Pith. sign in

REVIEW 3 major objections 7 minor 46 references

A High Magnifications Histopathology Image Dataset for Oral Squamous Cell Carcinoma Diagnosis and Prognosis

T0 review · 3 major / 7 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 This paper introduces Multi-OSCC, a publicly released histopathology image dataset of 1,325 oral squamous cell carcinoma patients annotated for six diagnostic and prognostic tasks, and reports benchmark results that reach 94.72% AUC for…

desk verdict A genuinely useful new OSCC dataset, but its benchmark numbers need a revision pass before they are citable. read the letter →

arxiv 2507.16360 v1 pith:CLBXCLCR submitted 2025-07-22 eess.IV cs.CV

Jinquan Guan , Junhong Guo , Qi Chen , Jian Chen , Yongkang Cai , Yilin He , Zhiquan Huang , Yan Wang
show 1 more author
Yutong Xie
This is my paper
classification eess.IVcs.CV
keywords oralsquamouscellcarcinomahistopathologyimagedatasetdiagnosisandprognosisrecurrencepredictionmulti-tasklearningstainnormalizationdeepbenchmarklymphnodemetastasis
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 authors aim to close a gap in oral squamous cell carcinoma (OSCC) research: existing public histopathology datasets are small or cover only diagnosis or only prognosis. They introduce Multi-OSCC, which links both diagnostic and prognostic labels for 1,325 patients, each represented by six high-magnification images from the tumor core and edge. They argue this is the first publicly available OSCC histopathology dataset designed for multiple diagnostic and prognostic targets, and they back this with systematic benchmarks. The benchmark findings include strong recurrence prediction (94.72% AUC), a clear benefit from pathology-specific pretraining, a task-dependent effect of stain normalization, and a measurable difficulty in combining all six tasks in one multi-task model.

What carries the argument

The central object is the Multi-OSCC dataset itself: six images per patient (two per magnification level, one from the lesion core and one from the edge) with patient-level clinical labels drawn from electronic medical records for six tasks. The accompanying benchmark pipeline combines a vision backbone (with ImageNet or pathology-specific pretrained weights), a feature fusion module (concatenation, low-rank multimodal fusion, gated fusion, or Transformer), and task-specific classification heads, evaluated by AUC with bootstrap confidence intervals. This setup lets the authors isolate the contribution of pretraining data, multi-image fusion, stain normalization, and multi-task optimization, and it provides the baseline numbers that future models on this dataset are meant to beat.

What would settle it

Have two independent pathologists re-annotate a random sample of about 100 patients from the test split for all six labels and measure inter-observer agreement; then recompute the reported AUCs on the consensus re-read labels instead of the original record-derived labels. If agreement is low or the AUCs shift substantially, the benchmark numbers are contaminated by label noise.

Watch

Extended reading notes

Core claim

Multi-OSCC is, to the authors' knowledge, the first publicly available histopathology image dataset specifically designed for OSCC research with multiple diagnostic and prognostic targets. It contains 1,325 patients, each with six H&E-stained images captured at ×200, ×400, and ×1000 magnifications from both the core and edge of the lesion, annotated for recurrence (REC), lymph node metastasis (LNM), tumor differentiation (TD), tumor invasion (TI), cancer embolus (CE), and perineural invasion (PI). The authors demonstrate that when the best model is selected per task, all six tasks exceed 70% AUC, with recurrence reaching 94.72% and tumor differentiation 81.23%. They further report that pathology-specific pretrained encoders outperform ImageNet-pretrained encoders on average, that stain normalization improves five diagnostic tasks but degrades recurrence prediction, and that multi-task learning with hard parameter sharing suffers an average 3.34% AUC degradation compared with single-task models.

Load-bearing premise

The clinical labels for all six tasks are taken directly from hospital electronic medical records without an independent pathology re-review or inter-observer validation, so any label errors, especially for rare events like cancer embolus and for recurrences after the two-year follow-up window, would directly lower every benchmark number.

