Auditing Frontier Vision-Language Models for Trustworthy Medical VQA: Grounding Failures, Format Collapse, and Domain Adaptation

• URL: http://arxiv.org/abs/2604.27720v1
• Date: Thu, 30 Apr 2026 11:11:47 GMT
• Title: Auditing Frontier Vision-Language Models for Trustworthy Medical VQA: Grounding Failures, Format Collapse, and Domain Adaptation
• Abstract summary: We audit five vision-language models (VLMs) on Medical VQA along two trust-relevant axes.<n>Models localize anatomical and pathological targets poorly.<n>A self-grounding pipeline, where the same model localizes then answers, degrades VQA accuracy.
• Score: 4.177028541321909
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Deploying vision-language models (VLMs) in clinical settings demands auditable behavior under realistic failure conditions, yet the failure landscape of frontier VLMs on specialized medical inputs is poorly characterized. We audit five recent frontier and grounding-aware VLMs (Gemini~2.5~Pro, GPT-5, o3, GLM-4.5V, Qwen~2.5~VL) on Medical VQA along two trust-relevant axes. Perception: all models localize anatomical and pathological targets poorly -- the best model reaches only 0.23 mean IoU and 19.1% Acc@0.5 -- and exhibit clinically dangerous laterality confusion. Pipeline integration: a self-grounding pipeline, where the same model localizes then answers, degrades VQA accuracy for every model -- driven by both inaccurate localization and format-compliance failures under the two-step prompt (parse failure rises to 70%--99% for Gemini and GPT-5 on VQA-RAD). Replacing predicted boxes with ground-truth annotations recovers and improves VQA accuracy, consistent with the failure residing in the perception module rather than in the decomposition itself. These observational findings identify grounding quality as a primary trustworthiness bottleneck in our SLAKE bounding-box setting. As a complementary fine-tuning follow-up, supervised fine-tuning of Qwen~2.5~VL on combined Med-VQA training data attains the highest reported SLAKE open-ended recall (85.5%) among comparable methods, suggesting that the VQA-level gap is tractable with domain adaptation; whether this also closes the perception/trustworthiness bottleneck is left to future work.

