Context Matters: Query-aware Dynamic Long Sequence Modeling of Gigapixel Images

Hao Chen; Jiabo Ma; Jinzhuo Wang; Lishuang Feng; Qichen Sun; Zhengrui Guo

REVIEW 3 major objections 6 minor 1 cited by

Context Matters: Query-aware Dynamic Long Sequence Modeling of Gigapixel Images

T0 review · 3 major / 6 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read Querent claims a query-aware selective attention mechanism keeps full self-attention's expressiveness within a bounded approximation error while reducing WSI modeling cost from O(N^2) to O(NK), and reports state-of-the-art results on 11…

desk verdict The empirical core is honest and the query-adaptive selection idea is worth a look, but the paper's central complexity claim is contradicted by its own appendix and the approximation theorem does not go through. read the letter →

arxiv 2501.18984 v2 pith:WHMIXNOW submitted 2025-01-31 cs.CV

Zhengrui Guo , Qichen Sun , Jiabo Ma , Lishuang Feng , Jinzhuo Wang , Hao Chen This is my paper · ORCID

classification cs.CV

keywords wholeslideimageanalysisquery-awareattentionself-attentionapproximationcomputationalpathologygigapixelimagesmultipleinstancelearninglongsequencemodelingsurvivalprediction

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

Querent is a framework for analyzing whole-slide pathology images that tries to keep the modeling power of full transformer self-attention while escaping its quadratic cost. The central move is to let each patch decide, through cheap region-level summaries, which parts of a gigapixel slide are worth attending to, and then run exact attention only on those selected regions. The paper claims this query-aware selection produces a theoretically bounded approximation of full self-attention and reduces complexity from O($N^{2}$) to O(NK), and reports state-of-the-art accuracy, AUC, F1, and C-index on 11 public datasets across biomarker prediction, mutation prediction, subtyping, and survival analysis. If correct, the method would make whole-slide attention practical at 100,000-patch scale with roughly 1 GB memory and 500 GFLOPs, versus 37 GB and 10,000 GFLOPs for full attention.

What carries the argument

The load-bearing object is the region-level min/max metadata pair (m_min_i, m_max_i) and the score si defined in Eq. (3). The min and max vectors are meant to bracket all patch features in a region, so that the maximum absolute dot product between a projected query and these two summary vectors upper-bounds the relevance of any patch in the region. That bound is what lets the top-K selection in Eq. (4) act as a cheap proxy for full attention, and it is the first step of Lemma B.4 that feeds Theorem B.6.

What would settle it

On real WSI patch features, compute for many (query, region) pairs the gap si minus the true maximum interaction max_j |<q, x_j>|; if a non-negligible fraction of gaps are negative, the selection rule can miss the truly relevant regions, contradicting Lemma B.4. A simpler end-to-end check is to compare the regions Querent selects to the true top-K regions from full attention on a held-out slide, since low overlap means the bound is not doing the work.

Watch

Extended reading notes

Core claim

The paper's central claim is that dynamic, query-dependent sparsification does not sacrifice full self-attention's expressiveness. For each query patch q, Querent partitions the slide into regions, computes per-region element-wise min and max summary vectors, projects them with learned networks, and scores every region by si = max(|<q_hat,m_min_hat>|, |<q_hat,m_max_hat>|). The top-K regions are then used to compute dense multi-head self-attention between the query and all patches in those regions, giving each patch its own attention pattern. The paper proves Theorem B.6, a Frobenius-norm bound ||A-B||_F <= (2 + B/sqrt(d)) epsilon between the query-aware attention matrix and the full self-attention matrix, under conditions on region diameter, spatial separation, attention decay exponent, and embedding dimension, and it reports that this bounded approximation outperforms prior MIL and transformer baselines on all evaluated tasks.

Load-bearing premise

The argument assumes the min/max summary of a region is a true upper bound on how relevant any patch in that region is to the query, so that the top-K regions chosen by that score are nearly the right ones; if the learned projections break that element-wise bound, the approximation guarantee no longer follows.

Editorial extensions

If this is right

Whole-slide transformers can process 100k+ patches with near-linear cost, opening gigapixel-scale self-attention to routine use.
The method yields consistent gains over prior attention-based MIL baselines, such as an average C-index of 0.670 versus 0.647 for the second-best method across eight TCGA survival tasks.
Query-aware selection is a general recipe: any task where relevant context varies per query could adopt the region-summarization-and-top-K attention pattern.
The theoretical bound, if it holds, gives users a worst-case guarantee rather than an empirical heuristic for sparse attention.