Editorial extensions

If this is right

  • Future OSCC histopathology models should initialize from pathology-specific pretrained encoders rather than ImageNet weights, since the average AUC across the six tasks is consistently higher with such pretraining.
  • Stain normalization is not uniformly beneficial: it should be applied selectively, because it improves diagnostic tasks while hurting recurrence prediction, suggesting recurrence models rely on color information that normalization removes.
  • Using images from both the core and the edge of the lesion, at three magnifications, improves average AUC over using either region alone, supporting multi-site, multi-resolution image collection.
  • Hard-parameter-sharing multi-task learning, even with GradNorm balancing, still loses about 3.34% average AUC relative to single-task models, so joint diagnosis-and-prognosis models will need more advanced task-balancing or task-grouping strategies.
  • The benchmark, with six tasks and per-task best results above 70% AUC, provides a reproducible reference point for evaluating future OSCC computer-aided diagnosis and prognosis systems.

Reading between the lines

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

  • Editorial inference: the resolution ablation shows that fine-tuning at 512×512 produces a much larger gain than simply using full-resolution images with a frozen encoder, which suggests that fine-tuning at the native 2592×1944 resolution could push performance higher; the paper does not test this because of GPU memory limits.
  • Editorial inference: because recurrence prediction is the most color-sensitive task and the dataset captures only six representative fields rather than whole slides, a whole-slide version of Multi-OSCC might reveal whether the strong recurrence AUC is driven by tissue architecture or by stain-related artifacts.
  • Editorial inference: the weak correlation between REC and the diagnostic tasks hints that recurrence depends on factors beyond tumor severity; pairing this image dataset with treatment and follow-up variables could let future models separate image-based risk from clinical risk.
  • Editorial inference: the 3.34% multi-task degradation, concentrated in REC, LNM, TD, and TI, suggests that task conflict is not uniform; grouping tasks by shared morphological patterns (e.g., invasion-related tasks) may recover some of the lost performance.
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

3 major / 7 minor

Summary. The paper introduces Multi-OSCC, a histopathology image dataset of 1,325 OSCC patients, with six high-resolution microscope images per patient (two each at ×200, ×400, and ×1000 from core and edge tumor regions) and patient-level labels for six tasks: recurrence (REC), lymph node metastasis (LNM), tumor differentiation (TD), tumor invasion (TI), cancer embolus (CE), and perineural invasion (PI). The authors benchmark visual encoders (ImageNet vs. pathology-specific pretraining), feature fusion methods, stain normalization, and multi-task learning. They report that pathology-specific pretraining helps, that stain normalization hurts REC but helps diagnostic tasks, and that multi-task learning degrades average AUC by 3.34%. The dataset and code are released publicly.

Significance. If the resource is as described, it is a valuable community contribution: it is among the largest OSCC-specific histopathology image collections with both diagnostic and prognostic labels, covering multiple clinically relevant tasks and magnifications. Strengths include the public release, fixed-seed training, bootstrap confidence intervals for headline metrics, and a systematic comparison of backbones, fusion, stain normalization, and multi-task learning methods. The central benchmark claims, however, need to be interpreted carefully because the final configuration was selected on the test set and because the underlying labels come from electronic medical records without independent verification; both issues affect the reliability of the reported AUCs as estimates of generalization.

