Pith. sign in

REVIEW 4 major objections 5 minor 46 references

DiagR1: A Vision-Language Model Trained via Reinforcement Learning for Digestive Pathology Diagnosis

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read A reinforcement-learning-trained vision-language model can write gastrointestinal pathology reports that outperform GPT-4o, by grounding generation in a pathology classifier's predictions and an explicit reasoning chain.

desk verdict A real dataset and a clear pipeline, but the clinical-relevance claims outrun the evidence; send to peer review with major revision. read the letter →

arxiv 2507.18433 v1 pith:4IUS4LYP submitted 2025-07-24 eess.IV cs.CV

Minxi Ouyang , Lianghui Zhu , Yaqing Bao , Qiang Huang , Jingli Ouyang , Tian Guan , Xitong Ling , Jiawen Li
show 5 more authors
Song Duan Wenbin Dai Li Zheng Xuemei Zhang Yonghong He
This is my paper · ORCID
classification eess.IVcs.CV
keywords gastrointestinalpathologyvision-languagemodelreportgenerationreinforcementlearningGRPOpromptaugmentationchain-of-thoughthallucinationreduction
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

DiagR1 is a vision-language model for gastrointestinal pathology that the authors train to generate both the microscopic-findings section and the pathological diagnosis of a report. The paper's central claim is that grounding the model with predictions from a pathology-specific image classifier plus biopsy-site information, forcing an explicit reasoning trace, and then optimizing structure, caption semantics, and diagnosis wording through reinforcement learning reduces hallucination and produces reports that beat strong open and proprietary baselines. On their internal and external test sets, the authors report BERTScore and BLEU gains over GPT-4o, and the abstract summarizes these as 18.7% higher clinical relevance, 32.4% improved structural completeness, and 41.2% fewer diagnostic errors. A sympathetic reader should care because the pipeline is a concrete route from noisy, partially annotated hospital reports to a reviewable automated draft.

What carries the argument

The load-bearing mechanism is prompt argumentation plus a three-segment output scaffold optimized by GRPO. A frozen UNIv2 image encoder and two small linear heads supply per-image lesion-class probabilities; that classification output, together with the biopsy site, is inserted into the prompt so the language model knows what region and lesion type it is looking at. The output scaffold <think>...<caption>...<answer>... forces an auditable reasoning chain, and the GRPO reward is a composite of a strict 0/1 format check, a BERT-embedding cosine similarity for the caption, and BLEU for the diagnosis. The mechanism converts a general image-to-text model into a structured report writer by making the classifier's prior and the report template part of the optimization signal.

What would settle it

Have board-certified pathologists, blinded to model identity, score 200 DiagR1 and GPT-4o reports for clinically significant diagnostic errors and unsupported histologic claims. If DiagR1 does not show a lower error rate and higher acceptance, the central claim of clinical utility fails even though its BLEU and BERTScore are higher.

Watch

Extended reading notes

Core claim

The discovery the paper argues for is that a relatively small open-weight multimodal model, Qwen2.5-VL-7B, can be turned into a digestive-pathology report generator by a post-training recipe: supervised fine-tuning on 11,904 expert reports with microscopic findings, then GRPO reinforcement learning on 12,627 additional reports, with prompts enriched by an ROI classifier built on a frozen UNIv2 encoder and by the anatomical site. The model must emit its reasoning in <think>...</think>, the microscopic findings in <caption>...</caption>, and the diagnosis in <answer>...</answer>, and the reward is a weighted sum of a format check, BERT-cosine similarity of the caption, and BLEU of the diagnosis. On the Liuzhou test set, DiagR1 reaches 88.55 BERTScore and 12.95 BLEU for microscopic findings and 92.80 BERTScore and 20.93 BLEU for diagnoses, against 82.66 and 1.12 and 85.07 and 2.69 for GPT-4o on the same two tasks; on the external Chongqing set, the diagnosis gains are smaller but still positive. The paper takes these results to show that prompt argumentation plus RL post-training yields reports that are more structured, more semantically aligned with reference reports, and less prone to unsupported histologic claims than general-purpose vision-language models.

Load-bearing premise

The evaluation hinges on BLEU and BERTScore computed against the original pathology reports being faithful proxies for clinical relevance and diagnostic correctness; if those text-similarity scores do not reflect clinical quality, the claimed superiority over GPT-4o does not establish clinical utility.

