Pith. sign in

REVIEW 4 major objections 6 minor 32 references

How Good are Foundation Models in Longitudinal MRI Disease Progression Reasoning?

T0 review · 4 major / 6 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read Current vision-language models can order longitudinal MRI scans but struggle to recognize what type of change occurred, this paper claims.

desk verdict Useful new longitudinal multi-view MRI benchmark, but the GPT-5 metadata privilege means the 'systematic failure on change-type recognition' claim needs a human baseline and an information-asymmetry ablation before it is fully calibrated. read the letter →

arxiv 2608.13309 v1 pith:2R32TRKD submitted 2026-08-13 cs.CV

Wafa Al Ghallabi , Ritesh Thawkar , Sara Ghaboura , Omkar Thawakar , Numan Saeed , Dana Al Nuaimi , Ajnas Alkatheeri , Salman Khan
show 1 more author
Fahad Shahbaz Khan
This is my paper · ORCID
classification cs.CV
keywords longitudinalMRIvision-languagemodelmedicalvisualquestionansweringdiseaseprogressiontemporalreasoningmulti-viewbenchmarkchangelocalization
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

This paper claims that the standard way of testing vision-language models in radiology—one image, one question—skips the part of the job that matters most: comparing scans across time and across anatomical planes. To close that gap, it introduces the Time-Aware Multi-View MRI Benchmark, a set of 3,920 expert-verified question-answer pairs drawn from 890 patients and more than 3,200 longitudinal MRI timepoints across seven cohorts and four disease groups. When the authors run 16 current vision-language models on this benchmark, they find that models can usually reconstruct the chronological order of scans but systematically fail to recognize what type of change occurred, such as growth versus regression or volumetric change. The paper's claim is that this failure is not an artifact of the test, because every retained question was independently approved by two radiologists. If the benchmark is valid, it gives the field a reusable instrument for measuring whether a model can actually follow a disease over time.

What carries the argument

The load-bearing instrument is the benchmark itself: 3,920 question-answer pairs in open-ended, multiple-choice, and binary formats, built by generating candidate questions from longitudinal multi-view MRI inputs and retaining only cases that two board-certified radiologists independently approved, a 72 percent acceptance decision. The paper's named evaluation object is the Time-Aware Composite (TAC) metric, defined as $TAC = 0.5 \times \text{TEDS} + 0.2 \times \text{Trend-F1} + 0.2 \times \text{SignAcc} + 0.1 \times \text{Coverage}$, where TEDS scores how well a model's predicted change sequence aligns with the reference sequence, Trend-F1 scores progression-versus-regression detection, SignAcc scores change-direction correctness, and Coverage scores completeness of interval reasoning. TAC is what turns raw answer accuracy into a claim about temporal coherence, and the five task categories are what let the authors separate simple chronology from genuine progression understanding.

What would settle it

Have a fresh panel of radiologists, blinded to the benchmark's gold labels, independently re-answer a random subsample of the 3,920 questions, and separately score the questions that were discarded during dual review; if the panel disagrees with the gold labels on a substantial fraction, or if the discarded cases are the ambiguous growth-versus-pseudoprogression ones, then the benchmark's validity—and the conclusion that models fail on change-type recognition—is measuring the generator's blind spots rather than clinical competence.

Watch

Extended reading notes

Core claim

The central discovery, on the paper's own terms, is a capability split: temporal ordering is tractable for current vision-language models, while change-type recognition is not. Across the 16 evaluated models, Time-Aware Composite scores are moderate (0.57–0.80) and chronology scores are high, but Trend-F1 and sign accuracy are low, meaning models identify that something changed less reliably than they identify when it changed. The benchmark also produces a second finding: feeding axial, coronal, and sagittal views together improves spatial localization in a two-stage agent workflow, but the extra views degrade temporal ordering in smaller open-source models. The paper interprets this as evidence that current 2D-input architectures lack the geometric and cross-timepoint machinery needed for longitudinal MRI reasoning, and that domain specialization alone does not fix it.

Load-bearing premise

The load-bearing premise is that the expert-approved gold answers are clinically correct for the questions asked, so that a model missing them is actually failing at longitudinal MRI reasoning.

Editorial extensions

If this is right

  • Zero-shot performance on this benchmark becomes a reusable baseline: later work can compare longitudinal MRI models on identical questions and splits.
  • Current vision-language models are not ready to drive clinical progression decisions such as growth-versus-regression or volumetric measurement, even when they localize the right region.
  • Adding more anatomical views is not universally helpful; compact architectures need adaptive view selection to avoid degrading temporal reasoning.
  • The gap between strong localization and weak quantification points future work toward explicit 3D geometric priors and cross-timepoint attention rather than larger versions of the same 2D models.