major comments (3)
  1. [Section 5.1, Tables 5, 6, 8, 10] The benchmark configuration is selected using test AUC: Table 5 selects PathoBench ViT-Small for the highest mean test AUC, Table 6 selects Concat fusion, and Table 8's per-task choice of no stain normalization for REC but Reinhard for other tasks is made from test-set results. The same test set is then used to report the final single-task benchmark in Table 10 and the multi-task benchmark in Table 11. Because model and preprocessing choices are tuned on the test set, the reported AUCs are optimistically biased, and the benchmark is not a fair basis for comparing future methods. Please fix the analysis configuration before touching the test set (for example, using a separate selection/validation set or nested cross-validation), or explicitly characterize the selection bias, for instance by reporting the performance of the final pipeline on a truly held-out test set.
  2. [Abstract and Section 6 vs. Table 10] The abstract states that 'all tasks surpass 70% AUC' and Section 6 states that 'the optimal AUC for other tasks exceeded 70%.' Under the final single-task benchmark in Table 10, however, PI attains an AUC of 66.19 (95% CI 56.52–75.42), and under the multi-task benchmark in Table 11 PI is 67.41. The statement is true only if one selects the best backbone per task from Table 5, where PI peaks at 74.17 with CONCH, but that is not the configuration reported as the dataset benchmark. Please correct the abstract and Discussion to reflect the final benchmark numbers, or state explicitly that the >70% claim refers to per-task best backbones rather than the unified benchmark.
  3. [Section 3.1] The patient-level annotations for all six tasks are obtained from the hospital's electronic medical records, with no independent pathology re-review, central adjudication, or inter-observer agreement reported. The paper also notes that nerve fibers and blood vessels may not be present in every captured image, so CE and PI negative labels partly reflect sampling of fields rather than confirmed absence of invasion. No label-noise estimate or sensitivity analysis is provided. Because every benchmark AUC is computed against these labels, unmeasured label error, especially for rare positive classes such as CE (66/925 training positives), can materially bias the reported numbers. Please report the label extraction protocol in more detail, provide any quality-control steps, and, if feasible, quantify label noise on a re-reviewed subset or perform a sensitivity analysis under plausible mislabeling rates.
minor comments (7)
  1. [Section 4.2.2] In the Gated Fusion description, the output is written as Z = ∑_{i=0}^5 (e_i · t_i), but the sigmoid output a_i is not used; please clarify whether the intended combination is a_i ⊙ t_i or another gating operation.
  2. [Section 4.1.1] For the validation and test sets, the reference image for Reinhard, Vahadane, and Macenko normalization is not specified; please state how the stain normalization reference is chosen at inference time, as this is needed for reproducibility.
  3. [Section 4.2.1] Typo: 'PahtoBench' should be 'PathoBench'.
  4. [Section 3.2] Please state explicitly that the train/validation/test split is performed at the patient level (one patient's six images stay in one split), since Table 4 lists 925/200/200 patients and each patient contributes six images.
  5. [Section 5.1] Please provide the exact random seed(s) and the full set of hyperparameters (for example, number of warmup epochs and weight decay schedule) to support the reproducibility claim; the current text says 'fixed random seeds' but does not name them.
  6. [Section 5.9] The statement that the benchmark uses 'a consistent setup across all tasks' is not literally true because REC is exempted from stain normalization; please rephrase to 'a shared backbone and fusion with task-specific stain normalization' or justify the exemption.
  7. [Section 6] The phrase 'electronic microscopy' should be 'optical microscopy' (or 'light microscopy') to match the Olympus microscope described in Section 3.1.

Circularity Check

1 steps flagged · score 2.0 of 10

Benchmark configuration is selected using the same test set later used to report the benchmark AUCs, mildly inflating the headline numbers; no self-citation or definitional circularity is present.

  1. fitted input called prediction [Section 5.5, Section 5.9, Table 8 and Table 10]
    "Therefore, in the single-task benchmark 5, we present the results without stain normalization for the REC task, while for the other tasks, we present the results with Reinhard stain normalization (Reinhard et al., 2001)."

    The per-task choice of stain normalization is made by comparing test-set AUCs in Table 8, and the same test set is then used to produce the 'Single-task Benchmark' in Table 10. The reported benchmark values are therefore not independent held-out estimates: they are a re-reporting of the configuration selected to maximize those very test AUCs. The same selection loop occurs for the backbone (Section 5.2), fusion method (Section 5.3), and multi-task optimizer (Section 5.6), all decided via test-set AUC and then codified in Section 5.9. This makes the headline numbers, such as the 94.72% REC AUC, selected maxima rather than predictions validated on unseen data.

