# Attention Without Grounding: Causal Evaluation of Visual Explanations in Medical VLMs

arXiv: https://arxiv.org/abs/2607.18577 · HTML: https://mlsanity.com/paper/2607.18577/

## Key facts

- Cited by 1 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 22 of its 26 references that are in the map.
- First posted on arXiv 20 Jul 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, 22 of 22)

- [Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization](https://mlsanity.com/paper/1610.02391/) (2016)
- [Axiomatic Attribution for Deep Networks](https://mlsanity.com/paper/1703.01365/) (2017)
- [Sanity Checks for Saliency Maps](https://mlsanity.com/paper/1810.03292/) (2018)
- [Publicly Available Clinical BERT Embeddings](https://mlsanity.com/paper/1904.03323/) (2019)
- [Attention is not Explanation](https://mlsanity.com/paper/1902.10186/) (2019)
- [RISE: Randomized Input Sampling for Explanation of Black-box Models](https://mlsanity.com/paper/1806.07421/) (2018)
- [Quantifying Attention Flow in Transformers](https://mlsanity.com/paper/2005.00928/) (2020)
- [Attention is not not Explanation](https://mlsanity.com/paper/1908.04626/) (2019)
- [Transformer Interpretability Beyond Attention Visualization](https://mlsanity.com/paper/2012.09838/) (2020)
- [MedGemma Technical Report](https://mlsanity.com/paper/2507.05201/) (2025)
- [Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2](https://mlsanity.com/paper/2408.05147/) (2024)
- [VinDr-CXR: An open dataset of chest X-rays with radiologist's annotations](https://mlsanity.com/paper/2012.15029/) (2020)
- [TorchXRayVision: A library of chest X-ray datasets and models](https://mlsanity.com/paper/2111.00595/) (2021)
- [CARES: A Comprehensive Benchmark of Trustworthiness in Medical Vision Language Models](https://mlsanity.com/paper/2406.06007/) (2024)
- [MedVH: Towards Systematic Evaluation of Hallucination for Large Vision Language Models in the Medical Context](https://mlsanity.com/paper/2407.02730/) (2024)
- [MIMO: A medical vision language model with visual referring multimodal input and pixel grounding multimodal output](https://mlsanity.com/paper/2510.10011/) (2025)
- [Seeing without Looking: Do Vision-Language Benchmarks Really Test Vision?](https://mlsanity.com/paper/2605.22903/) (2026)
- [PSF-Med: Measuring and Explaining Paraphrase Sensitivity in Medical Vision Language Models](https://mlsanity.com/paper/2602.21428/) (2026)
- [6 Fingers, 1 Kidney: Natural Adversarial Medical Images Reveal Critical Weaknesses of Vision-Language Models](https://mlsanity.com/paper/2512.04238/) (2025)
- [Mechanistically Guided LoRA Improves Paraphrase Consistency in Medical Vision-Language Models](https://mlsanity.com/paper/2603.00148/) (2026)
- [To Agree or To Be Right? The Grounding-Sycophancy Tradeoff in Medical Vision-Language Models](https://mlsanity.com/paper/2603.22623/) (2026)
- [Predictive Entropy as a Joint Screen for Error and Paraphrase Instability in Medical Vision-Language Models](https://mlsanity.com/paper/2604.08941/) (2026)

## Built on by (most cited first, 1 of 1)

- [Inherited Heads: Audio language models track speakers with their text backbone's attention, and an attention-mass ranking retrieves a different set](https://mlsanity.com/paper/2609.14174/) (2026)

## Abstract (arXiv)

Attention and saliency heatmaps are widely used to explain medical Vision-Language Model (VLM) outputs on chest X-rays, yet whether they truly highlight the image evidence driving predictions has not been causally tested. We audit faithfulness via overlap with radiologist bounding boxes on PadChest (n=637), attribution mass within radiologist masks on CheXlocalize (n=643), and 16x16 patch-occlusion maps that record which regions, when hidden, change the answer. We study three MedGemma-4B variants, cross-family probes on LLaVA-RAD and Qwen3-VL-8B-Instruct, and the specialist CheXagent-2-3b, with two CXR-trained classifiers (DenseNet121, ResNet50) as positive controls. A heatmap is faithful only if the model uses the image and attention concentrates on regions whose occlusion alters the prediction. No evaluated VLM meets both criteria. MedGemma and Qwen3-VL use the image, but attention anti-correlates with patch-occlusion importance (rho < 0 with 95% bootstrap CIs below zero). LLaVA-RAD's attention correlates positively, but the model is almost text-only (99.1% text-only agreement, near-zero causal mass), so correlation ties two near-zero signals. Attention also misses annotated anatomy: overlap with true regions never beats shifted or random controls, and no method places more than 22% of its mass inside radiologist masks. The two CXR classifiers pass all metrics, indicating the failure is specific to VLM heatmaps, not the evaluation. These heatmaps are visually reassuring but not faithful; clinical explanations require controlled localization metrics and causal perturbation, not visual inspection alone.