Reading between the lines

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

  • A human-radiologist baseline on the same 3,920 questions is not reported; without it, a 35 percent final-accuracy figure has no calibration for what expert-level performance looks like on this hard test.
  • Because 28 percent of generated candidates were rejected or discarded, the benchmark may under-represent the clinically ambiguous cases—precisely the ones where progression decisions are hardest.
  • High chronology scores could partly reflect side information such as image order, filenames, or metadata rather than visible disease features; a version that strips those cues would test whether models truly see progression.
  • If the benchmark's candidate-generation step carries a bias that survives radiologist review, the reported failure pattern measures the generator's blind spots; comparing the rejected candidates' content against the retained set would reveal this.
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 / 6 minor

Summary. The paper introduces the Time-Aware Multi-View MRI Benchmark, a longitudinal MRI reasoning benchmark with 3,920 expert-verified QA pairs derived from 890 patients across seven public cohorts and five task categories (Temporal Reasoning, Disease Progression, Structured Localization Guidance, Temporal Sequence Ordering, and Change Localization Over Time). Questions are generated by GPT-5 from longitudinal multi-sequence MRI inputs plus clinical metadata, then independently verified by two board-certified radiologists, with a 72% acceptance rate. The paper evaluates 16 vision-language models zero-shot on the benchmark, reporting final accuracy, a Reasoning Score, and a hand-weighted Time-Aware Composite (TAC) metric. The main findings are that models achieve moderate temporal alignment (TAC 0.57-0.80) but weak Trend-F1 (0.19-0.63) and Sign Accuracy (0.42-0.74), leading the authors to conclude that current VLMs handle temporal ordering reliably but consistently fail on change-type recognition. A secondary agentic-workflow experiment on a glioblastoma subset reports that multi-view inputs improve progression localization but degrade temporal ordering in smaller open-source models, and that all tested models score below 16% on change segmentation/quantification.

Significance. If the benchmark's validity holds, it would be a valuable and reusable resource: it is the first longitudinal, multi-view MRI QA benchmark of its kind, covers diverse pathologies (glioblastoma, brain metastases, neurodegeneration, vestibular schwannoma), and the authors release code, evaluation splits, and the dataset, which is a substantive community contribution. The falsifiable claim that current VLMs fail at change-direction recognition and volumetric quantification, despite strong temporal ordering, is clinically relevant and would be an important negative result for deployment decisions. However, the significance of these findings is contingent on the benchmark's ground truth being recoverable from the images the models actually see and on the evaluation metrics being statistically grounded. As written, the paper lacks a human radiologist baseline, inter-rater agreement statistics, confidence intervals, and a treatment of the metadata asymmetry between question generation and model evaluation. These gaps do not necessarily invalidate the benchmark, but they currently prevent the central 'systematic failure' conclusion from being fully supported.