full rationale

This is a dataset-and-benchmark paper rather than a derivation paper, so most circularity patterns do not apply. The six task labels come from the hospital's electronic medical records (Section 3.1), and the images are acquired by microscope; neither is defined in terms of the model outputs, so there is no self-definitional circularity. The benchmark models are trained on a fixed training split and evaluated on a separate test split, so the test AUCs are genuine measurements. No load-bearing self-citations were found: the cited prior datasets and pretraining methods are external works, and no uniqueness theorem or ansatz is imported from the authors' own prior publications. The mild circularity is in the benchmark protocol: the choice of backbone, fusion block, stain normalization, and multi-task optimizer is made by inspecting test-set AUCs in Sections 5.2-5.6, and the final 'benchmark' in Section 5.9 then reports AUCs on that same test set. This is a form of test-set selection that statistically forces the reported numbers upward, but it does not reduce the paper's central dataset contribution to its inputs. A separate correctness concern, not circularity, is that all ground-truth labels are taken from EMR without independent pathology re-review or a label-noise estimate, as stated in Section 3.1; this affects every benchmark number but is an annotation-validity issue rather than a circular derivation.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The central claim does not rest on a mathematical derivation, so no fitted numerical constants are used beyond the data-dependent benchmark configuration. The main unproved premises are the accuracy and completeness of the clinical labels, the sufficiency of a two-year follow-up for recurrence, the representativeness of the manually selected core and edge microscopy regions, and the transferability of pretrained feature extractors to this cohort. The benchmark configuration itself was selected using test-set AUC, which is a data-dependent design choice.

free parameters (1)
  • Benchmark configuration choices selected via test AUC = PathoBench ViT-Small + Concat + Reinhard for diagnostic tasks, original for REC; GradNorm for multi-task
    The final benchmark pipeline was chosen by maximizing test AUC across the explored alternatives (Sections 5.2-5.6). These choices are data-dependent and affect the reported scores.
assumptions (4)
  • domain assumption Clinical labels obtained from electronic medical records are accurate and complete
    Section 3.1: 'Patient-level annotations, including diagnosis and prognosis, are obtained from the hospital's electronic medical records.' No independent pathology re-review or inter-observer agreement is reported.
  • domain assumption A 2-year follow-up period is sufficient to define recurrence status
    Section 3.1 inclusion criteria (3): 'participation in follow-up care for at least two years post-surgery.' Recurrences after 2 years may be mislabeled as non-recurrence.
  • domain assumption The manually selected core and edge regions at three magnifications capture sufficient information for the six tasks
    Section 3.1: pathologists selected sites and preserved pictures; no guarantee nerve fibers or blood vessels are present in every image. The usefulness of this sampling for CE and PI is uncertain.
  • domain assumption Pre-trained models transfer to this dataset
    The entire method relies on transfer learning from ImageNet or pathology-specific pretraining (Section 4.2.1); this is a standard but unproved assumption for this specific cohort.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A High Magnifications Histopathology Image Dataset for Oral Squamous Cell Carcinoma Diagnosis and Prognosis." pith.science (2026). https://pith.science/paper/CLBXCLCR

@misc{pith2026250716360,
  author       = {Pith},
  title        = {Pith review of: A High Magnifications Histopathology Image Dataset for Oral Squamous Cell Carcinoma Diagnosis and Prognosis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CLBXCLCR}},
  note         = {Machine review of arXiv:2507.16360}
}
read the original abstract

Oral Squamous Cell Carcinoma (OSCC) is a prevalent and aggressive malignancy where deep learning-based computer-aided diagnosis and prognosis can enhance clinical assessments.However, existing publicly available OSCC datasets often suffer from limited patient cohorts and a restricted focus on either diagnostic or prognostic tasks, limiting the development of comprehensive and generalizable models. To bridge this gap, we introduce Multi-OSCC, a new histopathology image dataset comprising 1,325 OSCC patients, integrating both diagnostic and prognostic information to expand existing public resources. Each patient is represented by six high resolution histopathology images captured at x200, x400, and x1000 magnifications-two per magnification-covering both the core and edge tumor regions.The Multi-OSCC dataset is richly annotated for six critical clinical tasks: recurrence prediction (REC), lymph node metastasis (LNM), tumor differentiation (TD), tumor invasion (TI), cancer embolus (CE), and perineural invasion (PI). To benchmark this dataset, we systematically evaluate the impact of different visual encoders, multi-image fusion techniques, stain normalization, and multi-task learning frameworks. Our analysis yields several key insights: (1) The top-performing models achieve excellent results, with an Area Under the Curve (AUC) of 94.72% for REC and 81.23% for TD, while all tasks surpass 70% AUC; (2) Stain normalization benefits diagnostic tasks but negatively affects recurrence prediction; (3) Multi-task learning incurs a 3.34% average AUC degradation compared to single-task models in our multi-task benchmark, underscoring the challenge of balancing multiple tasks in our dataset. To accelerate future research, we publicly release the Multi-OSCC dataset and baseline models at https://github.com/guanjinquan/OSCC-PathologyImageDataset.

