{"paper":{"title":"Physics-Grounded Adversarial Stain Augmentation with Calibrated Coverage Guarantees","license":"http://creativecommons.org/licenses/by/4.0/","headline":"CASA performs adversarial augmentation in Macenko stain space with DKW-calibrated budgets to cover unseen hospital variations.","cross_cats":["cs.CV","cs.LG"],"primary_cat":"eess.IV","authors_text":"Mingi Hong","submitted_at":"2026-05-12T04:39:33Z","abstract_excerpt":"Stain variation across hospitals degrades histopathology models at deployment. Existing augmentation methods perturb color spaces with arbitrary hyperparameters, lacking both a principled budget and coverage guarantees for unseen centers. We propose \\textbf{C}alibrated \\textbf{A}dversarial \\textbf{S}tain \\textbf{A}ugmentation (\\textbf{CASA}), which performs adversarial augmentation in the Macenko stain parameter space with a budget calibrated from multi-center statistics via the DKW inequality. On Camelyon17-WILDS (5 seeds), CASA achieves $93.9\\% \\pm 1.6\\%$ slide-level accuracy -- outperformin"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"On Camelyon17-WILDS (5 seeds), CASA achieves 93.9% ± 1.6% slide-level accuracy -- outperforming HED-strong (88.4% ± 7.3%), RandStainNA (85.2% ± 6.7%), and ERM (63.9% ± 11.3%) -- with the highest worst-group accuracy (84.9% ± 0.9%) among all 10 compared methods.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That calibrating an adversarial budget from multi-center statistics via the DKW inequality in Macenko space supplies coverage guarantees for truly unseen centers without post-hoc tuning or access to target-domain data.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"CASA achieves 93.9% slide-level accuracy on Camelyon17-WILDS by adversarially augmenting stains in Macenko space with DKW-calibrated coverage, outperforming baselines including in worst-group accuracy.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"CASA performs adversarial augmentation in Macenko stain space with DKW-calibrated budgets to cover unseen hospital variations.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"026f711cadc86b08dcdedbfff917baded2e05a567572000de67ad0d0d5484b7f"},"source":{"id":"2605.13889","kind":"arxiv","version":1},"verdict":{"id":"f9b262b4-40ea-4f66-be44-2aaa982f3bc8","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T05:53:26.338758Z","strongest_claim":"On Camelyon17-WILDS (5 seeds), CASA achieves 93.9% ± 1.6% slide-level accuracy -- outperforming HED-strong (88.4% ± 7.3%), RandStainNA (85.2% ± 6.7%), and ERM (63.9% ± 11.3%) -- with the highest worst-group accuracy (84.9% ± 0.9%) among all 10 compared methods.","one_line_summary":"CASA achieves 93.9% slide-level accuracy on Camelyon17-WILDS by adversarially augmenting stains in Macenko space with DKW-calibrated coverage, outperforming baselines including in worst-group accuracy.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That calibrating an adversarial budget from multi-center statistics via the DKW inequality in Macenko space supplies coverage guarantees for truly unseen centers without post-hoc tuning or access to target-domain data.","pith_extraction_headline":"CASA performs adversarial augmentation in Macenko stain space with DKW-calibrated budgets to cover unseen hospital variations."},"references":{"count":10,"sample":[{"doi":"","year":2019,"title":"Medical Image Analysis , volume=","work_id":"22c55b13-bde3-46e0-8068-23c42693cab2","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2009,"title":"IEEE International Symposium on Biomedical Imaging (ISBI) , pages=","work_id":"ad90dd3b-9b7d-419c-9cb0-36d47c75be93","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Shen, Yiqing and Luo, Yulin and Shen, Dinggang and Ke, Jing , booktitle=. 2022 , publisher=","work_id":"70841798-69eb-4054-b8b6-b73c21a8c98f","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Zheng, Guangtao and Huai, Mengdi and Zhang, Aidong , booktitle=","work_id":"6a70ff8d-9443-445b-b582-6a6ed624694b","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Advances in Neural Information Processing Systems (NeurIPS) , year=","work_id":"6b06916d-7ccd-42d6-a3e9-4cc014e12b3a","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":10,"snapshot_sha256":"ab8d89747287e7c552f4d031bae701742cce96f2dcb4427816da2f8c9fa08e5f","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}