The abstract is read at the publisher; the summary is JClub's.
DELTA, a training-free label-free post-hoc correction method, improves patch-level representation robustness on PathoROB across all 20 pathology foundation models.
- Why it matters: Staining and scanning variation limits pathology foundation model generalization, and existing embedding correction methods require extra matched data or training.
- What they did: DELTA estimates recurrent embedding directions from paired original-versus-counterfactual stain-appearance variants using only 4% of eligible patches per source dataset, then attenuates aligned components via a fixed projection.
- The result: On PathoROB it improves robustness versus raw embeddings and preserves classification gains after patch aggregation; it also improves survival discrimination on TCGA-BRCA, screening in a clinical cohort, and reduces embedding variation across PLISM without refitting.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.