Editorial extensions

If this is right

  • If the reported gains hold, the same supervised-fine-tuning-plus-GRPO recipe can be applied to other structured medical reports whenever a decent organ-level classifier and site metadata are available.
  • Prompt argumentation can be a cheaper alternative to larger models: a high-accuracy classifier constraint appears to reduce VLM hallucination without retraining the visual encoder.
  • The <think> chain produces outputs a pathologist can audit, which is a prerequisite for real clinical workflow use rather than just a text-generation benchmark.
  • The larger gains on the in-distribution Liuzhou set and smaller gains on the external Chongqing set suggest the method is sensitive to hospital-specific report style and data completeness.

Reading between the lines

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

  • Because the rewards are n-gram and embedding similarity, the RL stage may be rewarding style and lexical mimicry of the original reports rather than diagnostic truth; a pathologist-adjudicated study of clinically significant errors is the real test.
  • If the ROI classifier's prior is wrong, the prompt could anchor the model toward an incorrect lesion class, so the method should be stress-tested on cases where the classifier is confident but wrong.
  • The headline 'clinical relevance' and 'fewer diagnostic errors' numbers are derived from text-similarity metrics; reading them as clinician-verified would be an extrapolation beyond the paper.
  • The same three-part scaffold and reward mix could transfer to endoscopy or radiology structured reporting, where templates and site labels are equally standardized.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper describes DiagR1, a vision-language model for gastrointestinal pathology report generation. The authors construct a large dataset of pathology reports with microscopic findings and diagnoses, augment prompts with predictions from a UNIv2-based ROI classifier plus anatomical site information, and then train Qwen2.5-VL-7B through supervised fine-tuning followed by GRPO reinforcement learning. The RL reward is a weighted combination of a format reward, a caption reward based on BERT embedding cosine similarity (Eq. 5), and a diagnosis reward based on BLEU (Eq. 6). Experiments compare DiagR1 with GPT-4o, Gemini-2.5-pro, Llama-3.2-11B-Vision, DeepSeek-VL2, Qwen2.5-VL-7B, and DiagSFT on internal (LZ) and external (CQ) test sets using BERTScore and BLEU. The authors report consistent improvements and also provide ablation studies for prompt augmentation and chain-of-thought prompting. The abstract claims that DiagR1 achieves 18.7% higher clinical relevance, 32.4% improved structural completeness, and 41.2% fewer diagnostic errors compared to state-of-the-art models.

Significance. If the clinical utility claims were validated, the paper would address an important need in pathology report generation, and the combination of prompt augmentation with RL post-training is a sensible engineering direction. The internal/external split and the ablation studies are useful contributions, and the paper is clearly written. However, the evaluation is the central weakness: the two metrics used in Table I (BERTScore and BLEU) are exactly the caption and diagnosis rewards optimized in Eqs. (5) and (6). Consequently, the reported improvements partly reflect reward overfitting rather than clinical quality, and the abstract's headline numbers are not derivable from any reported table. The paper provides no clinician-based validation, no diagnostic-accuracy evaluation, and no statistical significance testing. Since the paper's stated contribution is clinical utility, this is a load-bearing gap. The paper also does not release code or data, which limits reproducibility, but that alone would not preclude publication if the evaluation were sound.

