Pith. sign in

REVIEW 5 major objections 5 minor 2 cited by

Kvasir-VQA-x1: A Multimodal Dataset for Medical Reasoning and Robust MedVQA in Gastrointestinal Endoscopy

T0 review · 5 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper claims that Kvasir-VQA-x1 turns GI-endoscopy VQA from simple recall into a reasoning benchmark, and that fine-tuning on its 159,549 complexity-scored QA pairs lifts MedGemma and Qwen2.5-VL from 30–45% to 87–90% accuracy.

desk verdict A genuinely useful GI-endoscopy VQA dataset whose ground truth is the load-bearing assumption: the 'expert validation' is undocumented, so the benchmark's correctness is unverified, but the resource merits a serious referee if that audit is provided. read the letter →

arxiv 2506.09958 v1 pith:SJ7KPMZL submitted 2025-06-11 cs.CV cs.LG

Sushant Gautam , Michael A. Riegler , P{aa}l Halvorsen This is my paper
classification cs.CVcs.LG
keywords medicalvisualquestionansweringgastrointestinalendoscopymultimodaldatasetvision-languagemodelsperturbationsrobustnessbenchmarkcomplexityscoring
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper is trying to establish that the main barrier to reasoning-capable medical VQA is not the models but the datasets, and it builds a dataset that targets that barrier for gastrointestinal endoscopy. It expands the Kvasir-VQA corpus from 58,849 to 159,549 QA pairs by combining one, two, or three atomic questions into single prompts answered in naturalized clinical language, and it scores each new prompt by reasoning complexity from 1 to 3. It also supplies a robustness track of weakly perturbed images meant to mimic real endoscopy artifacts, with separate evaluation tracks for standard accuracy and robustness. The empirical case is that domain fine-tuning on this resource transforms off-the-shelf vision-language models: MedGemma reaches 87% and Qwen2.5-VL reaches 90% mean accuracy, against 30–45% for their base checkpoints. A sympathetic reader should care because the paper is aiming to give the field a harder, clinically oriented benchmark on which future MedVQA progress can be measured.

What carries the argument

The load-bearing mechanism is a three-level complexity scaffold built by combinatorial question merging. Starting from atomic QA pairs, sets of 1, 2, or 3 pairs are sampled; a local Qwen3-30B-A3B server naturalizes answers and merges questions into a coherent prompt, and the number of merged pairs becomes the integer complexity score from 1 to 3. This scaffold is what makes the dataset more than a larger pile of examples: it creates a graded sequence from factual recall to multi-hop synthesis, and it also powers the paper's LLM-as-adjudicator evaluation, which scores each answer aspect, such as polyp type or instrument presence, as correct or incorrect rather than relying on n-gram overlap alone.

What would settle it

Sample, say, 200 Level 2 and Level 3 question-answer pairs that were not used in any training run, have two or more independent gastroenterologists answer each from the image alone, and measure agreement with the dataset's labels; if a substantial fraction are judged unanswerable or answered differently across experts, the claim that the new pairs are clinically valid and image-grounded fails.

Watch

Extended reading notes

Core claim

On its own terms, the central discovery is that a curated expansion of an existing GI-endoscopy VQA dataset can turn a simple pattern-recognition benchmark into a reasoning benchmark without abandoning image-answerable ground truth. The dataset is built from 6,500 images from HyperKvasir and Kvasir-Instrument and the 58,849 atomic QA pairs of Kvasir-VQA; a local Qwen3-30B-A3B backbone merges up to three atomic pairs into one complex question and naturalizes the short answers into fluent medical prose, and the paper states that the pipeline's output was expert-validated. Each new pair carries a complexity score from 1 to 3 and references its original atomic pairs, which allows evaluation to be stratified by reasoning depth. The same images are weakly augmented with random resized cropping, rotation, affine shifts, and color jitter to produce a transformed robustness track. Fine-tuned MedGemma and Qwen2.5-VL checkpoints score 87% and 90% mean accuracy on LLM-adjudicated aspects, versus 30–45% for the base models, and the augmented-trained Qwen variant loses almost nothing on clean images while remaining stable on perturbed ones.

Load-bearing premise

The argument stands or falls on whether the new merged questions and naturalized answers are genuinely clinically correct and answerable from the image, and the paper reports that 'expert validation' happened without saying who did it, how many pairs were checked, or what the agreement was.

Editorial extensions

If this is right

  • Fine-tuning on Kvasir-VQA-x1 raises MedVQA accuracy from 30–45% to 87% for MedGemma and 90% for Qwen2.5-VL, showing that in-domain data, rather than scale alone, drives clinical competency.
  • Training on weakly augmented images keeps performance stable on both clean and perturbed inputs, with no measured trade-off on original images.
  • Complexity scoring enables stratified evaluation, and Level 2 questions often score above Level 1, suggesting that merged prompts can be less ambiguous than isolated atomic questions.
  • Level 3 questions remain the hardest, with errors compounding under all-or-nothing scoring, so the benchmark sets a clear target for multi-hop reasoning.
  • The dual-track normal/transformed evaluation provides a repeatable protocol for testing robustness to endoscopy-like imaging artifacts.

