One-Slide Calibration of Pathology Foundation Models
Quick Answer
This paper shows that SlideRuler calibrates pathology foundation models by using internal controls from tissue slides, achieving a 16.3-38.5% reduction in embedding distance across various scanners.
Quick Take
This method allows for consistent model application across imaging systems without altering the foundation model, enhancing diagnostic reliability.
Key Points
- SlideRuler uses regions within a slide as internal controls for calibration.
- Achieves a 16.3-38.5% reduction in target-to-source embedding distance.
- Demonstrates positive same-slide contributions across four evaluation settings.
- Source-anchored variant reduces source-feature displacement by 47.7-83.6%.
- Offers a path for consistent use of frozen pathology models across scanners.
DeepSignal Analysis
What happened
The SlideRuler method calibrates pathology foundation models by utilizing internal controls from tissue slides. This approach has demonstrated a reduction in embedding distance by 16.3-38.5% across various scanners, allowing for consistent model application without modifying the foundation model.
Key evidence
- SlideRuler uses regions within a slide as internal controls to estimate and correct shifts caused by acquisition, enabling calibration from a single scan.
- The method achieved a mean target-to-source embedding distance reduction of 16.3-38.5% when compared to raw embeddings across two encoders and five SCORPION scanners.
- A source-anchored variant of SlideRuler reduced source-feature displacement by 47.7-83.6% relative to learned transfer while maintaining most alignment gains.
Why it matters
This calibration method enhances the reliability of pathology models across different imaging systems, which is crucial for consistent diagnostic outcomes. By leveraging internal slide data, it minimizes the need for extensive model retraining, potentially streamlining the integration of AI in clinical settings.
What to watch
Paper Resources
📖 Reader Mode
~2 min readAbstract: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. 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. 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. By drawing calibration information from the slide itself, SlideRuler offers a path toward more consistent use of frozen pathology models across imaging systems.
| Comments: | Accepted at the NeurIPS 2026 Workshop AI at Scale for Clinical Impact (ASCI): Cancer Pathology Foundation Models |
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.08944 [cs.CV] |
| (or arXiv:2610.08944v1 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2610.08944 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Tianyi Huang [view email]
[v1]
Tue, 6 Oct 2026 18:09:36 UTC (2,184 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CV
See more →ProMoE-FL: Prototype-conditioned Mixture of Experts for Multimodal Federated Learning with Missing Modalities
ProMoE-FL introduces a Prototype-conditioned Mixture-of-Experts framework for multimodal federated learning, effectively addressing missing modalities. It outperforms existing methods on four chest X-ray datasets, demonstrating superior feature synthesis capabilities in both homogeneous and heterogeneous settings.