pith. sign in

arxiv: 2604.22858 · v1 · submitted 2026-04-22 · 💻 cs.CV

A Digital Pathology Resource for Liver Cancer Quantification with Datasets, Benchmarks, and Tools

Ying Xiao , Shimiao Tang , Xitong Ling , Weiming Chen , Jun Wang , Jiawen Li , Huaitian Yuan , Jianghui Yang
show 6 more authors
Bowen Li Huan Li Yiting Meng Tian Guan Yonghong He Hongfang Yin
This is my paper

Pith reviewed 2026-05-10 00:10 UTC · model grok-4.3

classification 💻 cs.CV
keywords digital pathologyliver cancerhepatocellular carcinomatissue classificationwhole slide imageregional quantificationHepatoBenchmachine learning
0
0 comments X p. Extension

The pith

The paper releases HepatoBench, a patch-level dataset annotated for seven liver tissue categories, and builds HepatoQuant by combining a tissue classifier with whole-slide segmentation to produce end-to-end regional quantification.

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

The authors note that fine-grained tissue identification on pathology slides is essential for estimating recurrence risk in liver cancer, yet suitable annotated resources remain scarce. They address this by releasing HepatoBench, a collection of annotated image patches covering seven key tissue types, and training a deep-learning classifier on it. They further train a segmentation model that identifies tumor regions across entire whole-slide images and integrate the two into HepatoQuant, which parses tissue composition and generates quantitative statistics directly from new slides.

Core claim

By creating HepatoBench with annotations for seven key tissue categories and training both a patch-level classification model and a WSI-level tumor segmentation model, the authors produce HepatoQuant, an integrated system that delivers a unified workflow from raw whole-slide images to parsed tissue composition and quantitative statistics.

What carries the argument

The integration of the patch-level tissue classifier with the WSI-level segmentation model, which restricts tissue recognition to automatically localized lesion regions and yields slide-wide composition statistics.

If this is right

  • A single pipeline converts raw whole-slide images into tissue composition maps and numerical statistics without manual region selection.
  • Open release of HepatoBench and the benchmarking protocol supports direct, reproducible comparisons among future liver-cancer quantification methods.
  • The resulting quantitative outputs can be linked to clinical endpoints such as recurrence risk to support prognostic modeling.
  • The same modular structure (patch classifier plus slide segmentation) can be retrained on new disease-specific categories while preserving the end-to-end workflow.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the quantitative outputs prove prognostic, the same dataset and workflow could be adapted to other solid tumors by redefining the tissue categories.
  • Linking the automated scores to longitudinal patient records would test whether the seven-category breakdown adds predictive value beyond standard tumor-staging metrics.
  • The benchmark protocol could serve as a template for constructing similar resources in other pathology domains that currently lack fine-grained annotations.

Load-bearing premise

The seven tissue categories capture the clinically relevant variations and the models trained on the annotated patches will perform reliably on new, unseen whole-slide images.

What would settle it

Run the released models on an independent collection of liver cancer whole-slide images that were never used in training or validation; if tissue classification accuracy falls substantially below the reported benchmark levels or if the resulting quantitative scores diverge markedly from expert pathologist review, the claim of practical utility collapses.

Figures

Figures reproduced from arXiv: 2604.22858 by Bowen Li, Hongfang Yin, Huaitian Yuan, Huan Li, Jianghui Yang, Jiawen Li, Jun Wang, Shimiao Tang, Tian Guan, Weiming Chen, Xitong Ling, Ying Xiao, Yiting Meng, Yonghong He.

