Are Large Vision Language Models Truly Grounded in Medical Images? Evidence from Italian Clinical Visual Question Answering

• URL: http://arxiv.org/abs/2511.19220v2
• Date: Wed, 26 Nov 2025 20:23:20 GMT
• Title: Are Large Vision Language Models Truly Grounded in Medical Images? Evidence from Italian Clinical Visual Question Answering
• Abstract summary: We investigate whether large vision language models (VLMs) demonstrate genuine visual grounding when answering Italian medical questions.<n>We test four state-of-the-art models: Claude Sonnet 4.5, GPT-4o, GPT-5-mini, and Gemini 2.0 flash exp.
• Score: 27.308408027453012
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large vision language models (VLMs) have achieved impressive performance on medical visual question answering benchmarks, yet their reliance on visual information remains unclear. We investigate whether frontier VLMs demonstrate genuine visual grounding when answering Italian medical questions by testing four state-of-the-art models: Claude Sonnet 4.5, GPT-4o, GPT-5-mini, and Gemini 2.0 flash exp. Using 60 questions from the EuropeMedQA Italian dataset that explicitly require image interpretation, we substitute correct medical images with blank placeholders to test whether models truly integrate visual and textual information. Our results reveal striking variability in visual dependency: GPT-4o shows the strongest visual grounding with a 27.9pp accuracy drop (83.2% [74.6%, 91.7%] to 55.3% [44.1%, 66.6%]), while GPT-5-mini, Gemini, and Claude maintain high accuracy with modest drops of 8.5pp, 2.4pp, and 5.6pp respectively. Analysis of model-generated reasoning reveals confident explanations for fabricated visual interpretations across all models, suggesting varying degrees of reliance on textual shortcuts versus genuine visual analysis. These findings highlight critical differences in model robustness and the need for rigorous evaluation before clinical deployment.