major comments (4)
  1. [Section IV.A and Eqs. (5)-(7)] The evaluation protocol uses the same quantities that the model was trained to maximize. The caption reward R_cap in Eq. (5) is the cosine similarity of BERT embeddings, and the diagnosis reward R_ans in Eq. (6) is BLEU; Table I then reports BERTScore and BLEU on the test sets. Because GRPO optimizes these exact objectives on training samples, the observed gains in Table I could be a direct consequence of reward optimization rather than of improved clinical reasoning or diagnostic correctness. The claim that DiagR1 'outperforms' baselines in clinical relevance and diagnostic accuracy is therefore not established by the current evidence. An evaluation with clinician-scored outputs, diagnostic error rates verified by pathologists, or metrics distinct from the rewards (e.g., factual consistency against structured diagnostic categories) is required.
  2. [Abstract and Section IV.C] The abstract's headline numbers (18.7% higher clinical relevance, 32.4% improved structural completeness, 41.2% fewer diagnostic errors) do not appear anywhere in the manuscript, and their mapping to Table I is unclear. For example, on diagnosis BERTScore for LZ, DiagR1 improves over GPT-4o by 7.73 percentage points (from 85.07 to 92.80), which is a relative gain of about 9.1%, not 18.7%; BLEU increases from 2.69 to 20.93, which is far more than 18.7% in relative terms. The terms 'clinical relevance' and 'structural completeness' are not defined operationally. The authors should either remove these numbers or state precisely how they were computed, ideally with the corresponding table rows.
  3. [Table I] No error bars, confidence intervals, or significance tests are reported for the benchmark numbers. Given the low absolute BLEU values (e.g., 4.20 on the CQ diagnosis task), small differences in raw BLEU can translate into large relative differences, and the reported improvements over GPT-4o (e.g., 1.81 percentage points in BLEU on CQ) may not be statistically significant. The claim of 'significantly outperforms' in Section IV.C requires either repeated runs with variance estimates or an appropriate significance test across the test set.
  4. [Section IV.A (dataset split)] The internal LZ split is at the level of reports, not patients, and all LZ training, test, and GRPO samples come from the same hospital. If a single patient contributed multiple biopsy reports across the SFT, GRPO, and test sets, the test results would be inflated by patient-level leakage. The authors should state whether the split was performed at the patient level and, if not, re-run the evaluation with a patient-exclusive split. This is relevant because pathology reports are highly templated and reports from the same patient are likely near-duplicates.
minor comments (5)
  1. [Throughout] The text consistently uses 'prompt argumentation' where 'prompt augmentation' appears to be intended. Please correct this in the title of Section III.B and elsewhere to avoid confusion with formal argumentation theory.
  2. [Fig. 4] In the figure caption, 'DigeR1 Generation' should be 'DiagR1 Generation'.
  3. [References] Reference [5] contains an incomplete citation: 'Official journal of the American College of Gastroenterology— ACG , pp. 10–14309, 2022' appears to be truncated. Please provide the full article title and volume/page details.
  4. [Table I and Section IV.C] Section IV.C states that DiagR1 outperforms GPT-4o by 11.83 percentage points in BLEU on the microscopic findings task of LZ, but Table I shows a difference of 12.95 - 1.12 = 11.83, which is consistent; however, the sentence says '5.90 percentage points' for BERTScore while Table I gives 88.55 - 82.66 = 5.89, a minor rounding inconsistency. Please align these values.
  5. [Section IV.E.1] The relative improvement of GPT-4o BERTScore from 79.23 to 85.07 is described as approximately 7.40%, but the actual relative gain is (85.07 - 79.23)/79.23 = 7.37%; the source of the 7.40% figure is unclear. Please check the arithmetic.

Circularity Check

1 steps flagged · score 6.0 of 10

RL rewards and evaluation metrics are the same BLEU/BERT-similarity quantities, so the reported clinical-relevance gains partly reduce to optimizing the evaluation metric.

  1. fitted input called prediction [Section III.C.3 (Eqs. 5-7) and Section IV.A (evaluation protocol); Table I]
    "To ensure the generated diagnosis is accurate and linguistically coherent, we adopt the BLEU score as the reward for the <answer> segment. ... This similarity score serves as a soft reward signal that encourages semantic alignment between generated and reference findings. ... For evaluation, we adopt a dual-perspective protocol: ... we first encode model outputs using a pre-trained BERT model and compute cosine similarity to assess semantic alignment."

    The caption reward in Eq. (5) is cosine similarity of BERT embeddings and the diagnosis reward in Eq. (6) is BLEU; both are combined into the final reward in Eq. (7). Section IV.A then evaluates the model with exactly the same two quantities: BERT cosine similarity and BLEU-1..4. Table I is therefore a re-measurement of the objective the model was trained to maximize, not an independent test of clinical relevance or diagnostic correctness. The abstract's claims of 'clinical relevance' and 'fewer diagnostic errors' are not given separate clinical definitions or clinician scoring, so the reported gains can reflect reward optimization on templated reports rather than independent clinical utility.

full rationale

The central derivation chain is otherwise self-contained: the model is initialized from Qwen2.5-VL, fine-tuned with SFT and GRPO, and compared against external baselines, with no self-citation chain or imported uniqueness theorem carrying the argument. However, the evaluation is not independent of the training objective. The diagnosis reward is literally BLEU and the caption reward is literally BERT embedding cosine similarity, and the evaluation section reports the same two metrics. This means the headline superiority of DiagR1 over GPT-4o on Table I is partly a statement about how well the model optimized its own reward function on similarly templated reports, not an independently validated measure of clinical relevance or diagnostic error reduction. The abstract's quantitative claims (18.7%, 32.4%, 41.2%) are not derived from any table or defined metric in the text, and no clinician-based validation is provided. This is a partial circularity: the reported 'clinical relevance' and 'diagnostic errors' are not established beyond the metrics used as rewards, though the engineering contribution remains plausible.

