{"paper":{"title":"FedStain: Modeling Higher-Order Stain Statistics for Federated Domain Generalization in Computational Pathology","license":"http://creativecommons.org/licenses/by/4.0/","headline":"FedStain lets sites share skewness and kurtosis of stain colors to train pathology models that generalize across institutions.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Fengyi Zhang, Junya Zhang, Wenzhuo Sun","submitted_at":"2026-05-14T09:00:04Z","abstract_excerpt":"Robust whole-slide image (WSI) analysis under strict data-governance remains challenging due to substantial cross-institutional stain heterogeneity. Domain generalization (DG) mitigates these shifts but typically requires centralized data, conflicting with privacy regulations. Federated learning (FedL) provides a decentralized alternative; however, existing FedL and federated DG (FedDG) approaches rely almost exclusively on low-order statistics, assuming Gaussian-like stain distributions. In contrast, real-world staining processes often produce asymmetric, heavy-tailed color distributions due "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"To our knowledge, FedStain is the first FedDG approach to explicitly model higher-order stain statistics, enabling robust cross-institutional deployment in computational pathology. FedStain yields consistent improvements, outperforming state-of-the-art FedL, DG, and FedDG baselines by up to +3.9% absolute accuracy.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That skewness and kurtosis, when exchanged as compact descriptors, sufficiently capture the dominant non-Gaussian stain variability and that the contrastive cross-site aggregation produces stain-invariant representations without relaxing privacy constraints.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"FedStain improves federated domain generalization in computational pathology by exchanging higher-order stain moments (skewness, kurtosis) to capture non-Gaussian variability, outperforming baselines by up to 3.9% accuracy.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"FedStain lets sites share skewness and kurtosis of stain colors to train pathology models that generalize across institutions.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"81f3f9931b70907dc7008d144109939f157b4b006ffdb3da4f2f9e2a210263e6"},"source":{"id":"2605.14590","kind":"arxiv","version":1},"verdict":{"id":"2f03f8ff-836f-4258-807e-2ade9aacf3b6","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T01:36:07.728944Z","strongest_claim":"To our knowledge, FedStain is the first FedDG approach to explicitly model higher-order stain statistics, enabling robust cross-institutional deployment in computational pathology. FedStain yields consistent improvements, outperforming state-of-the-art FedL, DG, and FedDG baselines by up to +3.9% absolute accuracy.","one_line_summary":"FedStain improves federated domain generalization in computational pathology by exchanging higher-order stain moments (skewness, kurtosis) to capture non-Gaussian variability, outperforming baselines by up to 3.9% accuracy.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That skewness and kurtosis, when exchanged as compact descriptors, sufficiently capture the dominant non-Gaussian stain variability and that the contrastive cross-site aggregation produces stain-invariant representations without relaxing privacy constraints.","pith_extraction_headline":"FedStain lets sites share skewness and kurtosis of stain colors to train pathology models that generalize across institutions."},"references":{"count":76,"sample":[{"doi":"","year":2017,"title":"A Survey on Deep Learning in Medical Image Analysis,","work_id":"30a9da38-3ce6-4b34-8400-52507b80cdcc","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2018,"title":"Machine Learning Methods for Histopathological Image Analysis,","work_id":"b147eb21-6236-44e9-94e9-7afd02d0c722","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1996,"title":"Summary of the HIPAA Privacy Rule,","work_id":"20b311cb-737e-45aa-965e-90e379822879","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2016,"title":"Regulation (EU) 2016/679: General Data Protection Regulation,","work_id":"2cabb7d2-6fc9-4872-84df-01d9a2751872","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2019,"title":"Quantifying the Effects of Data Augmentation and Stain Color Normalization in Convolutional Neural Networks for Computational Pathology,","work_id":"d1378c34-76c7-47ac-a0b6-e83a3d6f6277","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":76,"snapshot_sha256":"66bdba23564574b52a08e0f4c0f1f86975d487cd71b02a1c70d5192c36d36ea5","internal_anchors":0},"formal_canon":{"evidence_count":1,"snapshot_sha256":"9eb84f22fea6ae9b400ed4c71de60f13fe83c2018317ff77b944a3ee0485c82f"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}