# PSF-Med: Measuring and Explaining Paraphrase Sensitivity in Medical Vision Language Models

arXiv: https://arxiv.org/abs/2602.21428 · HTML: https://mlsanity.com/paper/2602.21428/

## Key facts

- Cited by 4 papers in the map (data through 14 Oct 2026).
- Placed in Medical vision-language models, one of 16 directions in Multimodal LLMs / vision-language models.
- Builds on 27 of its 38 references that are in the map.
- First posted on arXiv 24 Feb 2026.

## Placed in

- 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)
- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [Locating and Editing Factual Associations in GPT](https://mlsanity.com/paper/2202.05262/) (2022)
- [Making the V in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering](https://mlsanity.com/paper/1612.00837/) (2016)
- [Sparse Autoencoders Find Highly Interpretable Features in Language Models](https://mlsanity.com/paper/2309.08600/) (2023)
- [Refusal in Language Models Is Mediated by a Single Direction](https://mlsanity.com/paper/2406.11717/) (2024)
- [Beyond Accuracy: Behavioral Testing of NLP models with CheckList](https://mlsanity.com/paper/2005.04118/) (2020)
- [Publicly Available Clinical BERT Embeddings](https://mlsanity.com/paper/1904.03323/) (2019)
- [Attention is not Explanation](https://mlsanity.com/paper/1902.10186/) (2019)
- [Contrastive Learning of Medical Visual Representations from Paired Images and Text](https://mlsanity.com/paper/2010.00747/) (2020)
- [Attention is not not Explanation](https://mlsanity.com/paper/1908.04626/) (2019)
- [SLAKE: A Semantically-Labeled Knowledge-Enhanced Dataset for Medical Visual Question Answering](https://mlsanity.com/paper/2102.09542/) (2021)
- [Causal Abstractions of Neural Networks](https://mlsanity.com/paper/2106.02997/) (2021)
- [Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2](https://mlsanity.com/paper/2408.05147/) (2024)
- [Don't Just Assume; Look and Answer: Overcoming Priors for Visual Question Answering](https://mlsanity.com/paper/1712.00377/) (2017)
- [Making the Most of Text Semantics to Improve Biomedical Vision--Language Processing](https://mlsanity.com/paper/2204.09817/) (2022)
- [PadChest: A large chest x-ray image dataset with multi-label annotated reports](https://mlsanity.com/paper/1901.07441/) (2019)
- [Adversarial Example Generation with Syntactically Controlled Paraphrase Networks](https://mlsanity.com/paper/1804.06059/) (2018)
- [VinDr-CXR: An open dataset of chest X-rays with radiologist's annotations](https://mlsanity.com/paper/2012.15029/) (2020)
- [Causal Mediation Analysis for Interpreting Neural NLP: The Case of Gender Bias](https://mlsanity.com/paper/2004.12265/) (2020)
- [OmniMedVQA: A New Large-Scale Comprehensive Evaluation Benchmark for Medical LVLM](https://mlsanity.com/paper/2402.09181/) (2024)
- [Med-HALT: Medical Domain Hallucination Test for Large Language Models](https://mlsanity.com/paper/2307.15343/) (2023)
- [ProSA: Assessing and Understanding the Prompt Sensitivity of LLMs](https://mlsanity.com/paper/2410.12405/) (2024)
- [Cycle-Consistency for Robust Visual Question Answering](https://mlsanity.com/paper/1902.05660/) (2019)
- [VHELM: A Holistic Evaluation of Vision Language Models](https://mlsanity.com/paper/2410.07112/) (2024)
- [Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models](https://mlsanity.com/paper/2601.14004/) (2026)
- [Vision-Language Introspection: Mitigating Overconfident Hallucinations in MLLMs via Interpretable Bi-Causal Steering](https://mlsanity.com/paper/2601.05159/) (2026)

## Built on by (most cited first, 4 of 4)

- [Predictive Entropy as a Joint Screen for Error and Paraphrase Instability in Medical Vision-Language Models](https://mlsanity.com/paper/2604.08941/) (2026)
- [Attention Without Grounding: Causal Evaluation of Visual Explanations in Medical VLMs](https://mlsanity.com/paper/2607.18577/) (2026)
- [Lightweight Stylistic Consistency Profiling: Robust Detection of LLM-Generated Textual Content for Multimedia Moderation](https://mlsanity.com/paper/2605.05950/) (2026)
- [Reliability Stress Tests and Decision-Time Routing for Chest X-ray Vision-Language Models](https://mlsanity.com/paper/2610.02270/) (2026)

## Abstract (arXiv)

Medical Vision Language Models (VLMs) can change their answers when clinicians rephrase the same question, a failure mode that threatens deployment safety. We introduce PSF-Med, a benchmark of 26,850 chest X-ray questions paired with 92,856 meaning-preserving paraphrases across MIMIC-CXR, PadChest, and VinDr-CXR, spanning clinical populations in the US, Spain, and Vietnam. Every paraphrase is validated by an LLM judge using a bidirectional clinical entailment rubric, with 91.6% cross-family agreement. Across nine VLMs, including general-purpose models, we find flip rates from 3% to 37%. However, low flip rate does not imply visual grounding: text-only baselines show that some models stay consistent even when the image is removed, suggesting they rely on language priors. To study mechanisms in one model, we apply GemmaScope 2 Sparse Autoencoders (SAEs) to MedGemma 4B and analyze FlipBank, a curated set of 158 flip cases. We identify a sparse feature at layer 17 that correlates with prompt framing and predicts decision margin shifts. In causal patching, removing this feature's contribution recovers 45% of the yes-minus-no logit margin on average and fully reverses 15% of flips. Acting on this finding, we show that clamping the identified feature at inference reduces flip rates by 31% relative with only a 1.3 percentage-point accuracy cost, while also decreasing text-prior reliance. These results suggest that flip rate alone is not enough; robustness evaluations should test both paraphrase stability and image reliance.