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

The paper is an empirical systems paper; it introduces no new theoretical entities or physical laws. Its central claim rests on the assumptions that the curated dataset is accurate and representative, that BLEU/BERTScore capture clinical quality, that the RL rewards do not induce reward hacking, and that the prompt-augmentation classifier generalizes to the test sets.

free parameters (1)
  • Reward weighting coefficients lambda_format, lambda_cap, lambda_ans = 1/3 each
    Section IV.B sets all three reward weights to 1/3 by hand; no tuning, sensitivity analysis, or justification is reported, and the reward decomposition directly shapes the final model.
assumptions (4)
  • domain assumption The curated gastrointestinal pathology reports and ROI labels are accurate and representative of clinical GI pathology practice.
    All training and evaluation relies on expert-annotated reports and ROI classifications; errors or biases in these labels would propagate into the model and the reported metrics.
  • domain assumption BLEU and BERTScore against reference reports are valid proxies for clinical relevance and diagnostic correctness.
    Section IV.A defines the evaluation protocol with these metrics; if they do not reflect clinical quality, the reported improvements do not establish clinical utility.
  • domain assumption GRPO optimization with the composite reward does not lead to reward hacking or template overfitting that inflates metrics without clinical benefit.
    The RL objective uses BLEU and BERT similarity as rewards; without analysis of output diversity or clinician review, the model may exploit surface-level patterns.
  • domain assumption The ROI classifier's predictions used in prompt augmentation are correct on the test distribution.
    Prompts embed classifier outputs; incorrect or overconfident classifications could mislead generation, and the paper does not analyze classifier accuracy on the report test images.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DiagR1: A Vision-Language Model Trained via Reinforcement Learning for Digestive Pathology Diagnosis." pith.science (2026). https://pith.science/paper/4IUS4LYP

@misc{pith2026250718433,
  author       = {Pith},
  title        = {Pith review of: DiagR1: A Vision-Language Model Trained via Reinforcement Learning for Digestive Pathology Diagnosis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4IUS4LYP}},
  note         = {Machine review of arXiv:2507.18433}
}
read the original abstract

Multimodal large models have shown great potential in automating pathology image analysis. However, current multimodal models for gastrointestinal pathology are constrained by both data quality and reasoning transparency: pervasive noise and incomplete annotations in public datasets predispose vision language models to factual hallucinations when generating diagnostic text, while the absence of explicit intermediate reasoning chains renders the outputs difficult to audit and thus less trustworthy in clinical practice. To address these issues, we construct a large scale gastrointestinal pathology dataset containing both microscopic descriptions and diagnostic conclusions, and propose a prompt argumentation strategy that incorporates lesion classification and anatomical site information. This design guides the model to better capture image specific features and maintain semantic consistency in generation. Furthermore, we employ a post training pipeline that combines supervised fine tuning with Group Relative Policy Optimization (GRPO) to improve reasoning quality and output structure. Experimental results on real world pathology report generation tasks demonstrate that our approach significantly outperforms state of the art open source and proprietary baselines in terms of generation quality, structural completeness, and clinical relevance. Our solution outperforms state of the art models with 18.7% higher clinical relevance, 32.4% improved structural completeness, and 41.2% fewer diagnostic errors, demonstrating superior accuracy and clinical utility compared to existing solutions.

Figures

Figures reproduced from arXiv: 2507.18433 by the authors.

Figure 1
Figure 1. Challenges of general VLLMs in pathological image analysis. (a) Performance comparison between VLLMs and the pathology foundation model in digestive classification tasks. (b) Impact of classification prompts from the foundation model on pathology report generation. and 4-class intestinal classification tasks, pathology￾specific visual models exhibit significantly higher clas￾sification accuracy compared to general-p… view at source ↗
Figure 2
Figure 2. Pipeline for constructing a digestive pathology report dataset. The process involves two main stages: structured extraction of key information from raw pathology reports and the construction of an ROI classification dataset to train pathology classifiers. The trained classifiers are then used to predict on image-text samples, and the classification results are incorporated into the original text for prompt augmentat… view at source ↗
Figure 3
Figure 3. Overview of the proposed DiagR1 training and inference framework. In the training phase, enhanced image-text data are used for SFT and reinforcement learning based on GRPO, where task-specific reward functions are designed to optimize the quality of generated microscopic findings and diagnoses for digestive pathology. This training pipeline enables DiagR1 to generate pathology reports efficiently and reliably during… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Comparison of generation results between DiagR1 and GPT-4o on two test datasets. After applying prompt argumentation to the reports from both datasets, the inputs were processed by the models, resulting in the GPT-4o response and DiagR1 generation. The results clearly …