Figure 1
Figure 1. Figure 1: End-to-end quantitative analysis pipeline for hepatocellular carcinoma whole slide images. 7/8 [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Benchmark results of pathology foundation models on HepatoBench. 8/8 [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
read the original abstract

Liver cancer, especially hepatocellular carcinoma (HCC), imposes a substantial global disease burden. Accurate diagnosis and prognostic assessment directly influence treatment selection and patient survival, and pathological examination remains the gold standard for liver cancer diagnosis. Identifying diverse tissue components and pathological subtypes on histopathology slides is crucial for estimating postoperative recurrence risk and overall prognosis. However, most publicly available resources are still provided at the whole-slide image (WSI) level, and well-annotated datasets for fine-grained tissue component identification in liver cancer are scarce, which hinders reproducible model development and the deployment of quantitative analysis tools. To address this gap, we release HepatoBench, a patch-level image database for liver cancer with annotations for seven key tissue categories. Based on HepatoBench, we train and open-source a deep learning classification model as a tissue recognition tool. Furthermore, we train a WSI-level tumor/non-tumor segmentation model to automatically localize lesion regions across entire slides. By integrating the patch-level tissue classifier with the WSI-level segmentation model, we build HepatoQuant, an end-to-end, disease-specific regional quantification tool for liver cancer, enabling a unified workflow from WSIs to tissue composition parsing and quantitative statistics. We also open-source HepatoBench, the benchmarking protocol, and supporting tools, providing a solid foundation for automated regional quantification and fair method comparison in liver cancer pathology.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper releases HepatoBench, a patch-level dataset for liver cancer histopathology annotated with seven tissue categories. It trains and open-sources a patch-level tissue classifier and a WSI-level tumor/non-tumor segmentation model, then integrates them into HepatoQuant, an end-to-end tool claimed to enable unified regional quantification, tissue composition parsing, and quantitative statistics from whole-slide images. The dataset, benchmarking protocol, and supporting tools are also released to facilitate reproducible research and method comparison.

Significance. If the models achieve reliable accuracy and the dataset proves high-quality and representative, the work could meaningfully advance digital pathology for liver cancer by supplying a scarce fine-grained resource and standardized tools. Open-sourcing the data, models, and protocol is a concrete strength that supports reproducibility and fair benchmarking, potentially aiding prognostic assessment through automated tissue quantification.

major comments (2)
  1. [Abstract] Abstract: the central claim that HepatoQuant provides an 'end-to-end, disease-specific regional quantification tool' enabling 'accurate quantification' is unsupported because no performance metrics, validation protocols, Dice scores, accuracy figures, or error analysis are reported for either the patch-level classifier or the WSI segmentation model.
  2. [Abstract] Abstract (integration description): the assertion that combining the patch-level classifier with WSI segmentation yields reliable tissue-composition statistics on arbitrary new slides rests on untested generalization; no held-out external WSI cohort, multi-center test set, or scanner-variation experiment is described to quantify performance drop.
minor comments (1)
  1. [Abstract] The abstract would be strengthened by including at least one key quantitative result (e.g., classification accuracy or segmentation Dice) to allow readers to gauge the current state of the released models.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thorough review and constructive feedback on our manuscript. We address each of the major comments below and have made revisions to strengthen the abstract and clarify the validation aspects of our work.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that HepatoQuant provides an 'end-to-end, disease-specific regional quantification tool' enabling 'accurate quantification' is unsupported because no performance metrics, validation protocols, Dice scores, accuracy figures, or error analysis are reported for either the patch-level classifier or the WSI segmentation model.

    Authors: We appreciate this observation. While the full paper details the benchmarking results—including accuracy and F1 scores for the patch-level classifier on the HepatoBench test set, as well as Dice coefficients for the WSI-level tumor segmentation model on held-out slides—the abstract did not reference these figures. We have revised the abstract to include a concise summary of the key performance metrics and the internal validation protocol employed, thereby providing support for the quantification claims. revision: yes

  2. Referee: [Abstract] Abstract (integration description): the assertion that combining the patch-level classifier with WSI segmentation yields reliable tissue-composition statistics on arbitrary new slides rests on untested generalization; no held-out external WSI cohort, multi-center test set, or scanner-variation experiment is described to quantify performance drop.