Reading between the lines

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

  • A testable extension is to use the complexity scores for curriculum learning, starting with Level 2 questions as the paper's own 'synthesis sweet spot' observation suggests, and comparing against curricula that begin with Level 1.
  • The LLM-as-adjudicator protocol inherits a homogeneity bias because the judge and the best-performing model share architectural lineage; an independent human-scored sample would strengthen any cross-model comparison.
  • Because augmented images are released only as regeneration scripts, a reproducibility test is to regenerate the exact transformed train/test splits and verify that the reported scores replicate.
  • The heavy class imbalance, with rare classes like landmark color having only hundreds of pairs, will push future work toward targeted augmentation or resampling rather than simply scaling total QA count.
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

5 major / 5 minor

Summary. The paper introduces Kvasir-VQA-x1, an expanded MedVQA dataset for gastrointestinal endoscopy built on Kvasir-VQA. It adds 159,549 new question-answer pairs generated with the Qwen3-30B-A3B language model by merging one, two, or three atomic QA pairs and naturalizing answers; each pair is assigned a complexity score (1–3) equal to the number of merged atomic pairs. The dataset also defines a robustness track using weak image augmentations (cropping, rotation, affine transforms, color jitter) and is released with code and data on GitHub and Hugging Face. The authors fine-tune MedGemma and Qwen2.5-VL with LoRA and evaluate them on original and transformed images using n-gram metrics and an LLM-based adjudicator, reporting that fine-tuning raises accuracy from roughly 30–45% to 87–90%.

Significance. If the dataset construction is sound, Kvasir-VQA-x1 would be a substantial community resource: it is large, publicly released, FAIR-aligned, stratified by a complexity score, and includes a robustness track with reproducible augmentation scripts. The paper is transparent about its pipeline and releases code, which supports reproducibility. The main claims—that the new QA pairs are clinically correct and that the complexity score reflects reasoning difficulty—are not yet established because the expert-validation step is undocumented and the complexity score is defined only as the number of merged atomic questions. The empirical evaluation is also weakened by an LLM adjudicator from the same family as the generator and one of the evaluated models, and by the absence of error bars and external comparisons. These issues are fixable with additional validation and analysis, so the work is a plausible candidate for a valuable benchmark, but the current manuscript does not fully support its central contribution.

major comments (5)
  1. [Section 2.3] The manuscript states that LLM-generated merged questions and naturalized answers were 'followed by expert validation' (Section 2.3) and thanks medical experts (Section 7), but it gives no details on the validation protocol: who the experts were, how many participated, how many QA pairs were reviewed, what inclusion/exclusion criteria were used, how disagreements were resolved, or any inter-annotator agreement metric. Since the 159,549 new QA pairs are the core dataset contribution and their correctness is the ground truth for all accuracy numbers in Section 4, this missing support is load-bearing. Please provide a detailed validation protocol and, ideally, a quantitative audit (e.g., number of pairs reviewed per expert, agreement rate, examples of corrected outputs), or at least a precise description of what 'expert validation' actually covered.
  2. [Section 2.3 vs. Section 1.4] Section 1.4 introduces 'quantitative complexity scores, capturing both visual and linguistic difficulty,' but Section 2.3 defines the complexity score as the number of atomic QA pairs merged (1, 2, or 3), and Table 4's per-class distributions follow exactly from that combinatorial sampling. The paper's own finding in Section 4.2.3 that Level 2 questions are sometimes answered more accurately than Level 1 (e.g., abnormality color: 67.16% vs. 56.69%) indicates that the score does not behave as a monotonic difficulty measure. Either revise the claim about what the score captures or provide evidence—such as human difficulty ratings or a separate visual- and linguistic-difficulty model—that the number of merged pairs tracks reasoning difficulty.
  3. [Section 2.5 and Section 4.2.5] The evaluation uses Qwen/Qwen3-30B-A3B as the LLM adjudicator, which is the same model family used in Section 2.3 to generate the QA pairs, and one of the evaluated models (Qwen2.5-VL) shares the Qwen lineage. The paper acknowledges this 'homogeneity bias' in Section 4.2.5 but still reports Qwen2.5-VL-ft accuracies of up to 90% as evidence of dataset effectiveness. Because the adjudicator's binary per-aspect scores underlie all categorical accuracies in Table 7, and no human-agreement study or independent-judge check is reported, the absolute accuracy numbers for Qwen-family models are not yet trustworthy. Please add at least one of: a human-annotated agreement study on a sample, an ensemble of heterogeneous adjudicators, or a sensitivity analysis showing that the relative model rankings are stable across judges.
  4. [Section 4.2.1 vs. Tables 5 and 7] Section 4.2.1 states that MedGemma-ft and Qwen2.5-VL-ft achieve 'mean accuracies of 87% and 90%', but Table 7 reports per-category accuracies without an overall mean row, and Table 5 reports eval accuracies of 84.97%, 85.91%, and 85.78% on a 1% held-out subset, which do not match the stated numbers. Please state exactly how the 87%/90% means are computed, including the aggregation formula over aspects and complexity levels, and reconcile these numbers with Table 5. Without a traceable computation, the headline result cannot be verified.
  5. [Section 2.2 vs. Section 2.3] Section 2.2 says 'Each entry in the dataset stores the original or augmented image, the newly formulated question, a naturalized answer, a complexity score, and the original question-answer pair(s),' but Section 2.3 says 'The released dataset only includes the original images, associated QA pairs, and metadata' and that augmented images are generated via scripts. This ambiguity affects the Transformed Setting: are the QA pairs in the transformed track identical to those in the original track with the image swapped, or were some QA pairs generated from augmented images so that their content depends on the augmentation? Please clarify the relationship between the two tracks and ensure that the release format and scripts make this unambiguous.
minor comments (5)
  1. [References] References [18] and [19] are the same CLEF paper, and references [34] and [35] are both Med-Flamingo; please deduplicate and cite each work once.
  2. [Abstract and Section 1] The text contains spacing artifacts such as 'T ransformed' and 'F AIR' (also in Section 3); these should be corrected for a polished final version.
  3. [Table 4] The 'finding presence' category appears only at Level 1, with no Level 2 or 3 rows; the accompanying discussion explains this, but it would help to add a footnote in the table itself so readers can see the reason without searching the text.
  4. [Section 2.6] The text says 'standardized hyperparameters' but Table 3 shows different effective batch sizes (36 for MedGemma, 32 for Qwen2.5); please rephrase or clarify which hyperparameters were held constant.
  5. [Section 3] The dataset structure lists 'img id' but does not specify whether this ID matches the original Kvasir-VQA image IDs or uses a new mapping; please clarify to avoid ambiguity for users who want to cross-reference the original dataset.

