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

2026-04-30

AI Summary

This study addresses critical trustworthiness challenges in state-of-the-art vision-language models for medical visual question answering (VQA), including inaccurate object localization, left-right confusion, and failures in parsing prompt formats. Through a systematic audit of multiple models’ perceptual and pipeline integration capabilities, the work reveals— for the first time—that self-localization-to-QA pipelines suffer significant performance degradation due to localization errors and prompt collapse. It further demonstrates that using ground-truth bounding boxes substantially recovers accuracy, identifying the perception module as the key bottleneck. Evaluations via bounding box metrics (IoU, Acc@0.5) and prompt analysis show that even the best model achieves only a mean IoU of 0.23 and 19.1% localization accuracy. Supervised fine-tuning of Qwen2.5-VL yields an 85.5% open-domain recall on SLAKE, establishing a new state-of-the-art.

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.

Problem

• Medical VQA
• Vision-Language Models
• Grounding Failures
• Trustworthiness
• Domain Adaptation

Innovation

• vision-language models
• medical VQA
• grounding failure
• format collapse
• domain adaptation