    Authors: We concur that the abstract's phrasing regarding the integration could suggest broader applicability than what was tested. The current implementation and evaluation are based on internal held-out data from the HepatoBench collection. In the revised manuscript, we have updated the abstract to specify that the tissue composition statistics are derived from validation on the dataset's own slides. Additionally, we have included a limitations section discussing the lack of external cohort testing and potential impacts from scanner variations, along with our intentions for future multi-center evaluations. revision: yes

Circularity Check

0 steps flagged

No circularity: direct dataset release and model integration pipeline

full rationale

The paper releases HepatoBench as a new annotated patch dataset for seven liver tissue categories, trains a patch classifier and WSI segmentation model on it, and combines the two outputs into the HepatoQuant tool. No equations, fitted parameters, or predictions appear; the workflow is a standard empirical sequence of data creation followed by supervised training and post-hoc integration. No self-citations serve as load-bearing premises, and no step reduces by construction to its own inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 2 invented entities

This is a dataset and software release paper rather than a theoretical derivation, so the ledger contains no free parameters or axioms; the new entities are the released resources themselves.

invented entities (2)
  • HepatoBench no independent evidence
    purpose: Patch-level annotated database for seven liver cancer tissue categories
    Newly created and released resource introduced to fill the stated data gap.
  • HepatoQuant no independent evidence
    purpose: End-to-end tool combining patch classifier and WSI segmenter for regional quantification
    Newly constructed workflow and software built on the released models.

pith-pipeline@v0.9.0 · 5584 in / 1156 out tokens · 36177 ms · 2026-05-10T00:10:11.566563+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

24 extracted references · 2 canonical work pages

  1. [1]

    journal cancer136, E359–E386 (2015)

    Ferlay, J.et al.Cancer incidence and mortality worldwide: sources, methods and major patterns in globocan 2012.Int. journal cancer136, E359–E386 (2015)

    2012

  2. [2]

    Bray, F.et al.Global cancer statistics 2022: Globocan estimates of incidence and mortality worldwide for 36 cancers in 185 countries.CA: a cancer journal for clinicians74, 229–263 (2024)

    2022

  3. [3]

    & Omata, M

    Kanda, T., Goto, T., Hirotsu, Y ., Moriyama, M. & Omata, M. Molecular mechanisms driving progression of liver cirrhosis towards hepatocellular carcinoma in chronic hepatitis b and c infections: a review.Int. journal molecular sciences20, 1358 (2019)

    2019

  4. [4]

    J., Cholankeril, G

    Kovalic, A. J., Cholankeril, G. & Satapathy, S. K. Nonalcoholic fatty liver disease and alcoholic liver disease: metabolic diseases with systemic manifestations.Transl. Gastroenterol. Hepatol.4, 65 (2019)

    2019

  5. [5]

    Oh, J. H. & Jun, D. W. The latest global burden of liver cancer: A past and present threat.Clin. molecular hepatology29, 355 (2023)

    2023

  6. [6]

    Cureus15(2023)

    Lampimukhi, M.et al.A review of incidence and related risk factors in the development of hepatocellular carcinoma. Cureus15(2023)

    2023

  7. [7]

    A., Groopman, J

    Rayapati, D., McGlynn, K. A., Groopman, J. D. & Kim, A. K. Environmental exposures and the risk of hepatocellular carcinoma.Hepatol. Commun.9, e0627 (2025). 8.Tang, Z.-Y . Hepatocellular carcinoma-cause, treatment and metastasis.World journal gastroenterology7, 445 (2001). 4/8

    2025

  8. [8]

    Edmondson, H. A. & Steiner, P. E. Primary carcinoma of the liver. a study of 100 cases among 48,900 necropsies.Cancer 7, 462–503 (1954). 10.Balogh, J.et al.Hepatocellular carcinoma: a review.J. hepatocellular carcinoma41–53 (2016)

    1954

  9. [9]

    & Rygiel, T

    Sas, Z., Cendrowicz, E., Weinhäuser, I. & Rygiel, T. P. Tumor microenvironment of hepatocellular carcinoma: challenges and opportunities for new treatment options.Int. J. Mol. Sci.23, 3778 (2022)

    2022

  10. [10]

    Kuo, F.-Y .et al.Tumor necrosis is an indicator of poor prognosis among hepatoma patients undergoing resection.J. Surg. Res.283, 1091–1099 (2023)

    2023

  11. [11]

    & Roskams, T

    Libbrecht, L. & Roskams, T. Hepatic progenitor cells in human liver diseases. InSeminars in cell & developmental biology, vol. 13, 389–396 (Elsevier, 2002)