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 28 canonical work pages

  1. [1]

    Computational pathology: A comprehensive review of recent developments in digital and intelligent pathology,

    Q. Huang, S. Wu, Z. Ou, and Y . Gao, “Computational pathology: A comprehensive review of recent developments in digital and intelligent pathology,” Intelligent Oncology, 2025

    work page 2025

  2. [2]

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

    H. Xu, N. Usuyama, J. Bagga, S. Zhang, R. Rao, T. Naumann, C. Wong, Z. Gero, J. Gonz ´alez, Y . Gu, et al. , “A whole-slide foundation model for digital pathology from real-world data,” Nature, vol. 630, no. 8015, pp. 181–188, 2024

    work page 2024

  3. [3]

    Burden and cost of gastrointestinal, liver, and pancreatic diseases in the united states: Update 2024,

    A. F. Peery, C. C. Murphy, C. Anderson, E. T. Jensen, S. Deutsch-Link, M. D. Egberg, J. L. Lund, D. Subramaniam, E. S. Dellon, A. D. Sperber, et al., “Burden and cost of gastrointestinal, liver, and pancreatic diseases in the united states: Update 2024,” Gastroenterology, vol. 168, no. 5, pp. 1000–1024, 2025

    work page 2024

  4. [4]

    Patterns of upper gas- trointestinal diseases among patients undergoing esophagogastroduo- denoscopy at three hospitals in asella town, southeast ethiopia,

    D. Zena, F. Hisa, Z. Hurrisa, and M. Kaso, “Patterns of upper gas- trointestinal diseases among patients undergoing esophagogastroduo- denoscopy at three hospitals in asella town, southeast ethiopia,”Scientific Reports, vol. 14, no. 1, p. 24067, 2024

    work page 2024

  5. [5]

    Quality indicators for upper gi endoscopy,

    R. Yadlapati, D. Early, P. G. Iyer, D. R. Morgan, N. Sengupta, P. Sharma, and N. J. Shaheen, “Quality indicators for upper gi endoscopy,” Official journal of the American College of Gastroenterology— ACG , pp. 10– 14309, 2022

    work page 2022

  6. [6]

    The current troubled state of the global pathology workforce: a concise review,

    E. Walsh and N. M. Orsi, “The current troubled state of the global pathology workforce: a concise review,” Diagnostic Pathology, vol. 19, no. 1, p. 163, 2024

    work page 2024

  7. [7]

    Enhancing whole slide image classification with discriminative and contrastive learning,

    P. Liang, H. Zheng, H. Li, Y . Gong, S. Bakas, and Y . Fan, “Enhancing whole slide image classification with discriminative and contrastive learning,” in International Conference on Medical Image Computing and Computer-Assisted Intervention , pp. 102–112, Springer, 2024

    work page 2024

  8. [8]

    Application of deep learning convolutional neural networks to identify gastric squamous cell carcinoma in mice,

    Y . Ren, S. Li, D. Zhang, Y . Zhao, Y . Yang, G. Huo, X. Zhou, X. Geng, Z. Lin, and Z. Qu, “Application of deep learning convolutional neural networks to identify gastric squamous cell carcinoma in mice,” Frontiers in Medicine, vol. 12, p. 1587417, 2025

    work page 2025

Show all 46 references
  1. [9]

    Qwen2. 5-vl technical report,

    S. Bai, K. Chen, X. Liu, J. Wang, W. Ge, S. Song, K. Dang, P. Wang, S. Wang, J. Tang, et al., “Qwen2. 5-vl technical report,” arXiv preprint arXiv:2502.13923, 2025

    2025 arXiv

  2. [10]

    Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning,

    D. Guo, D. Yang, H. Zhang, J. Song, R. Zhang, R. Xu, Q. Zhu, S. Ma, P. Wang, X. Bi, et al., “Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning,” arXiv preprint arXiv:2501.12948 , 2025