Circularity Check

2 steps flagged · score 5.0 of 10

Same Qwen3 model generates the ground-truth answers and adjudicates model outputs, making headline accuracies partly self-consistency scores; the complexity score is, by definition, only the number of merged QA pairs.

  1. other [Sections 2.3, 2.5, 4.2.1, 4.2.5]
    "For question merging and answer naturalization, we used a locally hosted inference server for Qwen3-30B-A3B [44] language model. ... The core of this evaluation is a programmatic pipeline that leverages the Qwen/Qwen3-30B-A3B model as an impartial medical examiner. ... Homogeneity bias in LLM-as-a-Judge: A key limitation of our evaluation protocol is the use of a Qwen-based LLM as the adjudicator, which introduces potential homogeneity bias."

    The same Qwen3-30B-A3B system that wrote the merged questions and naturalized answers (Section 2.3) is then used as the adjudicator that scores each model response against those LLM-written references (Section 2.5). The headline accuracies reported in Section 4.2.1 (MedGemma-ft 87%, Qwen2.5-VL-ft 90%) are produced by this pipeline, so they measure, at least in part, agreement with the judge-and-generator's own output distribution rather than independently established clinical correctness. The paper's own limitation note concedes that evaluated Qwen2.5-VL shares architectural lineage with the adjudicator, admitting self-enhancement bias.

  2. self definitional [Sections 1.4, 2.3, 4.2.3]
    "Each QA pair is additionally annotated with quantitative complexity scores, capturing both visual and linguistic difficulty. ... an integer complexity score ranging from 1 to 3, which reflects the number of original questions that have been combined."

    The dataset's 'complexity' variable is defined as the number of atomic QA pairs merged (1, 2, or 3), not as any independently measured reasoning or visual difficulty. Consequently, the claim that questions are 'stratified by complexity' to test 'deeper clinical reasoning' holds only by definition: Level 3 means 'three prompts were combined,' not 'empirically harder.' The Level-2-over-Level-1 pattern interpreted in Section 4.2.3 as a 'synthesis sweet spot' is then partly a construction artifact of merged prompts containing more referential context, rather than a discovered fact about reasoning difficulty. The complexity label and the merge count are the same variable by construction.

full rationale

This is a dataset-and-evaluation paper rather than a derivation chain, and its central resource has genuine external grounding: the images come from HyperKvasir and Kvasir-Instrument, and the new QA pairs are anchored to the human-annotated atomic QA pairs of Kvasir-VQA. That external anchor is real and keeps the paper from being wholly circular. However, the evaluation loop contains a serious self-reference: the same Qwen3-30B-A3B model produced the naturalized answers and merged questions and then served as the 'impartial medical examiner' scoring all model outputs against those self-generated references; the paper itself acknowledges the resulting homogeneity bias for same-lineage models such as Qwen2.5-VL. The headline fine-tuning gains are therefore not the result of an independent clinical measurement but are at least partly self-consistency scores. A milder definitional circularity also exists in the complexity construct, which is defined as the number of merged question pairs and then described as capturing visual and linguistic difficulty; the complexity-based analysis largely restates this construction choice. The undocumented 'expert validation' mentioned in Sections 2.3 and 7 is a missing-support issue rather than a circular step: it weakens the ground-truth claim but does not by itself reduce any number to its inputs. Overall, the paper is partially circular in its evaluation layer, but not equivalent to its inputs by definition.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The central claims rest primarily on the correctness of the LLM-generated QA pairs and the validity of the LLM-as-judge evaluation. The free parameters are design choices for the dataset and do not involve fitting a scientific model. No new physical or conceptual entities are introduced.

