mlsanityData to 7 Oct 2026

MedVIGIL: Evaluating Trustworthy Medical VLMs Under Broken Visual Evidence

arXiv 2605.07919 ↗First posted 8 May 2026cs.CVcited by 0

Not on the map. Why?

Abstract (arXiv)

Medical vision--language models (VLMs) are usually evaluated on intact image--question pairs, but trustworthy clinical use requires a stronger property: a model must recognise when the evidential basis for an answer has failed. We study this through silent failures under perturbed evidence, where a vision-required medical question is paired with a false premise, wording perturbation, knowledge-only rewrite, or ROI-corrupted image, yet the model returns a fluent non-refusal answer. We introduce medvigil, a 300-case evaluation suite drawn from four public medical VQA sources, supervised end to end by four board-certified radiologists: every gold answer, refusal option, candidate-answer set, paraphrase, false-premise trap, ROI box, and clinical risk tier is clinician-authored. Two attending radiologists annotate every case in parallel, a senior radiologist consolidates the released manifest, and a separate fourth radiologist independent of construction answers every probe to provide the human reference baseline. The release contains 2556 MCQ probes, 240 counterfactual triplets, physician-adjudicated risk-tier and answerability flags, ROI boxes, and a paired open-ended variant. We report seven correctness-conditioned audit metrics that summarise into the medvigil Composite Score (MCS), and audit 16 vision-capable models plus two text-only baselines. The independent radiologist scores MCS 83.3 at silent-failure rate 5.8%, leaving a 14.1-point composite headroom above the strongest audited model (Claude Opus 4.7 at 69.2). The benchmark and evaluation harness are publicly released.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 34 of its 39 references that are in the map. First posted on arXiv 8 May 2026.

Builds on

34 of 39 references.

References · 34

  1. 2019Benchmarking Neural Network Robustness to Common Corruptions and Perturbationscited by 2,857
  2. 2016Making the V in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answeringcited by 2,498
  3. 2022Large Language Models Encode Clinical Knowledgecited by 1,807
  4. 2023Evaluating Object Hallucination in Large Vision-Language Modelscited by 1,696
  5. 2019CheXpert: A Large Chest Radiograph Dataset with Uncertainty Labels and Expert Comparisoncited by 1,410
  6. 2018Datasheets for Datasetscited by 1,371
  7. 2023LLaVA-Med: Training a Large Language-and-Vision Assistant for Biomedicine in One Daycited by 1,287
  8. 2017Selective Classification for Deep Neural Networkscited by 799
  9. 2018Object Hallucination in Image Captioningcited by 661
  10. 2021SLAKE: A Semantically-Labeled Knowledge-Enhanced Dataset for Medical Visual Question Answeringcited by 444
  11. 2023Med-Flamingo: a Multimodal Medical Few-shot Learnercited by 387
  12. 2021Domain Adaptation for Medical Image Analysis: A Surveycited by 351
22 more
  1. 2023PMC-VQA: Visual Instruction Tuning for Medical Visual Question Answeringcited by 333
  2. 2024HuatuoGPT-Vision, Towards Injecting Medical Visual Knowledge into Multimodal LLMs at Scalecited by 194
  3. 2022Data Cards: Purposeful and Transparent Dataset Documentation for Responsible AIcited by 169
  4. 2025AbstentionBench: Reasoning LLMs Fail on Unanswerable Questionscited by 129
  5. 2024Know Your Limits: A Survey of Abstention in Large Language Modelscited by 116
  6. 2019Towards Causal VQA: Revealing and Reducing Spurious Correlations by Invariant and Covariant Semantic Editingcited by 108
  7. 2019Cycle-Consistency for Robust Visual Question Answeringcited by 105
  8. 2024Croissant: A Metadata Format for ML-Ready Datasetscited by 79
  9. 2023Self-Evaluation Improves Selective Generation in Large Language Modelscited by 74
  10. 2024Detecting and Evaluating Medical Hallucinations in Large Vision Language Modelscited by 57
  11. 2026MIRAGE: The Illusion of Visual Understandingcited by 57
  12. 2020VQA-LOL: Visual Question Answering under the Lens of Logiccited by 48
  13. 2024VLind-Bench: Measuring Language Priors in Large Vision-Language Modelscited by 44
  14. 2026BabyVision: Visual Reasoning Beyond Languagecited by 39
  15. 2025A Novel Evaluation Benchmark for Medical LLMs: Illuminating Safety and Effectiveness in Clinical Domainscited by 28
  16. 2024MedVH: Towards Systematic Evaluation of Hallucination for Large Vision Language Models in the Medical Contextcited by 19
  17. 2025Localizing Before Answering: A Hallucination Evaluation Benchmark for Grounded Medical Multimodal LLMscited by 14
  18. 2025MedHEval: Benchmarking Hallucinations and Mitigation Strategies in Medical Large Vision-Language Modelscited by 10
  19. 2026Knowing When Not to Answer: Evaluating Abstention in Multimodal Reasoning Systemscited by 6
  20. 2025Judge Before Answer: Can MLLM Discern the False Premise in Question?cited by 2
  21. 2026Benchmarking Deflection and Hallucination in Large Vision-Language Modelscited by 1
  22. 2026Why Does Grounding Hurt Medical VQA? Benchmarking, Diagnosis, and Fine-Tuning of Vision-Language Modelscited by 1

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025Evaluating Reasoning Faithfulness in Medical Vision-Language Models using Multimodal Perturbations
  2. 2024MediConfusion: Can you trust your AI radiologist? Probing the reliability of multimodal medical foundation models
  3. 2024SURE-VQA: Systematic Understanding of Robustness Evaluation in Medical VQA Tasks
  4. 2026Toward Guarantees for Clinical Reasoning in Vision Language Models via Formal Verification
  5. 2026Calibrated Triage, Not Autonomy: Confidence Estimation for Medical Vision-Language Models
  6. 2026Verification Mirage: Mapping the Reliability Boundary of Self-Verification in Medical VQA
  7. 2025Benchmarking and Mitigating Sycophancy in Medical Vision Language Models
  8. 2025MedVLM-R1: Incentivizing Medical Reasoning Capability of Vision-Language Models (VLMs) via Reinforcement Learning