Figures

Figures reproduced from arXiv: 2507.16360 by the authors.

Figure 1
Figure 1. (a) illustrates an abstract scene depicting the collection process of core/edge histopathology slides. Six images in (b,c) are from the same patient. The three in (b) are captured from tissue sections at the lesion core, with magnifications of ×200, ×400, and ×1000 from left to right, while the images in (c) are from the lesion boundary. In (b), the focus is on keratin pearl details and surrounding cells, while (c) … view at source ↗
Figure 2
Figure 2. Overview of the proposed pipeline, which includes stain normalization, pre-trained backbone analysis, feature fusion, and multi-task optimization. The pipeline processes six input images from a patient, outputting the probability for a single task in single-task mode or for all tasks in multi-task mode. 4. Method 4.1. Data Preprocessing 4.1.1. Stain Normalization In examining the effect of stain inconsistency on dia… view at source ↗
Figure 3
Figure 3. Visualization of model attention for each of the six clinical tasks, generated by the top-performing model for that respective task. In each subfigure, the original histopathology image (top) is paired with a heatmap (bottom) indicating the model’s focus areas. (a) High tumor differentiation, with keratin pearls highlighted as clinical evidence of high differen￾tiation. (b) Tumor invasion into surrounding tissues, s… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Performance comparison of Freezed (2592x1944), Freezed (512x512) and Tuned (512x512) models [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

46 extracted references · 36 canonical work pages

  1. [1]

    , author Mohammed, K.K

    author Afify, H.M. , author Mohammed, K.K. , author Hassanien, A.E. , year 2023 . title Novel prediction model on oscc histopathological images via deep transfer learning combined with grad-cam interpretation . journal Biomedical Signal Processing and Control volume 83 , pages 104704

    work page 2023

  2. [2]

    , author Thakur, A

    author Albalawi, E. , author Thakur, A. , author Ramakrishna, M.T. , author Bhatia Khan, S. , author SankaraNarayanan, S. , author Almarri, B. , author Hadi, T.H. , year 2024 . title Oral squamous cell carcinoma detection using efficientnet on histopathological images . journal Frontiers in Medicine volume 10 , pages 1349336

    work page 2024

  3. [3]

    , author Kiros, J.R

    author Ba, J.L. , author Kiros, J.R. , author Hinton, G.E. , year 2016 . title Layer normalization . journal arXiv preprint arXiv:1607.06450

    arXiv 2016

  4. [4]

    , author Laversanne, M

    author Bray, F. , author Laversanne, M. , author Sung, H. , author Ferlay, J. , author Siegel, R.L. , author Soerjomataram, I. , author Jemal, A. , year 2024 . title Global cancer statistics 2022: Globocan estimates of incidence and mortality worldwide for 36 cancers in 185 countries . journal CA: a cancer journal for clinicians volume 74 , pages 229--263

    work page 2024

  5. [5]

    , author Touvron, H

    author Caron, M. , author Touvron, H. , author Misra, I. , author J \'e gou, H. , author Mairal, J. , author Bojanowski, P. , author Joulin, A. , year 2021 . title Emerging properties in self-supervised vision transformers , in: booktitle Proceedings of the IEEE/CVF international conference on computer vision , pp. pages 9650--9660

    work page 2021

  6. [6]

    , author Sarkar, A

    author Chattopadhay, A. , author Sarkar, A. , author Howlader, P. , author Balasubramanian, V.N. , year 2018 . title Grad-cam++: Generalized gradient-based visual explanations for deep convolutional networks , in: booktitle 2018 IEEE winter conference on applications of computer vision (WACV) , organization IEEE . pp. pages 839--847

    work page 2018

  7. [7]

    , author Rai, A

    author Chaudhary, N. , author Rai, A. , author Rao, A.M. , author Faizan, M.I. , author Augustine, J. , author Chaurasia, A. , author Mishra, D. , author Chandra, A. , author Chauhan, V. , author Ahmad, T. , year 2024 . title High-resolution ai image dataset for diagnosing oral submucous fibrosis and squamous cell carcinoma . journal Scientific Data volum...