free parameters (3)
  • augmentation mix ratio = 23% original / 77% augmented
    Hand-chosen in Section 2.3; determines how often a QA pair is paired with an augmented image, but is not fitted to data.
  • complexity score = 1, 2, or 3 (count of merged atomic QA pairs)
    Defined in Section 2.3 as the number of original QA pairs combined; used to stratify difficulty, but does not measure linguistic or reasoning complexity independently.
  • augmentation parameters = RandomResizedCrop 0.9-1.0, rotation +/-10 degrees, translation 10%, ColorJitter 0.8-1.2
    Chosen to be weak (Section 2.3); no ablations justify these values.
assumptions (3)
  • domain assumption LLM-generated merged questions and naturalized answers are clinically correct and answerable from the image.
    Section 2.3 claims expert validation but does not describe the process; the dataset's ground truth depends on this.
  • domain assumption Weak augmentations preserve the semantic content required to answer each QA pair.
    Section 2.3 uses RandomResizedCrop and affine transforms; cropping could remove the anatomical region referenced by the question.
  • domain assumption Qwen3-30B-A3B as adjudicator produces binary aspect scores that correctly reflect clinical accuracy.
    Section 2.5; the LLM judge is not validated against human clinicians, and the paper acknowledges homogeneity bias with the Qwen-family models.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Kvasir-VQA-x1: A Multimodal Dataset for Medical Reasoning and Robust MedVQA in Gastrointestinal Endoscopy." pith.science (2026). https://pith.science/paper/SJ7KPMZL

@misc{pith2026250609958,
  author       = {Pith},
  title        = {Pith review of: Kvasir-VQA-x1: A Multimodal Dataset for Medical Reasoning and Robust MedVQA in Gastrointestinal Endoscopy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SJ7KPMZL}},
  note         = {Machine review of arXiv:2506.09958}
}
read the original abstract

Medical Visual Question Answering (MedVQA) is a promising field for developing clinical decision support systems, yet progress is often limited by the available datasets, which can lack clinical complexity and visual diversity. To address these gaps, we introduce Kvasir-VQA-x1, a new, large-scale dataset for gastrointestinal (GI) endoscopy. Our work significantly expands upon the original Kvasir-VQA by incorporating 159,549 new question-answer pairs that are designed to test deeper clinical reasoning. We developed a systematic method using large language models to generate these questions, which are stratified by complexity to better assess a model's inference capabilities. To ensure our dataset prepares models for real-world clinical scenarios, we have also introduced a variety of visual augmentations that mimic common imaging artifacts. The dataset is structured to support two main evaluation tracks: one for standard VQA performance and another to test model robustness against these visual perturbations. By providing a more challenging and clinically relevant benchmark, Kvasir-VQA-x1 aims to accelerate the development of more reliable and effective multimodal AI systems for use in clinical settings. The dataset is fully accessible and adheres to FAIR data principles, making it a valuable resource for the wider research community. Code and data: https://github.com/Simula/Kvasir-VQA-x1 and https://huggingface.co/datasets/SimulaMet/Kvasir-VQA-x1

Figures

Figures reproduced from arXiv: 2506.09958 by the authors.

Figure 1
Figure 1. Rank-normalized heatmap illustrating comparative performance rankings (1 = best, 5 [PITH_FULL_IMAGE:figures/full_fig_p014_1.png] view at source ↗
Figure 2
Figure 2. Radar plot showing absolute performance scores of five models (Gemma3-4B, [PITH_FULL_IMAGE:figures/full_fig_p015_2.png] view at source ↗
Figure 3
Figure 3. Model performance across different complexity levels. Accuracy scores are plotted for [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗

Discussion (0). Sign in to comment.

Forward citations

Cited by 2 Pith papers

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

  1. Measuring and Improving Complex-Atomic Answer Consistency in Endoscopic VQA

    cs.CV 2026-07 conditional novelty 5.0 of 10

    Vision-language models can answer complex endoscopic questions correctly while failing the same-image atomic sub-questions; EndoCA measures this gap and ASR reduces it.

  2. Multimodal AI for Gastrointestinal Diagnostics: Tackling VQA in MEDVQA-GI 2025

    cs.CV 2025-07 conditional novelty 3.0 of 10

    Fine-tuning Florence-2 on a 1% subset of Kvasir-VQA with medical image augmentations yields moderate VQA performance on gastrointestinal endoscopy questions.

Reference graph

Works this paper leans on

54 extracted references · 11 canonical work pages · cited by 2 Pith papers

  1. [1]

    Foundation metrics for evaluating effectiveness of healthcare conversations powered by generative AI.npj Digital Med., 7(82):1–14, March 2024

    Mahyar Abbasian, Elahe Khatibi, Iman Azimi, David Oniani, Zahra Shakeri Hossein Abad, Alexander Thieme, Ram Sriram, Zhongqi Yang, Yanshan Wang, Bryant Lin, et al. Foundation metrics for evaluating effectiveness of healthcare conversations powered by generative AI.npj Digital Med., 7(82):1–14, March 2024. ISSN 2398-6352. doi:10.1038/s41746- 024-01074-z

    doi:10.1038/s41746- 2024

  2. [2]

