arXiv:2607. 22861v1 Announce Type: cross Abstract: Pathology foundation models (FMs) produce powerful tile-level representations which remain sensitive to scanner and staining variability, undermining deployment across laboratories.
Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv AI.
arXiv:2607. 04401v1 Announce Type: cross Abstract: How robust and generalisable are pathology foundation models and have their scaling limites been reached?
HERO (Histology Encoder for Robust Representation in Oncology) is a ViT‑G/14 pathology foundation model trained with DINO and iBOT objectives and refined using high‑resolution Gram anchoring on a 500‑million‑tile corpus from about 575,000 clinical whole‑slide images. It demonstrates superior robustness to center, scanner, and stain variation compared to other state‑of‑the‑art foundation models, while maintaining competitive performance on tile‑level classification, segmentation, and gene‑expression prediction. Across 39 slide‑level clinical tasks, HERO ranks first on average and achieves the best average rank across six benchmark frameworks under an equal‑weighted analysis.
arXiv:2601. 05148v2 Announce Type: replace-cross Abstract: Pathology foundation models substantially advanced the possibilities in computational pathology --- yet tradeoffs in terms of performance, robustness, and computational requirements remained, which limited their clinical deployment.
arXiv:2602. 14010v2 Announce Type: replace-cross Abstract: Pathology foundation models (PFMs) generalize well across computational pathology tasks but remain costly for gigapixel whole-slide image analysis.
arXiv:2607. 25497v2 Announce Type: replace-cross Abstract: Pathology foundation models encode non-biological variation introduced by tissue preparation, staining and scanning, enabling shortcut learning that undermines generalisation across institutions.
TAP-Path is a task‑adaptive compression framework that restructures a pretrained Virchow2 encoder for histopathology. It selectively removes transformer blocks, prunes patch tokens, and adds a lightweight gated task head, reducing parameters by 24.96% and FLOPs by 35.20% while maintaining high accuracy on a 32‑class benchmark. The method achieves competitive test metrics and improved rare‑class performance, with strong external validation on CPTAC samples.