    work page 2024

  8. [8]

    , author Bowyer, K.W

    author Chawla, N.V. , author Bowyer, K.W. , author Hall, L.O. , author Kegelmeyer, W.P. , year 2002 . title Smote: synthetic minority over-sampling technique . journal Journal of artificial intelligence research volume 16 , pages 321--357

    work page 2002

Show all 46 references
  1. [9]

    , author Chen, C

    author Chen, R.J. , author Chen, C. , author Li, Y. , author Chen, T.Y. , author Trister, A.D. , author Krishnan, R.G. , author Mahmood, F. , year 2022 . title Scaling vision transformers to gigapixel images via hierarchical self-supervised learning , in: booktitle Proceedings...

    2022

  2. [10]

    , author Badrinarayanan, V

    author Chen, Z. , author Badrinarayanan, V. , author Lee, C.Y. , author Rabinovich, A. , year 2018 . title Gradnorm: Gradient normalization for adaptive loss balancing in deep multitask networks , in: booktitle International conference on machine learning , organization PMLR ....

    2018

  3. [11]

    , author Assam, P.N

    author Claridge-Chang, A. , author Assam, P.N. , year 2016 . title Estimation statistics should replace significance testing . journal Nature methods volume 13 , pages 108--109

    2016

  4. [12]

    , author Wang, X

    author Corredor, G. , author Wang, X. , author Zhou, Y. , author Lu, C. , author Fu, P. , author Syrigos, K. , author Rimm, D.L. , author Yang, M. , author Romero, E. , author Schalper, K.A. , et al., year 2019 . title Spatial architecture and arrangement of tumor-infiltrating...

    2019

  5. [13]

    , year 2020

    author Crawshaw, M. , year 2020 . title Multi-task learning with deep neural networks: A survey . journal arXiv preprint arXiv:2009.09796

    2020 arXiv

  6. [14]

    what is likely to happen

    author Croft, P. , author Altman, D.G. , author Deeks, J.J. , author Dunn, K.M. , author Hay, A.D. , author Hemingway, H. , author LeResche, L. , author Peat, G. , author Perel, P. , author Petersen, S.E. , et al., year 2015 . title The science of clinical practice: disease di...

    2015

  7. [15]

    , author Dong, W

    author Deng, J. , author Dong, W. , author Socher, R. , author Li, L.J. , author Li, K. , author Fei-Fei, L. , year 2009 . title Imagenet: A large-scale hierarchical image database , in: booktitle 2009 IEEE conference on computer vision and pattern recognition , organization I...

    2009

  8. [16]

    , author Efron, B

    author DiCiccio, T.J. , author Efron, B. , year 1996 . title Bootstrap confidence intervals . journal Statistical science volume 11 , pages 189--228

    1996

  9. [17]

    , author Beyer, L

    author Dosovitskiy, A. , author Beyer, L. , author Kolesnikov, A. , author Weissenborn, D. , author Zhai, X. , author Unterthiner, T. , author Dehghani, M. , author Minderer, M. , author Heigold, G. , author Gelly, S. , et al., year 2020 . title An image is worth 16x16 words: ...

    2020 arXiv

  10. [18]

    , author Chen, Y

    author Fu, Q. , author Chen, Y. , author Li, Z. , author Jing, Q. , author Hu, C. , author Liu, H. , author Bao, J. , author Hong, Y. , author Shi, T. , author Li, K. , et al., year 2020 . title A deep learning algorithm for detection of oral cavity squamous cell carcinoma fro...

    2020

  11. [19]

    , author Bordes, A

    author Glorot, X. , author Bordes, A. , author Bengio, Y. , year 2011 . title Deep sparse rectifier neural networks , in: booktitle Proceedings of the fourteenth international conference on artificial intelligence and statistics , organization JMLR Workshop and Conference Proc...