major comments (4)
  1. [Section 2.3 and Section 4] The reference answers are generated by GPT-5 prompted with 'clinical annotations where available (RANO scores, MGMT/IDH status, tumor volumes)' in Section 2.3, while the zero-shot evaluation in Section 4 uses 'multi-sequence, multi-view MRI inputs' only. If any Disease Progression or Change Localization questions depend on those annotations, the low Trend-F1 and Sign Accuracy values in Table 1 are an expected consequence of withholding information from the models, not evidence of a specific deficiency in change-type recognition. Please provide an ablation that either generates reference answers without metadata or supplies the same metadata to the evaluated models, and show that the conclusions are unchanged. Without this, the central claim in Section 4 that current VLMs 'consistently fail on change-type recognition' is not established.
  2. [Section 2.3] The validation pipeline reports a 72% acceptance rate and dual radiologist approval, but it does not report inter-rater agreement (e.g., Cohen's kappa) or a human radiologist baseline on the actual image-only inputs. The two radiologists were verifying GPT-5's candidates with access to the same clinical metadata, so their approval does not calibrate whether the correct answers are recoverable from images alone. Please report inter-rater agreement, include a human baseline on a representative sample using the same inputs as the models, and analyze the persistent disagreements that were discarded (for example, whether they are concentrated in clinically ambiguous cases that matter most).
  3. [Section 3, Eq. (1)] The TAC metric is defined as TAC = 0.5*TEDS + 0.2*Trend-F1 + 0.2*SignAcc + 0.1*Coverage with no justification, no sensitivity analysis, and no formal definitions of its four components. The paper's central claims about relative model capability are expressed through TAC and its components, so load-bearing conclusions rest on this hand-weighted composite. Please provide formal definitions of TEDS, Trend-F1, Sign Accuracy, and Coverage, test the sensitivity of the conclusions to the choice of weights, and report confidence intervals or bootstrap errors for the point estimates in Tables 1 and 2, which currently have none.
  4. [Section 4, Table 2] The multi-view ablation is performed on a single cohort subset (UCSF-GBM, 1,192 samples) with six models and no statistical error bars. The conclusion that 'multi-view inputs improve spatial localization yet degrade temporal reasoning in compact architectures' rests on per-model deltas such as Qwen3-VL-8B: -8.0 pp and MedGemma-4B: -5.8 pp, which could easily fall within prompt or sampling noise. Please provide repeated-run or bootstrap intervals, and ideally evaluate on at least one additional cohort, before generalizing this result.
minor comments (6)
  1. [Section 3 and Table 1] The Chronology metric reported in Table 1 is mentioned in Section 3 but never defined; please specify how it is computed and how it differs from Temporal Sequence Ordering accuracy.
  2. [Table 2] The column headers 'Global Ch., Seg. Ch., Quant., Temp. Ord., Prog. Loc.' are cryptic; please expand them to full task names in the table or a footnote.
  3. [Section 2.3] The Reasoning Score is said to be 'computed following LlamaV-o1 [21]' with no details of how reasoning consistency, temporal alignment, and factual accuracy are scored for these new MRI tasks; please provide the rubric or a reference to an appendix.
  4. [Section 2.1] The normalization description 'p1 top 99.5' and 'p 2 top 98' is ambiguous; clarify whether these are percentile clip bounds and whether the adaptive ceiling extension was applied consistently across all cohorts and sequences.
  5. [Section 2.3] The paper states there are 3,920 QA pairs from 890 patients over 3,200 timepoints but does not state whether multiple QA pairs from the same patient are kept in the same data split; please clarify the patient-level split strategy, as leakage across timepoints of the same patient could inflate scores.
  6. [Figure 2] Figure 2 is dense and hard to read at its current resolution; consider splitting it into separate panels or providing a higher-resolution version.

Circularity Check

0 steps flagged · score 1.0 of 10

No circular derivation found: the benchmark labels are radiologist-verified, the TAC weights are hand-set rather than fitted, and the only self-citation (LlamaV-o1 RS) is minor and not load-bearing.

full rationale

The paper's central claims are empirical benchmark measurements, not derivations that reduce to their inputs. Ground-truth QA pairs are generated by GPT-5, but each candidate is independently verified by two board-certified radiologists with a 72% acceptance rate and persistent disagreements discarded, so the reference answers are not defined by the evaluated models' outputs. The TAC metric (Eq. 1) uses hand-set weights (0.5/0.2/0.2/0.1) and is not fitted to model behavior; the headline finding of weak Trend-F1/SignAcc is read directly from Table 1 rather than produced by a weighted composite. The only self-referential element is the Reasoning Score, adopted from LlamaV-o1 [21], a paper with overlapping authors; however, RS is not load-bearing for the paper's main conclusion, which rests on final accuracy, Trend-F1, SignAcc, and TAC defined in this paper. The use of GPT-5 in label generation while GPT-5.2 is among the evaluated models, and the provision of clinical metadata to the generator but not to zero-shot models, are benchmark-validity concerns (information asymmetry and contamination) but not circularity: radiologist verification breaks the identity between the label source and the evaluated model outputs, and no equation or fitted parameter reduces to its own input.