    2025 arXiv

  3. [11]

    Medpath: Augmenting health risk prediction via medical knowledge paths,

    M. Ye, S. Cui, Y . Wang, J. Luo, C. Xiao, and F. Ma, “Medpath: Augmenting health risk prediction via medical knowledge paths,” in Proceedings of the Web Conference 2021 , pp. 1397–1409, 2021

    2021

  4. [12]

    Generating dermatopathology reports from gigapixel whole slide im- ages with histogpt,

    M. Tran, P. Schmidle, R. R. Guo, S. J. Wagner, V . Koch, V . Lupperger, B. Novotny, D. H. Murphree, H. D. Hardway, M. D’Amato, et al. , “Generating dermatopathology reports from gigapixel whole slide im- ages with histogpt,” Nature Communications, vol. 16, no. 1, pp. 1–17, 2025

    2025

  5. [13]

    Computational pathology, new horizons and challenges for anatomical pathology,

    W. L. dos Santos, L. A. de Freitas, A. A. Duarte, M. F. Angelo, and L. R. Oliveira, “Computational pathology, new horizons and challenges for anatomical pathology,” Surgical and Experimental Pathology, vol. 5, no. 1, p. 10, 2022

    2022

  6. [14]

    Closing the loop–the role of pathologists in digital and computational pathology research,

    T. T. Rau, W. Cross, R. R. Lastra, R. C.-L. Lo, A. Matoso, and C. S. Herrington, “Closing the loop–the role of pathologists in digital and computational pathology research,” 2024

    2024

  7. [15]

    Synoptic reporting by summarizing cancer pathology reports using large language models,

    S. Rajaganapathy, S. Chowdhury, X. Li, V . Buchner, Z. He, R. Zhang, X. Jiang, P. Yang, J. R. Cerhan, and N. Zong, “Synoptic reporting by summarizing cancer pathology reports using large language models,” npj Health Systems, vol. 2, no. 1, p. 11, 2025

    2025

  8. [16]

    A framework to assess clinical safety and hallucination rates of llms for medical text summarisation,

    E. Asgari, N. Montana-Brown, M. Dubois, S. Khalil, J. Balloch, and D. Pimenta, “A framework to assess clinical safety and hallucination rates of llms for medical text summarisation,” medRxiv, pp. 2024–09, 2024

    2024

  9. [17]

    Benchmark pathology report text corpus with cancer type classification,

    J. Kefeli and N. Tatonetti, “Benchmark pathology report text corpus with cancer type classification,” medRxiv, 2023

    2023

  10. [18]

    Impact of template-based synoptic reporting on completeness of surgical pathology reports,

    N. Schaad, S. Berezowska, A. Perren, and E. Hewer, “Impact of template-based synoptic reporting on completeness of surgical pathology reports,” Virchows Archiv, vol. 484, no. 1, pp. 31–36, 2024

    2024

  11. [19]

    The 1 million words pathology report or the challenge of a reproducible and meaningful message,

    C. Eloy, P. Seegers, E. Bazyleva, and F. Fraggetta, “The 1 million words pathology report or the challenge of a reproducible and meaningful message,” ESMO Real World Data and Digital Oncology , vol. 4, p. 100044, 2024. 10 IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS

    2024

  12. [20]

    Multimodal whole slide foundation model for pathology,

    T. Ding, S. J. Wagner, A. H. Song, R. J. Chen, M. Y . Lu, A. Zhang, A. J. Vaidya, G. Jaume, M. Shaban, A. Kim, et al., “Multimodal whole slide foundation model for pathology,”arXiv preprint arXiv:2411.19666, 2024

    2024 arXiv

  13. [21]

    A foundation model for clinical-grade computational pathology and rare cancers detection,

    E. V orontsov, A. Bozkurt, A. Casson, G. Shaikovski, M. Zelechowski, K. Severson, E. Zimmermann, J. Hall, N. Tenenholtz, N. Fusi, et al., “A foundation model for clinical-grade computational pathology and rare cancers detection,” Nature medicine , vol. 30, no. 10, pp. 2924–2935, 2024

    2024

  14. [22]

