arXiv cs.AI· Vikash Singh, Debargha Ganguly, Haotian Yu, Chengwei Zhou, Prerna Singh, Brandon Lee, Vipin Chaudhary, Gourav Datta·· 3 天前AI 评分30
面向视觉语言模型临床推理保证的神经符号形式化验证框架
Toward Guarantees for Clinical Reasoning in Vision Language Models via Formal Verification
AI 导读
研究者提出一种神经符号验证框架，将 VLM 生成的放射学报告自由文本自动形式化为结构化命题证据，并用 SMT 求解器 Z3 结合临床知识库判定每条诊断结论是被数学蕴含、幻觉还是遗漏。在五个胸部 X 光基准上评估七款 VLM，该验证器揭示了传统指标无法察觉的推理失败模式，如保守观察与随机幻觉。在标注数据集上，以求解器蕴含约束作为事后保证，可系统性消除无依据幻觉，显著提升诊断可靠性与精确率。
正文 · 原文
Abstract:Vision-language models (VLMs) show promise in drafting radiology reports, yet they frequently suffer from logical inconsistencies, generating diagnostic impressions unsupported by their own perceptual findings or missing logically entailed conclusions. Standard lexical metrics heavily penalize clinical paraphrasing and fail to capture these deductive failures in reference-free settings. Toward guarantees for clinical reasoning, we introduce a neurosymbolic verification framework that deterministically audits the internal consistency of VLM-generated reports. Our pipeline autoformalizes free-text radiographic findings into structured propositional evidence, utilizing an SMT solver (Z3) and a clinical knowledge base to verify whether each diagnostic claim is mathematically entailed, hallucinated, or omitted. Evaluating seven VLMs across five chest X-ray benchmarks, our verifier exposes distinct reasoning failure modes, such as conservative observation and stochastic hallucination, that remain invisible to traditional metrics. On labeled datasets, enforcing solver-backed entailment acts as a rigorous post-hoc guarantee, systematically eliminating unsupported hallucinations to significantly increase diagnostic soundness and precision in generative clinical assistants.
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Logic in Computer Science (cs.LO) |
| Cite as: | arXiv:2602.24111 [cs.CV] |
| (or arXiv:2602.24111v2 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2602.24111 arXiv-issued DOI via DataCite |
Submission history
From: Vikash Singh [view email]
[v1]
Fri, 27 Feb 2026 15:49:59 UTC (1,414 KB)
[v2]
Tue, 13 Oct 2026 14:15:15 UTC (1,641 KB)
来源：arXiv cs.AI · arxiv.org