Problem
Research questions and friction points this paper is trying to address.
Whole-Slide Image Condensation
Self-Supervised Learning
Distribution Matching
Computational Pathology
Nonparametric Distribution Matching for Self-Supervised Whole-Slide Image Condensation
🤖 AI Summary
This study addresses the computational bottleneck imposed by the ultra-high resolution of whole slide images (WSIs) and the inability of existing compression methods to preserve critical feature distributions. We reformulate WSI compression as a distribution matching problem and propose NICER, a framework that incorporates non-parametric priors and slice-adaptive capacity. By integrating self-supervised learning with differentiable approximation algorithms, NICER achieves principled, distribution-aware compression that efficiently retains task-relevant features. Extensive experiments across five datasets demonstrate an average accuracy improvement of 7.44%, significantly optimizing the efficiency–accuracy trade-off. Furthermore, the clinical utility of the proposed method has been validated by expert pathologists.
📝 Abstract
Histological whole-slide images (WSIs) are central to computational pathology but pose severe computational challenges due to their extremely high resolution, often spanning several gigabytes per slide. To enable scalable learning, existing methods apply self-supervised data condensation to reduce computational cost, but typically rely on heuristic prototype learning and do not explicitly preserve learning-relevant feature distributions for downstream tasks. In response, we introduce a principled reformulation of WSI condensation as a distribution-matching problem under a fixed representational lens, and develop NICER, a tractable approximation framework based on a nonparametric prior with slide-adaptive capacity. Experiments on five histopathology datasets, together with clinical evaluation from a board-certified pathologist, show that NICER consistently outperforms prior methods, achieving an average accuracy improvement of 7.44% while offering improved efficiency-accuracy trade-offs, highlighting the benefits of principled, distribution-aware condensation for scalable histological representation learning. Source codes are available in https://github.com/nmduonggg/NICER.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Nonparametric Distribution Matching
Whole-Slide Image Condensation
Self-Supervised Learning
Computational Pathology
NICER
🔎 Similar Papers
No similar papers found.