    Towards a general-purpose foundation model for computational pathology,

    R. J. Chen, T. Ding, M. Y . Lu, D. F. Williamson, G. Jaume, A. H. Song, B. Chen, A. Zhang, D. Shao, M. Shaban, et al. , “Towards a general-purpose foundation model for computational pathology,” Nature Medicine, vol. 30, no. 3, pp. 850–862, 2024

    2024

  15. [23]

    Transformer-based unsupervised contrastive learning for histopathological image classification,

    X. Wang, S. Yang, J. Zhang, M. Wang, J. Zhang, W. Yang, J. Huang, and X. Han, “Transformer-based unsupervised contrastive learning for histopathological image classification,” Medical image analysis, vol. 81, p. 102559, 2022

    2022

  16. [24]

    A visual-language foundation model for computational pathology,

    M. Y . Lu, B. Chen, D. F. Williamson, R. J. Chen, I. Liang, T. Ding, G. Jaume, I. Odintsov, L. P. Le, G. Gerber, et al. , “A visual-language foundation model for computational pathology,” Nature Medicine , vol. 30, no. 3, pp. 863–874, 2024

    2024

  17. [25]

    A visual–language foundation model for pathology image analysis using medical twitter,

    Z. Huang, F. Bianchi, M. Yuksekgonul, T. J. Montine, and J. Zou, “A visual–language foundation model for pathology image analysis using medical twitter,” Nature medicine, vol. 29, no. 9, pp. 2307–2316, 2023

    2023

  18. [26]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark, et al., “Learning transferable visual models from natural language supervision,” in International conference on machine learning , pp. 8748–8763, PmLR, 2021

    2021

  19. [27]

    Visual instruction tuning,

    H. Liu, C. Li, Q. Wu, and Y . J. Lee, “Visual instruction tuning,” Advances in neural information processing systems , vol. 36, pp. 34892– 34916, 2023

    2023

  20. [28]

    Improved baselines with visual instruction tuning,

    H. Liu, C. Li, Y . Li, and Y . J. Lee, “Improved baselines with visual instruction tuning,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pp. 26296–26306, 2024

    2024

  21. [29]

    Pathasst: A generative foundation ai assistant towards artificial general intelligence of pathology,

    Y . Sun, C. Zhu, S. Zheng, K. Zhang, L. Sun, Z. Shui, Y . Zhang, H. Li, and L. Yang, “Pathasst: A generative foundation ai assistant towards artificial general intelligence of pathology,” in Proceedings of the AAAI Conference on Artificial Intelligence , vol. 38, pp. 5034–5042, 2024

    2024

  22. [30]

    Towards a generalizable pathology foundation model via unified knowledge distillation,

    J. Ma, Z. Guo, F. Zhou, Y . Wang, Y . Xu, J. Li, F. Yan, Y . Cai, Z. Zhu, C. Jin, et al., “Towards a generalizable pathology foundation model via unified knowledge distillation,” arXiv preprint arXiv:2407.18449, 2024

    2024 arXiv

  23. [31]

    A multimodal knowledge-enhanced whole- slide pathology foundation model,

    Y . Xu, Y . Wang, F. Zhou, J. Ma, C. Jin, S. Yang, J. Li, Z. Zhang, C. Zhao, H. Zhou, et al. , “A multimodal knowledge-enhanced whole- slide pathology foundation model,” arXiv preprint arXiv:2407.15362 , 2024

    2024 arXiv

  24. [32]

    Biogpt: generative pre-trained transformer for biomedical text genera- tion and mining,

    R. Luo, L. Sun, Y . Xia, T. Qin, S. Zhang, H. Poon, and T.-Y . Liu, “Biogpt: generative pre-trained transformer for biomedical text genera- tion and mining,” Briefings in bioinformatics, vol. 23, no. 6, p. bbac409, 2022

    2022

  25. [33]

    A multimodal generative ai copilot for human pathology,

    M. Y . Lu, B. Chen, D. F. Williamson, R. J. Chen, M. Zhao, A. K. Chow, K. Ikemura, A. Kim, D. Pouli, A. Patel, et al. , “A multimodal generative ai copilot for human pathology,” Nature, vol. 634, no. 8033, pp. 466–473, 2024

    2024

  26. [34]

    4v (ision) system card https://cdn. openai. com/papers,

    G. OpenAi, “4v (ision) system card https://cdn. openai. com/papers,” GPTV System Card. pdf, 2023

    2023

  27. [35]

    Quilt-llava: Visual instruction tuning by extracting localized narratives from open-source histopathology videos,

    M. S. Seyfioglu, W. O. Ikezogwo, F. Ghezloo, R. Krishna, and L. Shapiro, “Quilt-llava: Visual instruction tuning by extracting localized narratives from open-source histopathology videos,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp...

