Beyond Counts: A Distributional Robustness Margin For Pathology Foundation Models

📄 arXiv: 2607.25497v1 📥 PDF

作者: Clément Grisi, Jeroen van der Laak, Geert Litjens

分类: cs.CV, cs.AI

发布日期: 2026-07-28

备注: Preprint

💡 一句话要点

提出CRoMa以解决病理基础模型的稳健性问题

🎯 匹配领域: 支柱九：具身大模型 (Embodied Foundation Models)

关键词: 病理学 稳健性评估 机器学习 模型选择 生物匹配 数据异质性

📋 核心要点

  1. 现有的稳健性指数（RI）在评估病理模型时忽略了样本级异质性，导致稳健性评估不够全面。
  2. 本文提出的CRoMa通过比较生物匹配和生物干扰的距离，提供了一种样本解析的稳健性度量，克服了RI的局限性。
  3. 实验结果显示，CRoMa在不同数据集上的排名一致性较高，并且与模型的适应性表现有显著关联。

📝 摘要（中文）

病理基础模型在临床应用中逐渐成熟，但仍然容易受到中心间系统性非生物变异的影响。现有的稳健性指数（RI）虽然能够量化局部表示几何的生物学与非生物学变异，但其基于计数的形式忽略了距离信息。本文提出了交叉混淆稳健性边际（CRoMa），作为一种样本解析的度量，直接比较生物匹配与生物干扰的距离，从而将稳健性重新定义为一个分布而非单一得分。通过对20个切片级编码器和4个幻灯片级编码器的评估，发现CRoMa在不同数据集上的排名一致性较高，并揭示了模型内部的显著异质性。

🔬 方法详解

问题定义：本文旨在解决病理基础模型在不同中心间由于非生物变异导致的稳健性不足问题。现有的稳健性指数（RI）由于其固定邻域设计，无法充分反映样本级的异质性，导致评估结果的片面性。

核心思路：提出CRoMa作为一种新的稳健性度量，通过样本解析的方式直接比较生物匹配与生物干扰的距离，从而将稳健性视为一个分布而非单一得分。这种设计能够更好地捕捉模型在不同样本上的表现差异。

技术框架：CRoMa的整体架构包括样本距离计算、交叉混淆比较和分布分析三个主要模块。首先计算样本间的距离，然后对生物匹配和生物干扰进行比较，最后分析稳健性分布。

关键创新：CRoMa的最大创新在于其样本解析的稳健性度量方法，能够揭示模型在不同样本上的表现差异，克服了RI的固定邻域设计带来的局限性。

关键设计：在CRoMa的实现中，关键参数包括样本选择策略和距离计算方法。通过对不同模型的评估，发现每个切片编码器在低尾部的混淆主导特征表现出显著差异。

🖼️ 关键图片

fig_0
fig_1
fig_2

📊 实验亮点

实验结果表明，CRoMa在20个切片级编码器和4个幻灯片级编码器的评估中，排名一致性较高，且模型的低尾部稳健性与适应性表现之间存在显著关联。这一发现为模型选择提供了新的视角。

🎯 应用场景

该研究的潜在应用领域包括医学影像分析、病理学诊断和机器学习模型的选择与评估。通过提供更为精确的稳健性评估，CRoMa能够帮助研究人员和临床医生选择更可靠的模型，从而提高病理分析的准确性和一致性。

📄 摘要（原文）

Pathology foundation models are approaching clinical deployment, yet remain vulnerable to systematic non-biological variation across centres. Differences in tissue preparation, staining and scanning are strongly encoded in their representations, enabling shortcut learning and weakening generalisation across cohorts and institutions. The Robustness Index (RI) quantifies whether local representation geometry is dominated by biology or by non-biological variation, but its count-based formulation discards distance information. We show that adding distance weights changes little because the deeper limitation lies in RI's pooled, fixed-neighbourhood design, which obscures sample-level heterogeneity and effectively evaluates only a model-dependent subset of samples. We introduce the Cross-confounder Robustness Margin (CRoMa), a sample-resolved measure that directly compares distances to cross-confounder biological matches and same-confounder biological distractors. CRoMa recasts robustness as a cohort-wide margin distribution rather than a single pooled score. We evaluated frozen representations from 20 tile-level encoders across three benchmarks and 4 slide-level encoders on a fourth. Rankings by median CRoMa were broadly consistent across datasets, while the underlying distributions revealed substantial within-model heterogeneity. Every tile encoder retained a confounder-dominated lower tail, whose prevalence and severity varied markedly across models. These distinct robustness profiles frame model selection as a Pareto trade-off between typical and lower-tail robustness. Higher CRoMa was also associated with smaller shortcut-induced performance drops after supervised adaptation. By turning representation geometry into a distributional robustness readout that anticipates downstream shortcut susceptibility, CRoMa provides a principled basis for robustness assessment and model selection.