Reading between the lines

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

Editorial: The guarantee is only as strong as the element-wise min/max bracketing after learned projections; if the projections break the inequality, the top-K selection can silently drop relevant regions, so a diagnostic check on si versus true max interaction would settle the practical validity.
Editorial: The same selective-attention idea could transfer to other gigapixel-scale imagery, such as satellite or electron-microscopy data, where local relevance is context-dependent.
Editorial: An even cheaper variant would score regions with the raw min/max vectors before projection; comparing that variant's top-K overlap with Querent's would isolate how much the learned projections help or hurt selection.

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

Zhengrui and colleagues have a real idea here: instead of fixed local-global attention for gigapixel WSIs, score each region with learned min/max metadata against the query, and run dense attention only on the top-K regions. That is genuinely new in the MIL/transformer WSI literature they cite, and the min/max scoring mechanism is concrete and ablatable. The experiments are the paper's strongest asset: 11 public datasets, 5-fold CV, standard baselines, ablations against random selection and a side network, and released code. The gains are modest but consistent (average C-Index 0.670 vs 0.647; AUC gains in the low single digits), and the ablation shows the query-aware selection adds value over random region selection, especially on UBC-OCEAN. Nothing about the empirical section feels cooked.

The soft spots are large and load-bearing. The headline claim—near-linear O(NK) complexity approximating full self-attention—is contradicted by the paper's own complexity analysis. Algorithm 1 Phase 2 scores every query against every region: N queries times M=N/K regions, which is O(N^2/K) dot products, and Appendix H.2.2 states O(NRd) with R=N/p. The total in H.3 is O(Nd+NRd+Nkpd)=O(N^2 d/p) for constants. That is quadratic in N, not near-linear. The 'chunking' discussion and Figure 6's near-linear plot do not account for this phase, so the practical-efficiency motivation collapses. The approximation theorem is also unsupported: Lemma B.4 assumes the element-wise min/max bound carries through the learned projections fmin/fmax, but those projections break the element-wise comparison, and the proof mixes original and projected feature spaces. The theorem may be repairable with a different argument, but as written it does not prove what it claims.

These are not minor technicalities; they are the paper's central claims. That said, the empirical contribution could still stand if the authors honestly report the quadratic region-scoring cost and compare against other subquadratic alternatives, or find a genuinely subquadratic importance estimator (e.g., hashing or coarse clustering). As written, the paper overclaims.

Who should read it: anyone working on efficient attention for gigapixel pathology, for the experimental design and the query-adaptive idea—skim the theory, distrust the complexity claims. This deserves a serious referee: the idea is novel and the experiments are extensive enough that the issues could be fixed in revision. I would send it out, but with the expectation that the main claims need either a real proof, a real subquadratic algorithm, or a repositioned paper.

Referee Report

3 major / 6 minor

Summary. The paper proposes Querent, a transformer-style MIL framework for whole-slide images. The method partitions patch tokens into spatial regions, summarizes each region by learned projections of element-wise minimum and maximum feature vectors, scores each query patch against all region summaries, and then computes dense self-attention only over the top-K regions for that query. The authors claim this reduces self-attention complexity from O(N^2) to O(NK), provide a theorem stating that the query-aware attention matrix approximates full self-attention with bounded error, and report state-of-the-art accuracy, AUC, F1, and C-index results across 11 public WSI datasets. The appendix contains pseudo-code, a complexity analysis, and the proof of the approximation theorem.

Significance. If the efficiency and approximation claims were valid, Querent would be a practically useful contribution to computational pathology, combining global context with sub-quadratic cost. The paper's strengths are its broad empirical evaluation across biomarker prediction, mutation prediction, subtyping, and survival analysis, its release of code, and its inclusion of a self-contained appendix with pseudo-code and complexity derivations. However, the central efficiency claim is internally contradicted by the paper's own complexity equations, and the theoretical guarantee is stated for random projections rather than for the trained model actually evaluated. These issues are load-bearing: they undermine the stated motivation for the method, independent of the empirical tables.

major comments (3)