    Flamingo: a Visual Language Model for Few-Shot Learning.arXiv, April 2022

    Jean-Baptiste Alayrac, Jeff Donahue, Pauline Luc, Antoine Miech, Iain Barr, Yana Hasson, Karel Lenc, Arthur Mensch, Katie Millican, Malcolm Reynolds, et al. Flamingo: a Visual Language Model for Few-Shot Learning.arXiv, April 2022. doi:10.48550/arXiv.2204.14198

    arXiv doi:10.48550/arxiv.2204.14198 2022

  3. [3]

    A deep learning framework for quality assessment and restoration in video endoscopy.arXiv, April 2019

    Sharib Ali, Felix Zhou, Adam Bailey, Barbara Braden, James East, Xin Lu, and Jens Rittscher. A deep learning framework for quality assessment and restoration in video endoscopy.arXiv, April 2019. doi:10.1016/j.media.2020.101900

    arXiv 2019

  4. [4]

    Qwen2.5-vl technical report.arXiv preprint arXiv:2502.13923, 2025

    Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, Humen Zhong, Yuanzhi Zhu, Mingkun Yang, Zhaohai Li, Jianqiang Wan, Pengfei Wang, Wei Ding, Zheren Fu, Yiheng Xu, Jiabo Ye, Xi Zhang, Tianbao Xie, Zesen Cheng, Hang Zhang, Zhibo Yang, Haiyang Xu, and Junyang Lin. Qwen2.5-vl technical report.ar...

    arXiv 2025

  5. [5]

    Vision–Language Model for Visual Question Answering in Medical Imagery.Bioengineering, 10(3):380, March 2023

    Yakoub Bazi, Mohamad Mahmoud Al Rahhal, Laila Bashmal, and Mansour Zuair. Vision–Language Model for Visual Question Answering in Medical Imagery.Bioengineering, 10(3):380, March 2023. ISSN 2306-5354. doi:10.3390/bioengineering10030380

    doi:10.3390/bioengineering10030380 2023

  6. [6]

    Smedsrud, Steven Hicks, Debesh Jha, Sigrun L

    Hanna Borgli, Vajira Thambawita, Pia H. Smedsrud, Steven Hicks, Debesh Jha, Sigrun L. Eskeland, Kristin Ranheim Randel, Konstantin Pogorelov, Mathias Lux, Duc Tien Dang Nguyen, Dag Johansen, Carsten Griwodz, H ˚ akon K. Stens- land, Enrique Garcia-Ceja, Peter T. Schmidt, Hugo L. Hammer, Michael A. Riegler, P ˚ al Halvorsen, and Thomas de Lange. HyperKvasi...

    2020

  7. [7]

    Iglovikov, and Alexandr A

    Alexander Buslaev, Alex Parinov, Eugene Khvedchenya, Vladimir I. Iglovikov, and Alexandr A. Kalinin. Albumentations: fast and flexible image augmentations.arXiv, September 2018. doi:10.3390/info11020125

    doi:10.3390/info11020125 2018

  8. [8]

    A Simple Framework for Contrastive Learning of Visual Representations.arXiv, February 2020

    Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A Simple Framework for Contrastive Learning of Visual Representations.arXiv, February 2020. doi:10.48550/arXiv.2002.05709

    arXiv doi:10.48550/arxiv.2002.05709 2020

Show all 54 references
  1. [9]

    R-LLaV A: Improving Med-VQA Understanding through Visual Region of Interest.arXiv, October 2024

    Xupeng Chen, Zhixin Lai, Kangrui Ruan, Shichu Chen, Jiaxiang Liu, and Zuozhu Liu. R-LLaV A: Improving Med-VQA Understanding through Visual Region of Interest.arXiv, October 2024. doi:10.48550/arXiv.2410.20327

    2024 doi arXiv

  2. [10]

    Generative Models in Medical Visual Question Answering: A Survey.Appl

    Wenjie Dong, Shuhao Shen, Yuqiang Han, Tao Tan, Jian Wu, and Hongxia Xu. Generative Models in Medical Visual Question Answering: A Survey.Appl. Sci., 15(6):2983, March 2025. ISSN 2076-3417. doi:10.3390/app15062983

    2025 doi

  3. [11]

    LLM-based NLG Evaluation: Current Status and Challenges.Computational Linguistics, pages 1–27, 2025

    Mingqi Gao, Xinyu Hu, Xunjian Yin, Jie Ruan, Xiao Pu, and Xiaojun Wan. LLM-based NLG Evaluation: Current Status and Challenges.Computational Linguistics, pages 1–27, 2025. doi:10.1162/coli a 00561

    2025 doi

  4. [12]

    Hicks, Vajira Thambawita, P ˚ al Halvorsen, and Michael A

    Sushant Gautam, Andrea Stor ˚ as, Cise Midoglu, Steven A. Hicks, Vajira Thambawita, P ˚ al Halvorsen, and Michael A. Riegler. Kvasir-vqa: A text-image pair gi tract dataset. InProceedings of the First International Workshop on Vision- Language Models for Biomedical Application...