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

The central claims rest on label validity, slice representativeness, and the TAC composite. None of these is externally benchmarked in the paper. The TAC weights are the main hand-set numbers that change the headline ranking; preprocessing percentiles are secondary. I found no invented physical entities.

free parameters (2)
  • TAC composite weights = 0.5, 0.2, 0.2, 0.1
    Equation (1). Hand-set weights directly determine the headline ranking; no sensitivity analysis or external validation is given.
  • Sequence-specific normalization percentiles = p1-99.5 for T1CE, p2-98 for T2 and FLAIR
    Section 2.1. Chosen by hand to preserve intensity extremes; affects the pixel inputs and therefore all model scores, though not the central task structure.
assumptions (4)
  • domain assumption GPT-5 generated candidates, dual-approved by two radiologists with persistent disagreements discarded, form unbiased and clinically correct ground truth.
    Section 2.3 reports a 72% acceptance rate and no inter-rater agreement or human performance baseline; label validity is the basis for every accuracy and TAC number.
  • domain assumption Extracting 9-12 2D slices through the tumor centroid per timepoint preserves the interval changes the tasks ask about.
    Section 2.2 relies on representative slices; changes outside the selected slices are invisible to the model and to the benchmark.
  • ad hoc to paper The hand-set TAC combination (0.5*TEDS + 0.2*Trend-F1 + 0.2*SignAcc + 0.1*Coverage) measures temporal reasoning fidelity.
    Equation (1) in Section 3 introduces the composite with weights chosen without external validation against clinical outcomes or human judgment.
  • domain assumption Radiologists triangulate findings across orthogonal planes, so multi-view input is the correct approximation of clinical reading.
    Section 2.2 states this clinical premise and builds the benchmark around it.

how reviews work

0 comments
Cite this review

Pith. "Pith review of How Good are Foundation Models in Longitudinal MRI Disease Progression Reasoning?." pith.science (2026). https://pith.science/paper/2R32TRKD

@misc{pith2026260813309,
  author       = {Pith},
  title        = {Pith review of: How Good are Foundation Models in Longitudinal MRI Disease Progression Reasoning?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2R32TRKD}},
  note         = {Machine review of arXiv:2608.13309}
}
read the original abstract

Magnetic Resonance Imaging (MRI) interpretation is fundamental to clinical decision-making, requiring radiologists to integrate multi-view anatomical planes across sequential timepoints while precisely localizing interval changes. However, existing vision-language benchmarks remain confined to single-timepoint, single-view interpretation, failing to capture the temporal-spatial reasoning essential to radiologic practice. We introduce the Time-Aware Multi-View MRI Benchmark, an evaluation framework unifying multi-view anatomical input, temporal reasoning across longitudinal scans, and structured localization guidance. The benchmark comprises 3,920 expert-verified question-answer pairs derived from 890 patients across over 3,200 longitudinal MRI timepoints, drawn from seven clinical cohorts covering glioblastoma, neurodegeneration, vestibular schwannoma, and brain metastases, in open-ended, multiple-choice, and binary formats, requiring models to identify anatomical regions of maximal change, characterize progression across sequences and views, and provide structured guidance specifying boundaries, imaging features, and confounders. Experiments across 16 vision-language models reveal moderate temporal alignment but systematic failure on change direction recognition and volumetric quantification, while multi-view inputs improve spatial localization yet degrade temporal reasoning in compact architectures. Our benchmark provides a systematic framework for evaluating progression tracking, interval change localization, and temporal ordering, which are essential for clinical deployment. Code, evaluation splits, and the dataset are available at: https://github.com/wafaAlghallabi/Time-Aware-MRI.

Figures

Figures reproduced from arXiv: 2608.13309 by the authors.

Figure 1
Figure 1. Comparison of reasoning dimension coverage across MRI benchmarks. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Time-Aware Multi-View MRI Benchmark pipeline: data selection, multi [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Representative samples from the benchmark illustrating diverse patholo [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

32 extracted references · 20 canonical work pages

  1. [1]

    American College of Radiology: ACR practice parameter for communication of diagnostic imaging findings (2024)

    work page 2024

  2. [2]

    NeuroImage (2011)

    Avants, B.B., Tustison, N.J., Song, G., Cook, P.A., Klein, A., Gee, J.C.: A repro- ducible evaluation of ANTs similarity metric performance in brain image registra- tion. NeuroImage (2011)

    work page 2011

  3. [3]

    arXiv preprint arXiv:2406.04449 (2024)

    Bannur,S.,Bouzid,K.,Castro,D.C.,Schwaighofer,A.,Thieme,A.,etal.:MAIRA- 2: Grounded radiology report generation. arXiv preprint arXiv:2406.04449 (2024)

    arXiv 2024

  4. [4]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2023)

    Bannur, S., Hyland, S., Liu, Q., Pérez-García, F., et al.: Learning to exploit tem- poral structure for biomedical vision-language processing. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2023)

    work page 2023

  5. [5]

    arXiv preprint arXiv:2505.14064 (2025)