    2011

  12. [20]

    , author Zhang, X

    author He, K. , author Zhang, X. , author Ren, S. , author Sun, J. , year 2016 . title Deep residual learning for image recognition , in: booktitle Proceedings of the IEEE conference on computer vision and pattern recognition , pp. pages 770--778

    2016

  13. [21]

    , author Liu, Z

    author Huang, G. , author Liu, Z. , author Van Der Maaten, L. , author Weinberger, K.Q. , year 2017 . title Densely connected convolutional networks , in: booktitle Proceedings of the IEEE conference on computer vision and pattern recognition , pp. pages 4700--4708

    2017

  14. [22]

    , author Song, H

    author Kang, M. , author Song, H. , author Park, S. , author Yoo, D. , author Pereira, S. , year 2023 . title Benchmarking self-supervised learning on diverse pathology datasets , in: booktitle Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition ,...

    2023

  15. [23]

    , author Schultz, H

    author Kayser, K. , author Schultz, H. , author Goldmann, T. , author G \"o rtler, J. , author Kayser, G. , author Vollmer, E. , year 2009 . title Theory of sampling and its application in tissue based diagnosis . journal Diagnostic Pathology volume 4 , pages 1--13

    2009

  16. [24]

    , author Hu, H

    author Liu, Z. , author Hu, H. , author Lin, Y. , author Yao, Z. , author Xie, Z. , author Wei, Y. , author Ning, J. , author Cao, Y. , author Zhang, Z. , author Dong, L. , et al., year 2022 . title Swin transformer v2: Scaling up capacity and resolution , in: booktitle Procee...

    2022

  17. [25]

    , author Shen, Y

    author Liu, Z. , author Shen, Y. , author Lakshminarasimhan, V.B. , author Liang, P.P. , author Zadeh, A. , author Morency, L.P. , year 2018 . title Efficient low-rank multimodal fusion with modality-specific factors . journal arXiv preprint arXiv:1806.00064

    2018 arXiv

  18. [26]

    , author Chen, B

    author Lu, M.Y. , author Chen, B. , author Williamson, D.F. , author Chen, R.J. , author Liang, I. , author Ding, T. , author Jaume, G. , author Odintsov, I. , author Le, L.P. , author Gerber, G. , et al., year 2024 . title A visual-language foundation model for computational ...

    2024

  19. [27]

    , author Williamson, D.F

    author Lu, M.Y. , author Williamson, D.F. , author Chen, T.Y. , author Chen, R.J. , author Barbieri, M. , author Mahmood, F. , year 2021 . title Data-efficient and weakly supervised computational pathology on whole-slide images . journal Nature biomedical engineering volume 5 ...

    2021

  20. [28]

    , author Niethammer, M

    author Macenko, M. , author Niethammer, M. , author Marron, J.S. , author Borland, D. , author Woosley, J.T. , author Guan, X. , author Schmitt, C. , author Thomas, N.E. , year 2009 . title A method for normalizing histology slides for quantitative analysis , in: booktitle 200...

    2009

  21. [29]

    , author Sieniek, M

    author McKinney, S.M. , author Sieniek, M. , author Godbole, V. , author Godwin, J. , author Antropova, N. , author Ashrafian, H. , author Back, T. , author Chesus, M. , author Corrado, G.S. , author Darzi, A. , et al., year 2020 . title International evaluation of an ai syste...

    2020

  22. [30]

    , author Pchelnikov, A

    author Nechaev, D. , author Pchelnikov, A. , author Ivanova, E. , year 2024 . title Hibou: A family of foundational vision transformers for pathology . journal arXiv preprint arXiv:2406.05074

    2024 arXiv

  23. [31]

    , author Darcet, T

    author Oquab, M. , author Darcet, T. , author Moutakanni, T. , author Vo, H. , author Szafraniec, M. , author Khalidov, V. , author Fernandez, P. , author Haziza, D. , author Massa, F. , author El-Nouby, A. , et al., year 2023 . title Dinov2: Learning robust visual features wi...

    2023 arXiv

  24. [32]

    , author Mahanta, L.B

    author Rahman, T.Y. , author Mahanta, L.B. , author Das, A.K. , author Sarma, J.D. , year 2020 . title Histopathological imaging database for oral cancer analysis . journal Data in brief volume 29 , pages 105114

    2020

  25. [33]

    , author Adhikhmin, M

    author Reinhard, E. , author Adhikhmin, M. , author Gooch, B. , author Shirley, P. , year 2001 . title Color transfer between images . journal IEEE Computer graphics and applications volume 21 , pages 34--41

    2001

  26. [34]

    , author Qi, M

    author Ren, J. , author Qi, M. , author Yuan, Y. , author Duan, S. , author Tao, X. , year 2020 . title Machine learning--based mri texture analysis to predict the histologic grade of oral squamous cell carcinoma . journal American Journal of Roentgenology volume 215 , pages 1...

