One-Slide Calibration of Pathology Foundation Models

Computer Science > Computer Vision and Pattern Recognition arXiv:2610.08944 (cs) [Submitted on 6 Oct 2026] Title: One-Slide Calibration of Pathology Foundation Models Authors: Ming Ren Hou, Tianyi Huang.

计算机科学 > 计算机视觉与模式识别 arXiv:2610.08944 (cs) [提交于 2026 年 10 月 6 日] 标题：病理学基础模型的单切片校准 作者：Ming Ren Hou, Tianyi Huang。

Abstract: Scanner variation changes how pathology foundation models represent the same tissue. We introduce SlideRuler, which uses regions within a slide as internal controls to estimate and correct acquisition-induced shifts in other regions.

摘要：扫描仪的差异会改变病理学基础模型对同一组织的处理方式。我们引入了 SlideRuler，它利用切片内的区域作为内部对照，以估计并校正其他区域中由采集引起的偏移。

A transfer map learned from paired rescans enables calibration from a single scan at inference while keeping the foundation model fixed. Across two encoders and five SCORPION scanners, learned transfer reduces mean target-to-source embedding distance by 16.3-38.5% relative to raw embeddings.

通过从配对重扫描中学习到的转换映射，可以在推理时仅通过单次扫描进行校准，同时保持基础模型不变。在两个编码器和五台 SCORPION 扫描仪的测试中，学习到的转换相对于原始嵌入，将平均目标到源的嵌入距离减少了 16.3% 至 38.5%。

Comparisons with unrelated same-scanner controls reveal a positive same-slide contribution across all four evaluation settings, including scanner holdout. A source-anchored variant reduces source-feature displacement by 47.7-83.6% relative to learned transfer while retaining most of its alignment gain.

与不相关的同扫描仪对照组相比，在包括扫描仪留出法在内的所有四种评估设置中，均显示出积极的同切片贡献。一种源锚定变体在保留大部分对齐增益的同时，将源特征位移相对于学习到的转换减少了 47.7% 至 83.6%。

By drawing calibration information from the slide itself, SlideRuler offers a path toward more consistent use of frozen pathology models across imaging systems.

通过从切片本身提取校准信息，SlideRuler 为在不同成像系统间更一致地使用冻结病理模型提供了一条途径。