[Section 3.2.3, Appendix H.2.2, H.3, Algorithm 1] The claimed reduction to O(NK) is contradicted by the paper's own complexity analysis. Algorithm 1 Phase 2 loops "for each query patch q" and, inside that loop, "for i = 1 to M", computing a dot product for every query-region pair. Appendix H.2.2 reports this region importance estimation as O(NRd), and Appendix H.3 aggregates the total as O(Nd + NRd + Nkpd). Since R = N/p, this simplifies to O(N^2 d/p), which is quadratic in N for constant p and k, not O(NK) as stated in Section 3.2.3. Consequently, Figure 6's near-linear scaling is not explained by the described algorithm; it either measures only the selective-attention phase or relies on an implementation shortcut that is not described anywhere in the manuscript.
[Appendix B.1.2, Lemma B.4, Eq. (3)] The proof of Lemma B.4 uses the element-wise inequality m_min <= x_j <= m_max for raw features, but the importance score in Eq. (3) is computed after learned projections fmin and fmax. There is no guarantee that fmin(m_min) <= fmin(x_j) <= fmax(m_max) holds element-wise, and the maximum of absolute dot products with projected metadata is not an upper bound on |<q, x_j>| for arbitrary learned query features that may have negative components. The first step of Lemma B.4 therefore fails, and the ranking-stability Lemma B.5 and Theorem B.6 inherit this gap. The Lipschitz constant L of the learned projections is also an assumed free parameter that is never specified or justified for the trained networks.
[Theorem 3.1 / Theorem B.6] The theorem is stated for "random projection matrices WQ, WK in R^{d x d}" and uses Johnson-Lindenstrauss inner-product preservation, but the model in Section 3.2.3 uses a learned Wqkv and learned projections fq, fmin, fmax. The proved object is therefore not the model that is trained and evaluated, so the abstract's claim of "a theoretically bounded approximation of full self-attention" is not supported for Querent as implemented. Additionally, the exponential spatial-decay assumption on attention scores is not established for the learned features, so the theorem's applicability to the actual setting is unverified.

minor comments (6)

[Section 3.2.1 and 3.2.2] The symbol K is used both for the number of patches per region and for the number of selected regions per query, while Appendix H uses p and k for the same concepts; this notational collision makes the complexity claims difficult to verify.
[Introduction] The phrase "10, 0002 ~ 100, 0002 pixels" appears to be a typo for "10,000^2 to 100,000^2 pixels" and should be corrected.
[Tables 1 and 2] Several cells in Tables 1 and 2 contain triplicated values (for example, the Querent row on BCNB-ER and the DTFD row on KIRC), which appears to be a table-generation error that should be fixed.
[Theorem 3.1] The statement of Theorem 3.1 in the main text contains a dangling clause, "The hidden dimension satisfies, and the number of selected regions per query satisfies:", before the numbered conditions; the quantifiers should be stated explicitly.
[Appendix G.2 vs. Section 3.2] Appendix G.2 describes a hybrid loss combining binary classification and ranking components for training the importance estimation module, but Section 3.2 and Algorithm 1 describe only end-to-end training with task losses; this discrepancy should be reconciled.
[Figure 4] The caption reports P<0.005 but does not state which statistical test was used or how many repeated runs were compared; this information should be included in the caption or in Appendix G.1.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: Querent's empirical claims rest on held-out benchmarks and the approximation theorem is an implication, not a restatement of its inputs; the paper's O(NK) complexity claim is internally inconsistent with its own Appendix H, but that is a correctness defect, not circularity.

full rationale

I walked the claimed derivation chain. The learned components (query projection fq, metadata projections fmin/fmax, attention Wqkv, pooling fa, classifier fc) are all trained end-to-end on slide labels, and all reported AUC/ACC/F1/C-index numbers come from 5-fold cross-validation against external baselines; no fitted parameter is renamed as a prediction. The approximation theorem (Theorem 3.1/B.6) is an implication: under stated assumptions (exponential attention decay, JL inner-product preservation via Kaban 2015, region diameter and spatial-separation constraints), it bounds ||A-B||_F. It does not define A in terms of B, nor does it import a uniqueness theorem from the authors' prior work. The region importance score in Eq. 3 is asserted (and Lemma B.4 attempts to prove) to upper-bound true interactions; whether the bound holds for learned projections is an open correctness gap, not a circular step. Self-citations (HistGen, Focus, Mamba-related works) appear as baselines, related work, or foundation models and are not load-bearing for the central derivation. The notable defect is not circularity: Section 3.2.3 claims O(NK) complexity, but Appendix H.2.2 reports region importance estimation as O(NRd) and H.3 totals O(Nd + NRd + Nkpd) with R=N/p, which is O(N^2 d/p) for constant p, contradicting the near-linear claim and Figure 6. That is an internal inconsistency that undermines the practical-efficiency claim, but it is not a reduction of the conclusion to the inputs by definition. Hence no circularity; score 0.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central efficiency and approximation claims rest on tuned constants (region size, number of selected regions) and unverified assumptions about the geometry of PLIP features. The theorem's alpha, L, B, and region-separation lower bounds are assumed but not estimated, and the implementation's top-16 and K are chosen empirically, so the theoretical guarantees do not apply to the evaluated configuration.

