization: Encoding convolutional activations with deep generative model. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages 6759–6768, 2017

2017
[15]

Adver- sarial fisher vectors for unsupervised representation learning

Shuangfei Zhai, Walter Talbott, Carlos Guestrin, and Joshua Susskind. Adver- sarial fisher vectors for unsupervised representation learning. Advances in Neural Information Processing Systems, 32, 2019

2019
[16]

Improving the fisher kernel for large-scale image classification

Florent Perronnin, Jorge S´ anchez, and Thomas Mensink. Improving the fisher kernel for large-scale image classification. In Computer Vision–ECCV 2010: 11th European Conference on Computer Vision, Heraklion, Crete, Greece, September 5-11, 2010, Proceedings, Part IV 11 , pages ...

2010
[17]

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

Hanwen Xu, Naoto Usuyama, Jaspreet Bagga, Sheng Zhang, Rajesh Rao, Tris- tan 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):181–188, 2024

2024
[18]

Prism: A multi-modal generative foundation model for slide-level histopathology

George Shaikovski, Adam Casson, Kristen Severson, Eric Zimmermann, Yi Kan Wang, Jeremy D Kunz, Juan A Retamero, Gerard Oakley, David Klimstra, Christopher Kanan, et al. Prism: A multi-modal generative foundation model for slide-level histopathology. arXiv preprint arXiv:2405.1...

2024 arXiv
[19]

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
[20]

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
[21]

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
[22]

Current pathology foun- dation models are unrobust to medical center differences

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

2025 arXiv
[23]

Are the latent representa- tions of foundation models for pathology invariant to rotation? arXiv preprint arXiv:2412.11938, 2024

Matouˇ s Elphick, Samra Turajlic, and Guang Yang. Are the latent representa- tions of foundation models for pathology invariant to rotation? arXiv preprint arXiv:2412.11938, 2024

2024 arXiv
[24]

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

2025
[25]

Densely connected convolutional networks

Gao Huang, Zhuang Liu, Laurens Van Der Maaten, and Kilian Q Weinberger. Densely connected convolutional networks. In Proceedings of the IEEE conference on computer vision and pattern recognition , pages 4700–4708, 2017

2017
[26]

Artificial intelligence and digital pathology: challenges and opportunities

Hamid Reza Tizhoosh and Liron Pantanowitz. Artificial intelligence and digital pathology: challenges and opportunities. Journal of pathology informatics, 9, 2018

2018
[27]

Deep learning for digital pathology image analysis: A comprehensive tutorial with selected use cases.Journal of pathol- ogy informatics, 7, 2016

Andrew Janowczyk and Anant Madabhushi. Deep learning for digital pathology image analysis: A comprehensive tutorial with selected use cases.Journal of pathol- ogy informatics, 7, 2016

2016
[28]

Image classification with the fisher vector: Theory and practice

Jorge S´ anchez, Florent Perronnin, Thomas Mensink, and Jakob Verbeek. Image classification with the fisher vector: Theory and practice. International journal of computer vision, 105(3):222–245, 2013

2013

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.

[1] [1]

On image search in histopathology

Hamid R Tizhoosh and Liron Pantanowitz. On image search in histopathology. Journal of Pathology Informatics , page 100375, 2024

work page 2024

[2] [2]

Shah, Joaquin J

Isaiah Lahr, Saghir Alfasly, Peyman Nejat, Jibran Khan, Luke Kottom, Vaishnavi Kumbhar, Areej Alsaafin, Abubakr Shafique, Sobhan Hemati, Ghazal Alabtah, Nneka Comfere, Dennis Murphree, Aaron Mangold, Saba Yasir, Chady Meroueh, Lisa Boardman, Vijay H. Shah, Joaquin J. Garcia, and H.R. Tizhoosh. Analy- sis and validation of image search engines in histopath...

work page 2024

[3] [3]

Yottixel–an image search engine for large archives of histopathology whole slide images

Shivam Kalra, Hamid R Tizhoosh, Charles Choi, Sultaan Shah, Phedias Diaman- dis, Clinton JV Campbell, and Liron Pantanowitz. Yottixel–an image search engine for large archives of histopathology whole slide images. Medical Image Analysis , 65:101757, 2020

work page 2020

[4] [4]

Deep sets

Manzil Zaheer, Satwik Kottur, Siamak Ravanbakhsh, Barnabas Poczos, Russ R Salakhutdinov, and Alexander J Smola. Deep sets. Advances in neural information processing systems, 30, 2017

2017

[5] [5]

Learn- ing permutation invariant representations using memory networks

Shivam Kalra, Mohammed Adnan, Graham Taylor, and Hamid R Tizhoosh. Learn- ing permutation invariant representations using memory networks. In Computer α Breast (Sparse) Breast (Binary) Accuracy Macro ¯F 1 Weighted ¯F 1 Accuracy Macro ¯F 1 Weighted ¯F 1 0.0 0.690 0.151 0.651 0.685 0.126 0.640 0.1 0.693 0.149 0.650 0.703 0.131 0.653 0.01 0.698 0.140 0.651 0...

work page 2020

[6] [6]

Pay attention with focus: A novel learning scheme for classification of whole slide images

Shivam Kalra, Mohammed Adnan, Sobhan Hemati, Taher Dehkharghanian, Shahryar Rahnamayan, and Hamid R Tizhoosh. Pay attention with focus: A novel learning scheme for classification of whole slide images. In Medical Image Computing and Computer Assisted Intervention–MICCAI 2021: 24th International Conference, Strasbourg, France, September 27–October 1, 2021,...

work page 2021

[7] [7]

Representation learn- ing of histopathology images using graph neural networks

Mohammed Adnan, Shivam Kalra, and Hamid R Tizhoosh. Representation learn- ing of histopathology images using graph neural networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops , pages 988–989, 2020

work page 2020

[8] [8]

Exploiting generative models in discrim- inative classifiers

Tommi Jaakkola and David Haussler. Exploiting generative models in discrim- inative classifiers. In Advances in neural information processing systems , pages 487–493, 1999

work page 1999