[论文] Beyond the Leaderboard: Design Lessons for Trustworthy Multimodal...

论文概要

研究领域: NLP 作者: Sushant Gautam, Vajira Thambawita, Michael A. Riegler 发布时间: 2025-07-16 arXiv: 2507.12494

中文摘要

医疗多模态AI必须在结合视觉与文本证据的同时保持可靠性和可解释性。本文以 MediaEval Medico 2025 为回顾性GI内镜案例研究，分析了九个已记录系统在问答和解释质量上的设计选择。预训练骨干网络的参数高效适配在挑战赛中表现强劲，但答案层面的提升并未一致地转化为忠实且完整的临床推理。强制执行结构化推理和显式定位的方法在不同类型问题中表现出更可靠的行为，尽管证据是相关性的而非基于消融实验的。这些结果推动了超越词汇重叠的评估、标准化证据关联解释、防泄露数据治理，以及轻量级鲁棒性和校准检验。研究结论支持基于数据融合、可解释性和弹性评估的可信多模态医疗AI。

原文摘要

Healthcare multimodal AI must combine visual and textual evidence while remaining reliable and interpretable. Using MediaEval Medico 2025 as a retrospective GI endoscopy case study, we analyze design choices across nine documented systems for question answering and explanation quality. Parameter-efficient adaptation of pretrained backbones provides strong challenge performance, but answer-level gains do not consistently translate into faithful and complete clinical reasoning. Methods enforcing structured reasoning and explicit grounding show more reliable behavior across heterogeneous question types, although the evidence is correlational rather than ablation-based. These results motivate evaluation beyond lexical overlap, standardized evidence-linked explanations, leakage-aware data governance, and lightweight robustness and calibration checks. The findings support trustworthy multimoda...

--- *自动采集于 2026-07-19*

#论文 #arXiv #NLP #小凯