free parameters (2)

Region size K = 16 (UBC-OCEAN), 24 (TCGA-LUAD), 16/24/28 depending on dataset
Ablation Fig. 5 shows performance peaks at K=24 for TP53 and K=16 for OCEAN; chosen per dataset and affects accuracy.
Number of selected regions per query (top-K) = 16
Set to 16 for all experiments; claimed empirical sufficiency, not derived from the theorem's epsilon, alpha, or delta.

assumptions (4)

domain assumption Exponential spatial decay of attention scores with rate alpha
Theorem 3.1/B.6 assumes scores decay as exp(-alpha d(i,j)); no empirical check in WSI feature space, and top-K selection is based on learned relevance, not spatial distance.
domain assumption Region diameter and region separation bounds in feature space
Theorem requires diam(R_i) <= min(epsilon/(L sqrt(d)), 1/alpha) and inter-region distance >= C3/alpha; fixed grid regions and PLIP features are not shown to satisfy these.
standard math Johnson-Lindenstrauss inner-product preservation with dimension d rather than a reduced dimension m
Kaban (2015) is cited for a JL guarantee, but the theorem applies d by d random projection matrices, so the dimension-reduction premise does not match a real JL embedding.
ad hoc to paper Lipschitz continuity of fmin and fmax with known constant L
Lemma B.4 assumes L-Lipschitz projections with known L; the single-layer perceptrons with GELU are not verified to satisfy a useful Lipschitz bound.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Context Matters: Query-aware Dynamic Long Sequence Modeling of Gigapixel Images." pith.science (2026). https://pith.science/paper/WHMIXNOW

@misc{pith2026250118984,
  author       = {Pith},
  title        = {Pith review of: Context Matters: Query-aware Dynamic Long Sequence Modeling of Gigapixel Images},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WHMIXNOW}},
  note         = {Machine review of arXiv:2501.18984}
}

read the original abstract

Whole slide image (WSI) analysis presents significant computational challenges due to the massive number of patches in gigapixel images. While transformer architectures excel at modeling long-range correlations through self-attention, their quadratic computational complexity makes them impractical for computational pathology applications. Existing solutions like local-global or linear self-attention reduce computational costs but compromise the strong modeling capabilities of full self-attention. In this work, we propose Querent, i.e., the query-aware long contextual dynamic modeling framework, which achieves a theoretically bounded approximation of full self-attention while delivering practical efficiency. Our method adaptively predicts which surrounding regions are most relevant for each patch, enabling focused yet unrestricted attention computation only with potentially important contexts. By using efficient region-wise metadata computation and importance estimation, our approach dramatically reduces computational overhead while preserving global perception to model fine-grained patch correlations. Through comprehensive experiments on biomarker prediction, gene mutation prediction, cancer subtyping, and survival analysis across over 10 WSI datasets, our method demonstrates superior performance compared to the state-of-the-art approaches. Codes are available at https://github.com/dddavid4real/Querent.

Figures

Figures reproduced from arXiv: 2501.18984 by the authors.

**Figure 1.** Illustration of context-dependent patch relationships in whole slide images. When a benign patch (A) interacts with cancerous patch C, it shows low correlation, while a cancerous patch (B) shows high correlation with patch C. This demonstrates how the same patch (C) can have fundamentally different relationships with other patches depending on the biological context. ligence and deep learning to analyze the growing… view at source ↗

**Figure 2.** Illustration of the proposed Querent framework, which models a WSI via four key steps: (1) region-level metadata summarization from the partitioned WSI, detailed in [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Illustration of the region-level metadata summarization process. Each region from the WSI is represented by summary vectors computed from its constituent patches. These summary vectors capture the statistical characteristics (minimum and maximum values) across all patches within each region, providing an efficient representation for subsequent importance estimation. data bounds: si = max(|⟨q, ˆ mˆ min i ⟩|, |⟨q, ˆ … view at source ↗