Related papers

• NovGauge: A Fine-Grained Benchmark for Diagnosing LLMs' Capability in Paper Novelty Assessment [54.4265627247104]
  Large language models (LLMs) are increasingly used in peer review at major AI conferences.<n>Existing benchmarks assess novelty as a single holistic score.<n>We present NovGauge, a human-anchored benchmark for fine-grained novelty assessment diagnosis.
  arXiv  Detail & Related papers  (2026-09-10T08:34:56Z)
• EVADE: Evidence-Verified Agentic Diagnosis with Escape [0.880899367147235]
  EVADE is an inferential, non-training method that enhances the safety of deploying a single frozen VLM.<n>It commits only when both the entire image and the zoomed view responses agree.<n>It reduces expected calibration error (ECE) by up to 45% compared to zero-shot.
  arXiv  Detail & Related papers  (2026-08-19T11:51:24Z)
• 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)
• A Controlled Audit of Pretraining Contamination in Public Medical Vision-Language Benchmarks [46.43161388634735]
  Medical vision-language models (VLMs) are evaluated on public benchmarks whose images and question-answer pairs have been freely downloadable for years.<n>We audit open VLMs on SLAKE-En, PathVQA, VQA-RAD, and an auxiliary public OmniMedVQA mirror using four detector families.
  arXiv  Detail & Related papers  (2026-06-08T18:40:10Z)
• Where, What, Why, and Importance: Structured Defect Grounding for Text-to-Image Feedback [51.08692072066352]
  Text-to-image (T2I) models still exhibit localized, subtle, and structurally complex failures.<n>Structured Defect Grounding casts T2I diagnosis as structured set prediction by modeling each defect as a location, type, reason, importance.<n>Our detector outperforms leading proprietary VLMs on structured defect grounding.
  arXiv  Detail & Related papers  (2026-06-04T13:03:16Z)
• MedFM-Robust: Benchmarking Robustness of Medical Foundation Models [17.68454119152175]
  We present a robustness benchmark comprising 40 perturbation types across eight imaging modalities.<n>Fine-tuning strategy dominates robustness, with LoRA exhibiting nearly double the degradation of full fine-tuning.<n>Medical-specific perturbations disproportionately damage segmentation, with 9 of 15 top corruptions being domain-specific.
  arXiv  Detail & Related papers  (2026-05-18T18:50:56Z)
• Predictive Entropy Links Calibration and Paraphrase Sensitivity in Medical Vision-Language Models [2.064612766965483]
  We show five uncertainty quantification methods on MedGemma 4BIT across in distribution MIMIC CXR and outof distribution PadChest chest X ray datasets, with cross architecture validation on LLaVA RAD7B.<n>For well calibrated single model methods, predictive entropy from one forward pass predicts which samples will flip under rephrasing AUROC 0.711 on MedGemma, 0.878 on LLaVARAD p 10 4, enabling a single entropy threshold to flag both unreliable and sensitive predictions.
  arXiv  Detail & Related papers  (2026-04-10T04:18:47Z)
• Edge Reliability Gap in Vision-Language Models: Quantifying Failure Modes of Compressed VLMs Under Visual Corruption [0.0]
  The rapid compression of large vision-language models for edge deployment raises an underexplored question: do compact models fail differently, not merely more often?<n>This study compares a 7-billion- parameter quantised VLM (Qwen2.5-VL-7B, 4-bit NF4) against a 500-million- parameter FP16 model (SmolVLM2-500M) across 4,000 samples from VQAv2 and COCO Captions.<n>A three-category error taxonomy (Object Blindness, Semantic Drift, Prior Bias) is applied as a diagnostic framework. Confidence calibration is measured via Expected Error (ECE) using mean token probability
  arXiv  Detail & Related papers  (2026-03-24T10:14:40Z)
• Beyond Human Performance: A Vision-Language Multi-Agent Approach for Quality Control in Pharmaceutical Manufacturing [1.5362004217750165]
  Colony-forming unit (CFU) detection is critical in pharmaceutical manufacturing.<n>To address this, we developed a multi-agent framework combining deep learning (DL) with vision-language models (VLMs)<n>Initial DL-based automation reduced human verification by 50 percent across vaccine manufacturing sites.
  arXiv  Detail & Related papers  (2026-02-24T04:48:05Z)
• Toward More Reliable Artificial Intelligence: Reducing Hallucinations in Vision-Language Models [0.0]
  Vision-language models (VLMs) frequently generate hallucinated content plausible but incorrect claims about image content.<n>We propose a training-free self-correction framework enabling VLMs to iteratively refine responses through uncertainty-guided visual re-attention.
  arXiv  Detail & Related papers  (2025-12-08T13:58:46Z)
• MedGRPO: Multi-Task Reinforcement Learning for Heterogeneous Medical Video Understanding [47.843626983298726]
  We introduce textbfMedVidBench, a large-scale benchmark of 531,850 video-instruction pairs across 8 medical sources spanning video, segment, and frame-level tasks.<n>While supervised fine-tuning on MedVidBench yields noticeable gains, standard Reinforcement Learning fails due to imbalanced reward scales across datasets.<n>We introduce textbfMedGRPO, a novel RL framework for balanced multi-dataset training with two key innovations.
  arXiv  Detail & Related papers  (2025-12-06T22:27:59Z)
• When to Trust the Answer: Question-Aligned Semantic Nearest Neighbor Entropy for Safer Surgical VQA [11.804829233549704]
  Safety and reliability are essential for deploying Visual Question Answering (VQA) in surgery.<n>We study uncertainty estimation as a key enabler of safer decision making.<n>We introduce Question Aligned Semantic Nearest Neighbor Entropy (QA-SNNE), a black box uncertainty estimator.
  arXiv  Detail & Related papers  (2025-11-03T11:18:21Z)
• Token-Level Inference-Time Alignment for Vision-Language Models [58.41370989069588]
  Vision-Language Models (VLMs) have become essential backbones of modern multimodal intelligence.<n>We present TITA, a lightweight framework that freezes the base VLM and instead trains a reward model to approximate its distribution.<n>During inference, implicit preference signals are extracted as log-probability ratios between the reward model and the target VLM, yielding dense autoregressive feedback.
  arXiv  Detail & Related papers  (2025-10-20T09:58:03Z)
• Continual Action Quality Assessment via Adaptive Manifold-Aligned Graph Regularization [53.82400605816587]
  Action Quality Assessment (AQA) quantifies human actions in videos, supporting applications in sports scoring, rehabilitation, and skill evaluation.<n>A major challenge lies in the non-stationary nature of quality distributions in real-world scenarios.<n>We introduce Continual AQA (CAQA), which equips with Continual Learning capabilities to handle evolving distributions.
  arXiv  Detail & Related papers  (2025-10-08T10:09:47Z)
• MedOmni-45°: A Safety-Performance Benchmark for Reasoning-Oriented LLMs in Medicine [69.08855631283829]
  We introduce Med Omni-45 Degrees, a benchmark designed to quantify safety-performance trade-offs under manipulative hint conditions.<n>It contains 1,804 reasoning-focused medical questions across six specialties and three task types, including 500 from MedMCQA.<n>Results show a consistent safety-performance trade-off, with no model surpassing the diagonal.
  arXiv  Detail & Related papers  (2025-08-22T08:38:16Z)
• Crane: Context-Guided Prompt Learning and Attention Refinement for Zero-Shot Anomaly Detection [50.343419243749054]
  Anomaly detection is critical in fields such as medical diagnostics and industrial defect detection.<n> CLIP's coarse-grained image-text alignment limits localization and detection performance for fine-grained anomalies.<n>Crane improves the state-of-the-art ZSAD from 2% to 28%, at both image and pixel levels, while remaining competitive in inference speed.
  arXiv  Detail & Related papers  (2025-04-15T10:42:25Z)
• Does Acceleration Cause Hidden Instability in Vision Language Models? Uncovering Instance-Level Divergence Through a Large-Scale Empirical Study [44.170933007736984]
  Vision-Language Models (VLMs) are powerful yet computationally intensive for widespread practical deployments.<n>Current acceleration evaluations primarily target minimal overall performance degradation, overlooking a crucial question: does the accelerated model still give the same answers to the same questions as it did before acceleration?<n>This is vital for stability-centered industrial applications where consistently correct answers for specific, known situations are paramount, such as in AI-based disease diagnosis.
  arXiv  Detail & Related papers  (2025-03-09T22:16:48Z)
• Generalized Focal Loss V2: Learning Reliable Localization Quality Estimation for Dense Object Detection [78.11775981796367]
  GFLV2 (ResNet-101) achieves 46.2 AP at 14.6 FPS, surpassing the previous state-of-the-art ATSS baseline (43.6 AP at 14.6 FPS) by absolute 2.6 AP on COCO tt test-dev. Code will be available at https://github.com/implus/GFocalV2.
  arXiv  Detail & Related papers  (2020-11-25T17:06:37Z)

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.