    2024

  5. [13]

    Medgemma hugging face, May 2025

    Google. Medgemma hugging face, May 2025. URLhttps://huggingface.co/collections/google/ medgemma-release-680aade845f90bec6a3f60c4. [Online; accessed 29. May 2025]

    2025

  6. [14]

    LaPA: Latent Prompt Assist Model For Medical Visual Question Answering.arXiv, April 2024

    Tiancheng Gu, Kaicheng Yang, Dongnan Liu, and Weidong Cai. LaPA: Latent Prompt Assist Model For Medical Visual Question Answering.arXiv, April 2024. doi:10.48550/arXiv.2404.13039

    2024 doi arXiv

  7. [15]

    DiN: Diffusion Model for Robust Medical VQA with Semantic Noisy Labels.arXiv, March 2025

    Erjian Guo, Zhen Zhao, Zicheng Wang, Tong Chen, Yunyi Liu, and Luping Zhou. DiN: Diffusion Model for Robust Medical VQA with Semantic Noisy Labels.arXiv, March 2025. doi:10.48550/arXiv.2503.18536

    2025 doi arXiv

  8. [16]

    Vision-language models for medical report generation and visual question answering: a review.Front

    Iryna Hartsock and Ghulam Rasool. Vision-language models for medical report generation and visual question answering: a review.Front. Artif. Intell., 7:1430984, November 2024. ISSN 2624-8212. doi:10.3389/frai.2024.1430984

    2024

  9. [17]

    DeBERTaV3: Improving DeBERTa using ELECTRA-Style Pre-Training with Gradient-Disentangled Embedding Sharing.arXiv, November 2021

    Pengcheng He, Jianfeng Gao, and Weizhu Chen. DeBERTaV3: Improving DeBERTa using ELECTRA-Style Pre-Training with Gradient-Disentangled Embedding Sharing.arXiv, November 2021. doi:10.48550/arXiv.2111.09543

    2021 doi arXiv

  10. [19]

    Overview of imageclefmedical 2023-medical visual question answering for gastrointestinal tract

    Steven Hicks, Andrea M Stor ˚ as, P ˚ al Halvorsen, Thomas de Lange, Michael Riegler, and Vajira Thambawita. Overview of imageclefmedical 2023-medical visual question answering for gastrointestinal tract. InCLEF (Working Notes), pages 1316–1327, 2023

    2023

  11. [20]

    Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen

    Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. LoRA: Low-Rank Adaptation of Large Language Models.arXiv, June 2021. doi:10.48550/arXiv.2106.09685

    2021 doi arXiv

  12. [21]

    Summers, and Yingying Zhu

    Xinyue Hu, Lin Gu, Qiyuan An, Mengliang Zhang, Liangchen Liu, Kazuma Kobayashi, Tatsuya Harada, Ronald M. Summers, and Yingying Zhu. Expert Knowledge-Aware Image Difference Graph Representation Learning for Difference- Aware Medical Visual Question Answering.arXiv, July 2023. ...

    2023

  13. [22]

    Sadman Hafiz, Jamin Rahman Jim, Md

    Tauhidul Islam, Md. Sadman Hafiz, Jamin Rahman Jim, Md. Mohsin Kabir, and M. F. Mridha. A systematic review of deep learning data augmentation in medical imaging: Recent advances and future research directions.Healthcare Analytics, 5:100340, June 2024. ISSN 2772-4425. doi:10.1...

    2024

  14. [23]

    Hicks, Vajira Thambawita, Enrique Garcia-Ceja, Michael A

    Debesh Jha, Sharib Ali, Krister Emanuelsen, Steven A. Hicks, Vajira Thambawita, Enrique Garcia-Ceja, Michael A. Riegler, Thomas de Lange, Peter T. Schmidt, H ˚ avard D. Johansen, Dag Johansen, and P ˚ al Halvorsen. Kvasir-Instrument: Diagnostic and Therapeutic Tool Segmentatio...

    2021 doi

  15. [24]

    Lau, Soumya Gayen, Asma Ben Abacha, and Dina Demner-Fushman

    Jason J. Lau, Soumya Gayen, Asma Ben Abacha, and Dina Demner-Fushman. A dataset of clinically generated vi- sual questions and answers about radiology images.Sci. Data, 5(180251):1–10, November 2018. ISSN 2052-4463. doi:10.1038/sdata.2018.251

    2018 doi

  16. [25]

    Meteor: an automatic metric for MT evaluation with high levels of correlation with human judgments

    Alon Lavie and Abhaya Agarwal. Meteor: an automatic metric for MT evaluation with high levels of correlation with human judgments. InDL Hosted proceedings, pages 228–231. Association for Computational Linguistics, June 2007. doi:10.5555/1626355.1626389

    2007

  17. [26]

    LLaV A-Med: Training a Large Language-and-Vision Assistant for Biomedicine in One Day.arXiv, June 2023

    Chunyuan Li, Cliff Wong, Sheng Zhang, Naoto Usuyama, Haotian Liu, Jianwei Yang, Tristan Naumann, Hoifung Poon, and Jianfeng Gao. LLaV A-Med: Training a Large Language-and-Vision Assistant for Biomedicine in One Day.arXiv, June 2023. doi:10.48550/arXiv.2306.00890

    2023 doi arXiv

  18. [27]

    Less Could Be Better: Parameter-efficient Fine-tuning Advances Medical Vision Foundation Models.arXiv, January 2024

    Chenyu Lian, Hong-Yu Zhou, Yizhou Yu, and Liansheng Wang. Less Could Be Better: Parameter-efficient Fine-tuning Advances Medical Vision Foundation Models.arXiv, January 2024. doi:10.48550/arXiv.2401.12215

    2024 doi arXiv

  19. [28]

    Candidate-Heuristic In-Context Learning: A new framework for enhancing medical visual question answering with LLMs.Information Processing & Management, 61(5):103805, September 2024

    Xiao Liang, Di Wang, Haodi Zhong, Quan Wang, Ronghan Li, Rui Jia, and Bo Wan. Candidate-Heuristic In-Context Learning: A new framework for enhancing medical visual question answering with LLMs.Information Processing & Management, 61(5):103805, September 2024. ISSN 0306-4573. d...