Figures from the paper (2 more)

**Figure 4.** Figure 4: Ablation on Querent using min, max, mean, and mean ±std strategies compared to our min-max method on TCGALUAD TP53 gene mutation dataset (details in Appendix G.1). tionships between regions before and after summarization. As shown in [PITH_FULL_IMAGE:figures/full_fig…

**Figure 6.** Figure 6: Computational efficiency comparison between full selfattention and our query-aware approach. (a) Memory requirements in gigabytes and (b) computational complexity in GFLOPs across different sequence lengths. See detailed analysis in Appendix H. cer subtyping (UBC-OCEA…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

FOCUS: Knowledge-enhanced Adaptive Visual Compression for Few-shot Whole Slide Image Classification
cs.CV 2024-11 conditional novelty 6.0 of 10

FOCUS improves few-shot whole slide image classification by progressively removing redundant patches guided by pathology foundation model features and language prompts, then aggregating the surviving tokens with cross...

Reference graph

Works this paper leans on

13 extracted references · 12 canonical work pages · cited by 1 Pith paper

[1]

The hidden dimension satisfies: d ≥ C1 · log(L/δ) ϵ2 where C1 = 8B4 (from JL inner-product preservation (Kaban, 2015))

work page 2015
[2]

The number of selected regions per query satisfies: k ≥ C2 α · log 1 ϵ where C2 = 2 (derived from Step 3)

work page
[3]

For each region Ri, the diameter satisfies: diam(Ri) ≤ min ϵ L · √ d , 1 α

work page
[4]

Proof of Theorem B.6

Regions are spatially separated such that: ∀i ̸= j, d (Ri, Rj) ≥ C3 α where C3 = 1 2 ensures P∞ m=k+1 e−C3m ≤ ϵ. Proof of Theorem B.6. Step 1: Region Metadata Summarization. By Lemma B.4, projections ˆmi min, ˆmi max approximate interactions within Ri with error ϵ1 = B · L · diam(Ri) ≤ Bϵ√ d. The condition diam(Ri) ≤ 1 α ensures region-level interactions ...

work page 2015
[5]

These images showcase the distinct morphological features of squamous cell carcinomas, including keratinization and intercellular bridges

TCGA-LUSC: Lung Squamous Cell Carcinoma dataset from TCGA, providing 454 WSIs and survival data for lung squamous cell carcinoma cases. These images showcase the distinct morphological features of squamous cell carcinomas, including keratinization and intercellular bridges. The dataset enables comparative studies between different types of lung cancers an...

work page
[6]

TCGA-BRCA: Breast Invasive Carcinoma dataset from TCGA, containing 1025 whole slide images with associated survival outcome data for breast cancer patients. This comprehensive collection represents one of the largest breast cancer datasets, featuring diverse histological patterns and molecular subtypes including ductal and lobular carcinomas. The survival...

work page
[7]

This dataset includes various histological grades and stages of endometrial carcinoma, providing valuable insights into the progression and prognosis of gynecologic cancers

TCGA-UCEC: Uterine Corpus Endometrial Carcinoma dataset from TCGA, encompassing 497 WSIs and survival data from endometrial cancer cases. This dataset includes various histological grades and stages of endometrial carcinoma, providing valuable insights into the progression and prognosis of gynecologic cancers. The images showcase diverse morphological pat...

work page
[8]

The dataset represents different anatomical locations within the stomach and various histological subtypes of gastric adenocarcinoma

TCGA-STAD: Stomach Adenocarcinoma dataset from TCGA, which includes 365 WSIs and patient survival infor- mation for gastric cancer cases. The dataset represents different anatomical locations within the stomach and various histological subtypes of gastric adenocarcinoma. The survival data is particularly valuable for understanding the relationship between...

work page

Show all 13 references

[9]

This collection captures the heterogeneous nature of lung adenocarcinomas, including various growth patterns and degrees of differentiation

TCGA-LUAD: Lung Adenocarcinoma dataset from TCGA, containing 457 WSIs and survival outcome data from lung adenocarcinoma patients. This collection captures the heterogeneous nature of lung adenocarcinomas, including various growth patterns and degrees of differentiation. The s...
[11]

This collection includes primary and metastatic melanoma cases, capturing the diverse histological patterns and progression stages of this aggressive skin cancer