    2002

  12. [12]

    Y ., Brown Jr, R

    Salomao, M., Woojin, M. Y ., Brown Jr, R. S., Emond, J. C. & Lefkowitch, J. H. Steatohepatitic hepatocellular carcinoma (sh-hcc): a distinctive histological variant of hcc in hepatitis c virus-related cirrhosis with associated nafld/nash.The Am. journal surgical pathology34, 1630–1636 (2010)

    2010

  13. [13]

    A.et al.Computational pathology in 2030: a delphi study forecasting the role of ai in pathology within the next decade.EBioMedicine88(2023)

    Berbís, M. A.et al.Computational pathology in 2030: a delphi study forecasting the role of ai in pathology within the next decade.EBioMedicine88(2023). 16.Bankhead, P. Developing image analysis methods for digital pathology.The J. pathology257, 391–402 (2022)

    2030

  14. [14]

    Liang, J.et al.Deep learning supported discovery of biomarkers for clinical prognosis of liver cancer.Nat. Mach. Intell.5, 408–420 (2023)

    2023

  15. [15]

    Data12, 138 (2025)

    Lou, S.et al.A large histological images dataset of gastric cancer with tumour microenvironment annotation for ai.Sci. Data12, 138 (2025)

    2025

  16. [16]

    Data12, 1381 (2025)

    Ling, X.et al.Comprehensive benchmark dataset for pathological lymph node metastasis in breast cancer sections.Sci. Data12, 1381 (2025)

    2025

  17. [17]

    InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 16000–16009 (2022)

    He, K.et al.Masked autoencoders are scalable vision learners. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 16000–16009 (2022)

    2022

  18. [18]

    A.et al.Pancancer insights from the cancer genome atlas: the pathologist’s perspective.The J

    Cooper, L. A.et al.Pancancer insights from the cancer genome atlas: the pathologist’s perspective.The J. pathology244, 512–524 (2018). 22.Kim, Y . J.et al.Paip 2019: Liver cancer segmentation challenge.Med. image analysis67, 101854 (2021)

    2018

  19. [19]

    E.et al.Diagnostic assessment of deep learning algorithms for detection of lymph node metastases in women with breast cancer.Jama318, 2199–2210 (2017)

    Bejnordi, B. E.et al.Diagnostic assessment of deep learning algorithms for detection of lymph node metastases in women with breast cancer.Jama318, 2199–2210 (2017)

    2017

  20. [20]

    doi: 10.1109/TMI.2018.2867350

    Bándi, P.et al.From detection of individual metastases to classification of lymph node status at the patient level: The camelyon17 challenge.IEEE Transactions on Med. Imaging38, 550–560, 10.1109/TMI.2018.2867350 (2019). 25.Weinstein, J. N.et al.The cancer genome atlas pan-cancer analysis project.Nat. genetics45, 1113–1120 (2013)

    work page doi:10.1109/tmi.2018.2867350 2018

  21. [21]

    pathology260, 564–577 (2023)

    Asif, A.et al.Unleashing the potential of ai for pathology: challenges and recommendations.The J. pathology260, 564–577 (2023)

    2023

  22. [22]

    J.et al.Towards a general-purpose foundation model for computational pathology.Nat

    Chen, R. J.et al.Towards a general-purpose foundation model for computational pathology.Nat. Medicine30, 850–862 (2024)

    2024

  23. [23]

    medicine30, 2924–2935 (2024)

    V orontsov, E.et al.A foundation model for clinical-grade computational pathology and rare cancers detection.Nat. medicine30, 2924–2935 (2024). 29.Xu, H.et al.A whole-slide foundation model for digital pathology from real-world data.Nature1–8 (2024)

    2024

  24. [24]

    In2025 47th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), 1–7 (IEEE, 2025)

    Bolhassani, M.et al.Lora-fine-tuned large vision models for automated assessment of post-sbrt lung injury. In2025 47th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), 1–7 (IEEE, 2025). 31.Xiao, Y .et al.Hepatobench (revision d78c1d3), 10.57967/hf/8231 (2026). Acknowledgements This work was supported by the B...

    work page doi:10.57967/hf/8231 2025