    2024

  20. [29]

    Rouge: A package for automatic evaluation of summaries

    Chin-Yew Lin. Rouge: A package for automatic evaluation of summaries. InText summarization branches out, pages 74–81, 2004

    2004

  21. [30]

    Medical visual question answering: A survey.Artif

    Zhihong Lin, Donghao Zhang, Qingyi Tao, Danli Shi, Gholamreza Haffari, Qi Wu, Mingguang He, and Zongyuan Ge. Medical visual question answering: A survey.Artif. Intell. Med., 143:102611, September 2023. ISSN 0933-3657. doi:10.1016/j.artmed.2023.102611

    2023

  22. [31]

    Slake: A Semantically-Labeled Knowledge-Enhanced Dataset For Medical Visual Question Answering

    Bo Liu, Li-Ming Zhan, Li Xu, Lin Ma, Yan Yang, and Xiao-Ming Wu. Slake: A Semantically-Labeled Knowledge-Enhanced Dataset For Medical Visual Question Answering. InIEEE 18th International Symposium on Biomedical Imaging (ISBI), pages 13–16. IEEE, 2021. doi:10.1109/ISBI48211.202...

    2021

  23. [32]

    Visual Instruction Tuning.arXiv, April 2023

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual Instruction Tuning.arXiv, April 2023. doi:10.48550/arXiv.2304.08485

    2023 doi arXiv

  24. [33]

    Peft: State-of- the-art parameter-efficient fine-tuning methods

    Sourab Mangrulkar, Sylvain Gugger, Lysandre Debut, Younes Belkada, Sayak Paul, and Benjamin Bossan. Peft: State-of- the-art parameter-efficient fine-tuning methods. InPeft: State-of-the-art parameter-efficient fine-tuning methods. 2022

    2022

  25. [35]

    Med-Flamingo: a Multimodal Medical Few-shot Learner.arXiv, July 2023

    Michael Moor, Qian Huang, Shirley Wu, Michihiro Yasunaga, Cyril Zakka, Yash Dalmia, Eduardo Pontes Reis, Pranav Rajpurkar, and Jure Leskovec. Med-Flamingo: a Multimodal Medical Few-shot Learner.arXiv, July 2023. doi:10.48550/arXiv.2307.15189

    2023 doi arXiv

  26. [36]

    GPT-4 Technical Report.arXiv, March 2023

    OpenAI, Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, et al. GPT-4 Technical Report.arXiv, March 2023. doi:10.48550/arXiv.2303.08774

    2023 doi arXiv

  27. [37]

    Chaudhari, et al

    Sophie Ostmeier, Justin Xu, Zhihong Chen, Maya Varma, Louis Blankemeier, Christian Bluethgen, Arne Edward Michal- son, Michael Moseley, Curtis Langlotz, Akshay S. Chaudhari, et al. GREEN: Generative Radiology Report Evaluation and Error Notation.arXiv, May 2024. doi:10.18653/v...

    2024 doi

  28. [38]

    BLEU: a method for automatic evaluation of ma- chine translation

    Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. BLEU: a method for automatic evaluation of ma- chine translation. InDL Hosted proceedings, pages 311–318. Association for Computational Linguistics, July 2002. doi:10.3115/1073083.1073135

    2002

  29. [39]

    chrF: character n-gram F-score for automatic MT evaluation.ACL Anthology, pages 392–395, September

    Maja Popovi´ c. chrF: character n-gram F-score for automatic MT evaluation.ACL Anthology, pages 392–395, September

  30. [40]

    ZeRO: Memory Optimizations Toward Training Trillion Parameter Models.arXiv, October 2019

    Samyam Rajbhandari, Jeff Rasley, Olatunji Ruwase, and Yuxiong He. ZeRO: Memory Optimizations Toward Training Trillion Parameter Models.arXiv, October 2019. doi:10.48550/arXiv.1910.02054

    2019 doi arXiv

  31. [41]

    Samaan, Reem Al Shabeeb, Farah Ladak, Jamie O

    Seyed Amir Ahmad Safavi-Naini, Shuhaib Ali, Omer Shahab, Zahra Shahhoseini, Thomas Savage, Sara Rafiee, Jamil S. Samaan, Reem Al Shabeeb, Farah Ladak, Jamie O. Yang, et al. Vision-Language and Large Language Model Perfor- mance in Gastroenterology: GPT, Claude, Llama, Phi, Mis...