Related papers

• Polish Medical Visual Question Answering: Vision-Language Models Underutilize Visual Evidence [2.779050345468656]
  The benchmark comprises image-containing questions spanning diverse medical specialties and visual domains.<n>We evaluate Polish-oriented, general-purpose open-weight, and commercial vision-language models.
  arXiv  Detail & Related papers  (2026-08-13T08:07:22Z)
• Do Medical Vision Language Models Actually See? A Counterfactual Grounding Framework and Hard-Negative Contrastive Training for Visually-Reliant Medical VLMs [6.775973739193119]
  Large vision language models (VLMs) report strong accuracy on medical question-answering, yet it remains unclear whether they reason from visual evidence or exploit textual shortcuts.<n>We introduce a counterfactual evaluation framework that decouples visual and textual contributions by substituting input images with controlled surrogates blank, pixel-shuffled, image-absent, and CLIP-retrieved hard negatives.
  arXiv  Detail & Related papers  (2026-07-04T00:06:35Z)
• Med-R2: An Adversarial Benchmark for Evidence-Grounded Reasoning in Medical VLMs [21.95646908343504]
  We introduce Med-R2 Bench, a hierarchical benchmark aligned with the clinical workflow to evaluate adversarial robustness with visual grounding.<n>We design stepwise QA tasks to assess whether reasoning chains are strictly grounded in visual evidence across the four clinical stages.
  arXiv  Detail & Related papers  (2026-05-23T09:50:19Z)
• JMed48k: A Multi-Profession Japanese Medical Licensing Benchmark for Vision-Language Model Evaluation [53.571125769505194]
  We introduce JMed48k, a Japanese healthcare licensing benchmark for evaluating vision-language models.<n>Built from official PDF materials released by the Japanese Ministry of Health, Labour and Welfare, JMed48k contains 48,862 exam questions and 20,142 images.<n>We derive JMed48k-Eval, a recent five-year evaluation subset with 12,484 scored questions, including 9,905 text-only questions and 2,579 questions with images.
  arXiv  Detail & Related papers  (2026-05-21T07:20:38Z)
• Medical thinking with multiple images [44.04557445622649]
  We introduce MedThinkVQA, an expert-annotated benchmark for thinking with multiple images.<n>The dataset contains 8,067 cases, including 720 test cases, with an average of 6.62 images per case.<n>On the test set, the best closed-source models, Claude-4.6-Opus, Gemini-3-Pro, and GPT-5.2-xhigh, reach only 57.2%, 55.3%, and 54.9% accuracy.
  arXiv  Detail & Related papers  (2026-04-14T18:51:07Z)
• DeepGI: Explainable Deep Learning for Gastrointestinal Image Classification [0.0]
  The study confronts common endoscopic challenges such as variable lighting, fluctuating camera angles, and frequent imaging artifacts.<n>The best performing models, VGG16 and MobileNetV2, each achieved a test accuracy of 96.5%.<n>The approach includes explainable AI via Grad-CAM visualization, enabling identification of image regions most influential to model predictions.
  arXiv  Detail & Related papers  (2025-11-26T22:35:57Z)
• Evaluating Reasoning Faithfulness in Medical Vision-Language Models using Multimodal Perturbations [19.488236277427358]
  Vision-language models (VLMs) often produce chain-of-thought (CoT) explanations that sound plausible yet fail to reflect the underlying decision process.<n>We present a clinically grounded framework for chest X-ray visual question answering (VQA) that probes CoT faithfulness via controlled text and image modifications.
  arXiv  Detail & Related papers  (2025-10-13T09:28:22Z)
• EchoBench: Benchmarking Sycophancy in Medical Large Vision-Language Models [82.43729208063468]
  Recent benchmarks for medical Large Vision-Language Models (LVLMs) emphasize leaderboard accuracy, overlooking reliability and safety.<n>We study sycophancy -- models' tendency to uncritically echo user-provided information.<n>We introduce EchoBench, a benchmark to systematically evaluate sycophancy in medical LVLMs.
  arXiv  Detail & Related papers  (2025-09-24T14:09:55Z)
• COREVQA: A Crowd Observation and Reasoning Entailment Visual Question Answering Benchmark [3.5018278981067685]
  COREVQA (Crowd Observations and Reasoning Entailment) is a benchmark of 5608 image and synthetically generated true/false statement pairs.<n>Our results show that even the top-performing VLMs achieve accuracy below 80%, with other models performing substantially worse.
  arXiv  Detail & Related papers  (2025-07-17T04:47:47Z)
• An Early Investigation into the Utility of Multimodal Large Language Models in Medical Imaging [0.3029213689620348]
  We explore the potential of the Gemini (textitgemini-1.0-pro-vision-latest) and GPT-4V models for medical image analysis. Both Gemini AI and GPT-4V are first used to classify real versus synthetic images, followed by an interpretation and analysis of the input images. Our early investigation presented in this work provides insights into the potential of MLLMs to assist with the classification and interpretation of retinal fundoscopy and lung X-ray images.
  arXiv  Detail & Related papers  (2024-06-02T08:29:23Z)
• Language Models Meet Anomaly Detection for Better Interpretability and Generalizability [13.674931689496225]
  This research explores the integration of language models and unsupervised anomaly detection in medical imaging. We introduce a new dataset for multi-image visual question-answering on brain magnetic resonance images. Our model achieves a 60.81% accuracy on closed questions, covering disease classification and severity across 15 different classes.
  arXiv  Detail & Related papers  (2024-04-11T10:16:44Z)
• Gemini vs GPT-4V: A Preliminary Comparison and Combination of Vision-Language Models Through Qualitative Cases [98.35348038111508]
  This paper presents an in-depth comparative study of two pioneering models: Google's Gemini and OpenAI's GPT-4V(ision) The core of our analysis delves into the distinct visual comprehension abilities of each model. Our findings illuminate the unique strengths and niches of both models.
  arXiv  Detail & Related papers  (2023-12-22T18:59:58Z)
• Multimodal Foundation Models Exploit Text to Make Medical Image Predictions [3.4230952713864373]
  We evaluate the mechanisms by which multimodal foundation models integrate and prioritize different data modalities, including images and text. Our results suggest that multimodal AI models may be useful in medical diagnostic reasoning but that their accuracy is largely driven, for better and worse, by their exploitation of text.
  arXiv  Detail & Related papers  (2023-11-09T18:48:02Z)
• Holistic Analysis of Hallucination in GPT-4V(ision): Bias and Interference Challenges [54.42256219010956]
  This benchmark is designed to evaluate and shed light on the two common types of hallucinations in visual language models: bias and interference. bias refers to the model's tendency to hallucinate certain types of responses, possibly due to imbalance in its training data. interference pertains to scenarios where the judgment of GPT-4V(ision) can be disrupted due to how the text prompt is phrased or how the input image is presented.
  arXiv  Detail & Related papers  (2023-11-06T17:26:59Z)
• HallusionBench: An Advanced Diagnostic Suite for Entangled Language Hallucination and Visual Illusion in Large Vision-Language Models [69.52245481329899]
  We introduce HallusionBench, a benchmark for the evaluation of image-context reasoning. The benchmark comprises 346 images paired with 1129 questions, all meticulously crafted by human experts. In our evaluation on HallusionBench, we benchmarked 15 different models, highlighting a 31.42% question-pair accuracy achieved by the state-of-the-art GPT-4V.
  arXiv  Detail & Related papers  (2023-10-23T04:49:09Z)
• Robust and Interpretable Medical Image Classifiers via Concept Bottleneck Models [49.95603725998561]
  We propose a new paradigm to build robust and interpretable medical image classifiers with natural language concepts. Specifically, we first query clinical concepts from GPT-4, then transform latent image features into explicit concepts with a vision-language model.
  arXiv  Detail & Related papers  (2023-10-04T21:57:09Z)
• Learning to Decompose Visual Features with Latent Textual Prompts [140.2117637223449]
  We propose Decomposed Feature Prompting (DeFo) to improve vision-language models. Our empirical study shows DeFo's significance in improving the vision-language models.
  arXiv  Detail & Related papers  (2022-10-09T15:40:13Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.