    2020

  27. [35]

    , author Koltun, V

    author Sener, O. , author Koltun, V. , year 2018 . title Multi-task learning as multi-objective optimization . journal Advances in neural information processing systems volume 31

    2018

  28. [36]

    , year 1961

    author Spearman, C. , year 1961 . title The proof and measurement of association between two things

    1961

  29. [37]

    , author Hinton, G

    author Srivastava, N. , author Hinton, G. , author Krizhevsky, A. , author Sutskever, I. , author Salakhutdinov, R. , year 2014 . title Dropout: a simple way to prevent neural networks from overfitting . journal The journal of machine learning research volume 15 , pages 1929--1958

    2014

  30. [38]

    , author Swain-Bowden, M.J

    author Stirling, D.R. , author Swain-Bowden, M.J. , author Lucas, A.M. , author Carpenter, A.E. , author Cimini, B.A. , author Goodman, A. , year 2021 . title Cellprofiler 4: improvements in speed, utility and usability . journal BMC bioinformatics volume 22 , pages 1--11

    2021

  31. [39]

    , author Peng, T

    author Vahadane, A. , author Peng, T. , author Sethi, A. , author Albarqouni, S. , author Wang, L. , author Baust, M. , author Steiger, K. , author Schlitter, A.M. , author Esposito, I. , author Navab, N. , year 2016 . title Structure-preserving color normalization and sparse ...

    2016

  32. [40]

    , author Shazeer, N

    author Vaswani, A. , author Shazeer, N. , author Parmar, N. , author Uszkoreit, J. , author Jones, L. , author Gomez, A.N. , author Kaiser, . , author Polosukhin, I. , year 2017 . title Attention is all you need . journal Advances in neural information processing systems volume 30

    2017

  33. [41]

    , author Hartmann, S

    author Vollmer, A. , author Hartmann, S. , author Vollmer, M. , author Shavlokhova, V. , author Brands, R.C. , author K \"u bler, A. , author Wollborn, J. , author Hassel, F. , author Couillard-Despres, S. , author Lang, G. , et al., year 2024 . title Multimodal artificial int...

    2024

  34. [42]

    , author Suebnukarn, S

    author Warin, K. , author Suebnukarn, S. , year 2024 . title Deep learning in oral cancer-a systematic review . journal BMC Oral Health volume 24 , pages 212

    2024

  35. [43]

    , author Sun, Q

    author Yan, R. , author Sun, Q. , author Jin, C. , author Liu, Y. , author He, Y. , author Guan, T. , author Chen, H. , year 2024 . title Shapley values-enabled progressive pseudo bag augmentation for whole-slide image classification . journal IEEE Transactions on Medical Imaging

    2024

  36. [44]

    , author Wei, C

    author Zhou, J. , author Wei, C. , author Wang, H. , author Shen, W. , author Xie, C. , author Yuille, A. , author Kong, T. , year 2021 . title ibot: Image bert pre-training with online tokenizer . journal arXiv preprint arXiv:2111.07832

    2021 arXiv

  37. [45]

    , author Wu, H

    author Zhou, J. , author Wu, H. , author Hong, X. , author Huang, Y. , author Jia, B. , author Lu, J. , author Cheng, B. , author Xu, M. , author Yang, M. , author Wu, T. , year 2024 . title A pathology-based diagnosis and prognosis intelligent system for oral squamous cell ca...

    2024

  38. [46]

    , author Jarosz, R

    author Zuley, M. , author Jarosz, R. , author Kirk, S. , author Lee, Y. , author Colen, R. , author Garcia, K. , author Delbeke, D. , author Pham, M. , author Nagy, P. , author Sevinc, G. , et al., year 2016 . title The cancer genome atlas head-neck squamous cell carcinoma col...

    2016

Pith tools

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