    2024 doi arXiv

  32. [42]

    Thibault Sellam, Dipanjan Das, and Ankur P. Parikh. BLEURT: Learning Robust Metrics for Text Generation.arXiv, April 2020. doi:10.48550/arXiv.2004.04696

    2020 doi arXiv

  33. [43]

    Pfohl, Heather Cole-Lewis, et al

    Karan Singhal, Tao Tu, Juraj Gottweis, Rory Sayres, Ellery Wulczyn, Mohamed Amin, Le Hou, Kevin Clark, Stephen R. Pfohl, Heather Cole-Lewis, et al. Toward expert-level medical question answering with large language models.Nat. Med., 31(3):943–950, March 2025. ISSN 1546-170X. d...

    2025 doi

  34. [44]

    Qwen3 technical report, 2025

    Qwen Team. Qwen3 technical report, 2025. URLhttps://arxiv.org/abs/2505.09388

    2025 arXiv

  35. [45]

    Sanders, Yuchen Liu, Kennarey Seang, Bach Xuan Tran, Atanas G

    Xiaofei Wang, Hayley M. Sanders, Yuchen Liu, Kennarey Seang, Bach Xuan Tran, Atanas G. Atanasov, Yue Qiu, Shenglan Tang, Josip Car, Ya Xing Wang, et al. ChatGPT: promise and challenges for deployment in low- and middle-income coun- tries.Lancet Regional Health – Western Pacifi...

    2023

  36. [46]

    Wilkinson, Michel Dumontier, IJsbrand Jan Aalbersberg, Gabrielle Appleton, Myles Axton, Arie Baak, Niklas Blomberg, Jan-Willem Boiten, Luiz Bonino da Silva Santos, Philip E

    Mark D. Wilkinson, Michel Dumontier, IJsbrand Jan Aalbersberg, Gabrielle Appleton, Myles Axton, Arie Baak, Niklas Blomberg, Jan-Willem Boiten, Luiz Bonino da Silva Santos, Philip E. Bourne, et al. The F AIR Guiding Principles for scien- tific data management and stewardship.Sc...

    2016 doi

  37. [47]

    Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, R´ emi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Yacine Jernite, Julien 23 Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger,...

    2020

  38. [48]

    Parameter-Efficient Fine-Tuning Methods for Pretrained Language Models: A Critical Review and Assessment.arXiv, December 2023

    Lingling Xu, Haoran Xie, Si-Zhao Joe Qin, Xiaohui Tao, and Fu Lee Wang. Parameter-Efficient Fine-Tuning Methods for Pretrained Language Models: A Critical Review and Assessment.arXiv, December 2023. doi:10.48550/arXiv.2312.12148

    2023 doi arXiv

  39. [49]

    MedFrameQA: A Multi-Image Medical VQA Benchmark for Clinical Reasoning.arXiv, May 2025

    Suhao Yu, Haojin Wang, Juncheng Wu, Cihang Xie, and Yuyin Zhou. MedFrameQA: A Multi-Image Medical VQA Benchmark for Clinical Reasoning.arXiv, May 2025. doi:10.48550/arXiv.2505.16964

    2025 doi

  40. [50]

    Fine-grained Adaptive Visual Prompt for Generative Medical Visual Question Answering.AAAI, 39(9):9662–9670, April 2025

    Ting Yu, Zixuan Tong, Jun Yu, and Ke Zhang. Fine-grained Adaptive Visual Prompt for Generative Medical Visual Question Answering.AAAI, 39(9):9662–9670, April 2025. ISSN 2374-3468. doi:10.1609/aaai.v39i9.33047

    2025 doi

  41. [51]

    Medical Visual Question Answering via Conditional Reasoning

    Li-Ming Zhan, Bo Liu, Lu Fan, Jiaxin Chen, and Xiao-Ming Wu. Medical Visual Question Answering via Conditional Reasoning. InACM Conferences, pages 2345–2354. Association for Computing Machinery, New York, NY, USA, October

  42. [52]

    Weinberger, and Yoav Artzi

    Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q. Weinberger, and Yoav Artzi. BERTScore: Evaluating Text Generation with BERT.arXiv, April 2019. doi:10.48550/arXiv.1904.09675

    2019 doi arXiv

  43. [53]

    PMC-VQA: Visual Instruction Tuning for Medical Visual Question Answering.arXiv, May 2023

    Xiaoman Zhang, Chaoyi Wu, Ziheng Zhao, Weixiong Lin, Ya Zhang, Yanfeng Wang, and Weidi Xie. PMC-VQA: Visual Instruction Tuning for Medical Visual Question Answering.arXiv, May 2023. doi:10.48550/arXiv.2305.10415

    2023 doi arXiv

  44. [54]

    Swift:a scalable lightweight infrastructure for fine-tuning, 2024

    Yuze Zhao, Jintao Huang, Jinghan Hu, Xingjun Wang, Yunlin Mao, Daoze Zhang, Zeyinzi Jiang, Zhikai Wu, Baole Ai, Ang Wang, Wenmeng Zhou, and Yingda Chen. Swift:a scalable lightweight infrastructure for fine-tuning, 2024. URL https://arxiv.org/abs/2408.05517. 24

    2024 arXiv

  45. [2015]

    doi:10.18653/v1/W15-3049

    doi

  46. [2020]

    doi:10.1145/3394171.3413761

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.