    Bercea, C.I., Li, J., Raffler, P., et al.: NOVA: A benchmark for anomaly localization and clinical reasoning in brain MRI. arXiv preprint arXiv:2505.14064 (2025)

    arXiv 2025

  6. [6]

    arXiv preprint arXiv:2406.06512 (2024)

    Blankemeier, L., Cohen, J.P., Kumar, A., Wentland, D., Moor, M., Reis, E., Ru- bin, N., Bluethgen, C., et al.: Merlin: A vision language foundation model for 3D computed tomography. arXiv preprint arXiv:2406.06512 (2024)

    arXiv 2024

  7. [7]

    Al Ghallabi et al

    Chadha, S., Weiss, D., Janas, A., Ramakrishnan, D., et al.: Yale-brain-mets- longitudinal: Yale longitudinal dataset of brain metastases on mri with associated clinical data (2025) 10 W. Al Ghallabi et al

    work page 2025

  8. [8]

    arXiv preprint arXiv:2408.03361 (2024)

    Chen, P., Ye, J., Wang, G., Li, Y., et al.: GMAI-MMBench: A comprehensive multimodal evaluation benchmark towards general medical AI. arXiv preprint arXiv:2408.03361 (2024)

    arXiv 2024

Show all 32 references
  1. [9]

    Radiology: Artificial Intelligence (2024)

    Fields, B.K.K., Calabrese, E., Mongan, J., Cha, S., et al.: The university of cali- fornia san francisco adult longitudinal post-treatment diffuse glioma MRI dataset. Radiology: Artificial Intelligence (2024)

    2024

  2. [10]

    Scientific Data (2026)

    Gagnon, L., Gupta, D., Nguyen, U., Correia de Verdier, M., Saluja, R., et al.: The university of california san diego post-treatment glioblastoma (UCSD-PTGBM) annotated multimodal MRI dataset. Scientific Data (2026)

    2026

  3. [11]

    Computer Vision and Image Understanding (2025)

    Ghouse, H., Behzad, M.: MOSAIC: A multi-view 2.5D organ slice selector with cross-attentional reasoning for anatomically-aware CT localization in medical or- gan segmentation. Computer Vision and Image Understanding (2025)

    2025

  4. [12]

    arXiv preprint arXiv:2508.17524 (2025)

    He,X.,Rofena,A.,Feng,R.,etal.:OmniMRI:Aunifiedvision-languagefoundation model for generalist MRI interpretation. arXiv preprint arXiv:2508.17524 (2025)

    2025 arXiv

  5. [13]

    PhysioNet (2025)

    Hu, X., Gu, L., An, Q., Zhang, M., Liu, L., et al.: Medical-Diff-VQA: A large-scale medical dataset for difference visual question answering on chest x-ray images. PhysioNet (2025)

    2025

  6. [14]

    arXiv preprint arXiv:2402.09181 (2024)

    Hu, Y., Li, T., Lu, Q., Shao, W., He, J., Qiao, Y., Luo, P.: OmniMedVQA: A new large-scale comprehensive evaluation benchmark for medical LVLM. arXiv preprint arXiv:2402.09181 (2024)

    2024 arXiv

  7. [15]

    Journal of Magnetic Res- onance Imaging (2008)

    JackJr.,C.R.,Bernstein,M.A.,Fox,N.C.,Thompson,P.M.,etal.:TheAlzheimer’s disease neuroimaging initiative (ADNI): MRI methods. Journal of Magnetic Res- onance Imaging (2008)

    2008

  8. [16]

    Kujawa, A., Dorent, R., Wijethilake, N., et al.: Segmentation of vestibular schwan- noma from magnetic resonance imaging: An annotated multi-center routine clinical dataset (Vestibular-Schwannoma-MC-RC) (2023)

    2023

  9. [17]

    Journal of Cognitive Neuroscience (2010)

    Marcus, D.S., Fotenos, A.F., Csernansky, J.G., Morris, J.C., Buckner, R.L.: Open access series of imaging studies (OASIS): Longitudinal MRI data in nondemented and demented older adults. Journal of Cognitive Neuroscience (2010)

    2010

  10. [18]

    arXiv preprint arXiv:2408.08704 (2025)

    Nan, Y., Zhou, H., Xing, X., et al.: Beyond the hype: A dispassionate look at vision- language models in medical scenario. arXiv preprint arXiv:2408.08704 (2025)

    2025 arXiv

  11. [19]

    arXiv preprint arXiv:2404.18416 (2024)

