# Evaluating Reasoning Faithfulness in Medical Vision-Language Models using Multimodal Perturbations

arXiv: https://arxiv.org/abs/2510.11196 · HTML: https://mlsanity.com/paper/2510.11196/

## Key facts

- Cited by 9 papers in the map (data through 14 Oct 2026).
- Placed in Clinical and biomedical applications, one of 17 directions in LLM reasoning.
- Builds on 27 of its 32 references that are in the map.
- First posted on arXiv 13 Oct 2025.

## Placed in

- LLM reasoning / [Clinical and biomedical applications](https://mlsanity.com/lr/clinical-and-biomedical-applications/)
- Multimodal LLMs / vision-language models / [Medical vision-language models](https://mlsanity.com/mllm/medical-vision-language-models/)

## Builds on (most cited first, 27 of 27)

- [GPT-4 Technical Report](https://mlsanity.com/paper/2303.08774/) (2023)
- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities](https://mlsanity.com/paper/2507.06261/) (2025)
- [gpt-oss-120b & gpt-oss-20b Model Card](https://mlsanity.com/paper/2508.10925/) (2025)
- [Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting](https://mlsanity.com/paper/2305.04388/) (2023)
- [MedGemma Technical Report](https://mlsanity.com/paper/2507.05201/) (2025)
- [Reasoning Models Don't Always Say What They Think](https://mlsanity.com/paper/2505.05410/) (2025)
- [MedXpertQA: Benchmarking Expert-Level Medical Reasoning and Understanding](https://mlsanity.com/paper/2501.18362/) (2025)
- [LlamaV-o1: Rethinking Step-by-step Visual Reasoning in LLMs](https://mlsanity.com/paper/2501.06186/) (2025)
- [HealthGPT: A Medical Large Vision-Language Model for Unifying Comprehension and Generation via Heterogeneous Knowledge Adaptation](https://mlsanity.com/paper/2502.09838/) (2025)
- [Making Reasoning Matter: Measuring and Improving Faithfulness of Chain-of-Thought Reasoning](https://mlsanity.com/paper/2402.13950/) (2024)
- [Benchmarking Large Language Models on Answering and Explaining Challenging Medical Questions](https://mlsanity.com/paper/2402.18060/) (2024)
- [More Thinking, Less Seeing? Assessing Amplified Hallucination in Multimodal Reasoning Models](https://mlsanity.com/paper/2505.21523/) (2025)
- [Diagnostic Reasoning Prompts Reveal the Potential for Large Language Model Interpretability in Medicine](https://mlsanity.com/paper/2308.06834/) (2023)
- [Are DeepSeek R1 And Other Reasoning Models More Faithful?](https://mlsanity.com/paper/2501.08156/) (2025)
- [Walk the Talk? Measuring the Faithfulness of Large Language Model Explanations](https://mlsanity.com/paper/2504.14150/) (2025)
- [MedCaseReasoning: Evaluating and learning diagnostic reasoning from clinical case reports](https://mlsanity.com/paper/2505.11733/) (2025)
- [MedFrameQA: A Multi-Image Medical VQA Benchmark for Clinical Reasoning](https://mlsanity.com/paper/2505.16964/) (2025)
- [ChestX-Reasoner: Advancing Radiology Foundation Models with Reasoning through Step-by-Step Verification](https://mlsanity.com/paper/2504.20930/) (2025)
- [Quantifying the Reasoning Abilities of LLMs on Real-world Clinical Cases](https://mlsanity.com/paper/2503.04691/) (2025)
- [Automated Structured Radiology Report Generation](https://mlsanity.com/paper/2505.24223/) (2025)
- [A Closer Look at Bias and Chain-of-Thought Faithfulness of Large (Vision) Language Models](https://mlsanity.com/paper/2505.23945/) (2025)
- [VFaith: Do Large Multimodal Models Really Reason on Seen Images Rather than Previous Memories?](https://mlsanity.com/paper/2506.11571/) (2025)
- [PRobELM: Plausibility Ranking Evaluation for Language Models](https://mlsanity.com/paper/2404.03818/) (2024)
- [Susceptibility of Large Language Models to User-Driven Factors in Medical Queries](https://mlsanity.com/paper/2503.22746/) (2025)
- [Medical Reasoning in LLMs: An In-Depth Analysis of DeepSeek R1](https://mlsanity.com/paper/2504.00016/) (2025)

## Built on by (most cited first, 9 of 9)

- [MedGround: Bridging the Evidence Gap in Medical Vision-Language Models with Verified Grounding Data](https://mlsanity.com/paper/2601.06847/) (2026)
- [MedOpenClaw and MedFlowBench: Auditing Medical Agents in Full-Study Workflows](https://mlsanity.com/paper/2603.24649/) (2026)
- [Position, Not Provenance: Separating Reasoning Mediation from Sycophancy in Medical Vision-Language Models](https://mlsanity.com/paper/2607.27304/) (2026)
- [Be Faithful When Response: Returning Fluent and Grounded Answers for Vision-Language Models Reinforcement Learning](https://mlsanity.com/paper/2606.29984/) (2026)
- [Marking the Wrong Symptoms: Evaluating LLM Watermarks in Medical Texts](https://mlsanity.com/paper/2607.20462/) (2026)
- [Evaluating the Diagnostic Robustness of Vision-Language Models Under Visual and Textual Perturbations](https://mlsanity.com/paper/2608.04885/) (2026)
- [The Plan, Not the Decoder: Diagnosing and Repairing Compositional Failure in Reasoning-Augmented Text-to-Image Generation](https://mlsanity.com/paper/2608.21713/) (2026)
- [Beyond the Verdict: Evidence-Aligned Evaluation of Visual Prompt-Injection Guardrails](https://mlsanity.com/paper/2609.05535/) (2026)
- [One Perturbation Is Not Enough: Identifiability and Blind Baselines for Behavioral AI Evaluation](https://mlsanity.com/paper/2609.06190/) (2026)

## Abstract (arXiv)

Vision-language models (VLMs) often produce chain-of-thought (CoT) explanations that sound plausible yet fail to reflect the underlying decision process, undermining trust in high-stakes clinical use. Existing evaluations rarely catch this misalignment, prioritizing answer accuracy or adherence to formats. We present a clinically grounded framework for chest X-ray visual question answering (VQA) that probes CoT faithfulness via controlled text and image modifications across three axes: clinical fidelity, causal attribution, and confidence calibration. In a reader study (n=4), evaluator-radiologist correlations fall within the observed inter-radiologist range for all axes, with strong alignment for attribution (Kendall's τb=0.670), moderate alignment for fidelity (τb=0.387), and weak alignment for confidence tone (τb=0.091), which we report with caution. Benchmarking six VLMs shows that answer accuracy and explanation quality can be decoupled, acknowledging injected cues does not ensure grounding, and text cues shift explanations more than visual cues. While some open-source models match final answer accuracy, proprietary models score higher on attribution (25.0% vs. 1.4%) and often on fidelity (36.1% vs. 31.7%), highlighting deployment risks and the need to evaluate beyond final answer accuracy.