arXiv cs.CL· Edric Castel C. Hao, Rez Samantha Z. Floresca·· 12 小时前AI 评分34
医疗 VLM 在渐进式证据缺失下的“行为海市蜃楼”：MedGemma 越过证据边界仍持续作答
Beyond the Evidence Boundary: Behavioral Mirage Under Progressive Degradation in Medical VLMs
AI 导读
研究提出基线锚定框架，追踪基线答对的病例在图像渐进退化下何时越过经验证据边界，并将边界处仍坚持作答（无论对错）定义为“海市蜃楼”。三个医疗数据集上的三款 VLM 中仅 MedGemma 基线正确样本量足够做轨迹分析，其边界后行为包括直接海市蜃楼、正确恢复、海市蜃楼消解、持续拒答与振荡，各数据集-模态组合的海市蜃楼率为 16.0% 至 100.0%。
正文 · 原文
Abstract:VLMs for medical image interpretation are typically evaluated on clean inputs, despite the progressive evidence loss that can occur in high stakes clinical settings. Existing robustness studies show that performance can degrade while confidence remains high, but do not establish when a previously reliable model-case pair ceases to behave reliably. We introduce a baseline-anchored framework that traces baseline-correct cases through progressive image degradation to identify an empirical evidence boundary, and define mirage as continued commitment at this boundary, regardless of correctness. Across three VLMs on three medical datasets, only MedGemma provides a sufficiently large baseline-correct pool for trajectory analysis. On VQA-RAD, the only dataset in which the question text varies, language-only answerability is associated with baseline correctness (AUROC 0.677) and with commitment at the boundary (AUROC 0.653 to 0.670). MedGemma's post-boundary behavior is heterogeneous, including direct mirage, correct recovery, mirage dissolution, consistent abstention, and oscillation, with mirage rates of 16.0 to 100.0% across dataset-modality pairs. Despite this behavioral heterogeneity, residual-stream probes decode commit-versus-abstain behavior at the boundary with AUROC of 0.895 to 0.994 under nested cross-validation. These results show that progressive evidence loss can produce persistent commitment that output entropy and abstention behavior do not consistently reflect, motivating boundary-aware evaluation, option-order controls, and improved abstention mechanisms for reliable medical foundation models operating under partial observation.
| Comments: | Accepted to WMHS Workshop @ NeurIPS 2026 |
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL) |
| ACM classes: | I.4.9; I.2.7; J.3 |
| Cite as: | arXiv:2610.28000 [cs.CV] |
| (or arXiv:2610.28000v1 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2610.28000 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Edric Castel Hao [view email]
[v1]
Tue, 20 Oct 2026 10:45:40 UTC (2,071 KB)
来源：arXiv cs.CL · arxiv.org