    2024

  28. [36]

    Histgen: Histopathology report generation via local-global feature encoding and cross-modal context interaction,

    Z. Guo, J. Ma, Y . Xu, Y . Wang, L. Wang, and H. Chen, “Histgen: Histopathology report generation via local-global feature encoding and cross-modal context interaction,” in International Conference on Medi- cal Image Computing and Computer-Assisted Intervention, pp. 189–199, S...

    2024

  29. [37]

    Vlm-r1: A stable and generalizable r1-style large vision-language model,

    H. Shen, P. Liu, J. Li, C. Fang, Y . Ma, J. Liao, Q. Shen, Z. Zhang, K. Zhao, Q. Zhang, et al., “Vlm-r1: A stable and generalizable r1-style large vision-language model,” arXiv preprint arXiv:2504.07615 , 2025

    2025 arXiv

  30. [38]

    Vision-r1: Incentivizing reasoning capability in multimodal large language models,

    W. Huang, B. Jia, Z. Zhai, S. Cao, Z. Ye, F. Zhao, Z. Xu, Y . Hu, and S. Lin, “Vision-r1: Incentivizing reasoning capability in multimodal large language models,” arXiv preprint arXiv:2503.06749 , 2025

    2025 arXiv

  31. [39]

    Openai o1 system card,

    A. Jaech, A. Kalai, A. Lerer, A. Richardson, A. El-Kishky, A. Low, A. Helyar, A. Madry, A. Beutel, A. Carney, et al. , “Openai o1 system card,” arXiv preprint arXiv:2412.16720 , 2024

    2024 arXiv

  32. [40]

    Llamav- o1: Rethinking step-by-step visual reasoning in llms,

    O. Thawakar, D. Dissanayake, K. More, R. Thawkar, A. Heakl, N. Ah- san, Y . Li, M. Zumri, J. Lahoud, R. M. Anwer, et al. , “Llamav- o1: Rethinking step-by-step visual reasoning in llms,” arXiv preprint arXiv:2501.06186, 2025

    2025 arXiv

  33. [41]

    Visual-rft: Visual reinforcement fine-tuning,

    Z. Liu, Z. Sun, Y . Zang, X. Dong, Y . Cao, H. Duan, D. Lin, and J. Wang, “Visual-rft: Visual reinforcement fine-tuning,” arXiv preprint arXiv:2503.01785, 2025

    2025 arXiv

  34. [42]

    Histogym: A reinforce- ment learning environment for histopathological image analysis,

    Z.-B. Liu, X. Pang, J. Wang, S. Liu, and C. Li, “Histogym: A reinforce- ment learning environment for histopathological image analysis,” arXiv preprint arXiv:2408.08847, 2024

    2024 arXiv

  35. [43]

    Dual attention model with reinforcement learning for classification of histology whole-slide images,

    M. Raza, R. Awan, R. M. S. Bashir, T. Qaiser, and N. M. Rajpoot, “Dual attention model with reinforcement learning for classification of histology whole-slide images,” Computerized Medical Imaging and Graphics, vol. 118, p. 102466, 2024

    2024

  36. [44]

    Med-r1: Reinforcement learning for generalizable medical reasoning in vision-language models,

    Y . Lai, J. Zhong, M. Li, S. Zhao, and X. Yang, “Med-r1: Reinforcement learning for generalizable medical reasoning in vision-language models,” arXiv preprint arXiv:2503.13939 , 2025

    2025

  37. [45]

    Prox- imal policy optimization algorithms,

    J. Schulman, F. Wolski, P. Dhariwal, A. Radford, and O. Klimov, “Prox- imal policy optimization algorithms,” arXiv preprint arXiv:1707.06347, 2017

    2017 arXiv

  38. [46]

    Pathgen-1.6 m: 1.6 million pathology image- text pairs generation through multi-agent collaboration,

    Y . Sun, Y . Zhang, Y . Si, C. Zhu, Z. Shui, K. Zhang, J. Li, X. Lyu, T. Lin, and L. Yang, “Pathgen-1.6 m: 1.6 million pathology image- text pairs generation through multi-agent collaboration,” arXiv preprint arXiv:2407.00203, 2024

    2024 arXiv

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.