TCGA-SKCM: Skin Cutaneous Melanoma dataset from TCGA, consisting of 417 WSIs and survival information from melanoma patients. This collection includes primary and metastatic melanoma cases, capturing the diverse histological patterns and progression stages of this aggressive s...
[12]

The dataset showcases the characteristic clear cell morphology and various grades of renal cell carcinoma

TCGA-KIRC: Kidney Renal Clear Cell Carcinoma dataset from TCGA, containing 500 WSIs with survival outcome data for kidney cancer patients. The dataset showcases the characteristic clear cell morphology and various grades of renal cell carcinoma. The survival information helps ...
[13]

This collection represents a distinct histological subtype of kidney cancer, featuring papillary architecture and different cellular patterns

TCGA-KIRP: Kidney Renal Papillary Cell Carcinoma dataset from TCGA, providing 263 WSIs and survival data for papillary renal cell carcinoma cases. This collection represents a distinct histological subtype of kidney cancer, featuring papillary architecture and different cellul...

2018
[199]

9 Context Matters: Query-aware Dynamic Long Sequence Modeling of Gigapixel Images Guo, Z., Xiong, C., Ma, J., Sun, Q., Feng, L., Wang, J., and Chen, H

Springer, 2024. 9 Context Matters: Query-aware Dynamic Long Sequence Modeling of Gigapixel Images Guo, Z., Xiong, C., Ma, J., Sun, Q., Feng, L., Wang, J., and Chen, H. Focus: Knowledge-enhanced adaptive visual compression for few-shot whole slide image classifica- tion. In Pro...

2024 arXiv

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.

[1] [1]

The hidden dimension satisfies: d ≥ C1 · log(L/δ) ϵ2 where C1 = 8B4 (from JL inner-product preservation (Kaban, 2015))

work page 2015

[2] [2]

The number of selected regions per query satisfies: k ≥ C2 α · log 1 ϵ where C2 = 2 (derived from Step 3)

work page

[3] [3]

For each region Ri, the diameter satisfies: diam(Ri) ≤ min ϵ L · √ d , 1 α

work page

[4] [4]

Proof of Theorem B.6

Regions are spatially separated such that: ∀i ̸= j, d (Ri, Rj) ≥ C3 α where C3 = 1 2 ensures P∞ m=k+1 e−C3m ≤ ϵ. Proof of Theorem B.6. Step 1: Region Metadata Summarization. By Lemma B.4, projections ˆmi min, ˆmi max approximate interactions within Ri with error ϵ1 = B · L · diam(Ri) ≤ Bϵ√ d. The condition diam(Ri) ≤ 1 α ensures region-level interactions ...

work page 2015

[5] [5]

These images showcase the distinct morphological features of squamous cell carcinomas, including keratinization and intercellular bridges

TCGA-LUSC: Lung Squamous Cell Carcinoma dataset from TCGA, providing 454 WSIs and survival data for lung squamous cell carcinoma cases. These images showcase the distinct morphological features of squamous cell carcinomas, including keratinization and intercellular bridges. The dataset enables comparative studies between different types of lung cancers an...

work page

[6] [6]

TCGA-BRCA: Breast Invasive Carcinoma dataset from TCGA, containing 1025 whole slide images with associated survival outcome data for breast cancer patients. This comprehensive collection represents one of the largest breast cancer datasets, featuring diverse histological patterns and molecular subtypes including ductal and lobular carcinomas. The survival...

work page

[7] [7]

This dataset includes various histological grades and stages of endometrial carcinoma, providing valuable insights into the progression and prognosis of gynecologic cancers

TCGA-UCEC: Uterine Corpus Endometrial Carcinoma dataset from TCGA, encompassing 497 WSIs and survival data from endometrial cancer cases. This dataset includes various histological grades and stages of endometrial carcinoma, providing valuable insights into the progression and prognosis of gynecologic cancers. The images showcase diverse morphological pat...

work page

[8] [8]

The dataset represents different anatomical locations within the stomach and various histological subtypes of gastric adenocarcinoma

TCGA-STAD: Stomach Adenocarcinoma dataset from TCGA, which includes 365 WSIs and patient survival infor- mation for gastric cancer cases. The dataset represents different anatomical locations within the stomach and various histological subtypes of gastric adenocarcinoma. The survival data is particularly valuable for understanding the relationship between...

work page