    Saab, K., Tu, T., Weng, W.H., Tanno, R., Stutz, D., Wulczyn, E., et al.: Capabil- ities of Gemini models in medicine. arXiv preprint arXiv:2404.18416 (2024)

    2024 arXiv

  12. [20]

    Scientific Data (2022)

    Suter, Y., Knecht, U., Valenzuela, W., Notter, M., et al.: The LUMIERE dataset: Longitudinal glioblastoma MRI with expert RANO evaluation. Scientific Data (2022)

    2022

  13. [21]

    arXiv preprint arXiv:2501.06186 (2025)

    Thawakar, O., Dissanayake, D., More, K., Thawkar, R., et al.: LlamaV-o1: Re- thinking step-by-step visual reasoning in LLMs. arXiv preprint arXiv:2501.06186 (2025)

    2025 arXiv

  14. [22]

    IEEE Transactions on Medical Imaging (2010)

    Tustison, N.J., Avants, B.B., Cook, P.A., et al.: N4ITK: Improved N3 bias correc- tion. IEEE Transactions on Medical Imaging (2010)

    2010

  15. [23]

    In: Advances in Neural Information Processing Systems

    Wang, Y., Peng, J., Dai, Y., Jones, C., et al.: Enhancing vision-language models for medical imaging: Bridging the 3D gap with innovative slice selection. In: Advances in Neural Information Processing Systems. Curran Associates, Inc. (2024)

    2024

  16. [24]

    In: Medical Image Computing and Computer Assisted Intervention – MICCAI 2025

    Wolf, D., Hillenhagen, H., Taskin, B., Bäuerle, A., Beer, M., et al.: Your other left! vision-language models fail to identify relative positions in medical images. In: Medical Image Computing and Computer Assisted Intervention – MICCAI 2025. Springer (2025)

    2025

  17. [25]

    Nature Com- munications (2025) Time-Aware Multi-View MRI Benchmark 11

    Wu, C., Zhang, X., Zhang, Y., Wang, Y., Xie, W.: Towards generalist foundation model for radiology by leveraging web-scale 2D&3D medical data. Nature Com- munications (2025) Time-Aware Multi-View MRI Benchmark 11

    2025

  18. [26]

    IEEE Journal of Biomedical and Health Informatics (2025)

    Xin, Y., Ates, G.C., Gong, K., Shao, W.: Med3DVLM: An efficient vision-language model for 3D medical image analysis. IEEE Journal of Biomedical and Health Informatics (2025)

    2025

  19. [27]

    arXiv preprint arXiv:2508.10947 (2025)

    Xu, R., Huang, Z., Wei, Y., et al.: MedAtlas: Evaluating LLMs for multi-round, multi-task medical reasoning across diverse imaging modalities and clinical text. arXiv preprint arXiv:2508.10947 (2025)

    2025 arXiv

  20. [28]

    In: Interna- tional Conference on Medical Image Computing and Computer-Assisted Interven- tion

    Yang, X., Miao, J., Yuan, Y., Wang, J., Dou, Q., Li, J., Heng, P.A.: Med-MIM: Medical large vision language models with multi-image visual ability. In: Interna- tional Conference on Medical Image Computing and Computer-Assisted Interven- tion. Springer (2025)

    2025

  21. [29]

    arXiv preprint arXiv:2505.16964 (2025)

    Yu, S., Wang, H., Wu, J., Xie, C., Zhou, Y.: MedFrameQA: A multi-image medical VQA benchmark for clinical reasoning. arXiv preprint arXiv:2505.16964 (2025)

    2025

  22. [30]

    arXiv preprint arXiv:2509.25143 (2025)

    Zhang, J., Gu, J.C., Hu, W., Zhou, Y., Piramuthu, R., Peng, N.: TemMed-Bench: Evaluating temporal medical image reasoning in vision-language models. arXiv preprint arXiv:2509.25143 (2025)

    2025 arXiv

  23. [31]

    arXiv preprint arXiv:2305.10415 (2023)

    Zhang, X., Wu, C., Zhao, Z., Lin, W., Zhang, Y., et al.: PMC-VQA: Visual instruc- tion tuning for medical visual question answering. arXiv preprint arXiv:2305.10415 (2023)

    2023 arXiv

  24. [32]

    Zhou, T., Xu, Y., Zhu, Y., et al.: DrVD-Bench: Do vision-language models reason like human doctors in medical image diagnosis? arXiv preprint arXiv:2505.24173 (2025)

    2025 arXiv

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.