BagShift: Measuring How Patch Selection Changes the Evidence Seen by Whole-Slide MIL

2026-08-16

AI Summary

This study addresses evidence shift and performance degradation caused by patch selector variations in whole-slide image MIL. We propose the BagShift protocol, which isolates and quantifies the impact of sampling strategies on observed evidence by fixing feature extractors and predictors while varying only selectors. Through paired evaluation that decouples selector responses from case-level biases, we demonstrate that patch quantity does not equate to effective evidence. Experiments reveal that local sampling reduces PANDA QWK by 17.96 points, whereas our proposed nonlinear joint aggregation strategy improves QWK by 7.87 points. These findings effectively elucidate the mechanisms underlying deployment environment fluctuations and enhance model robustness in computational pathology applications.

Abstract

Whole-slide multiple-instance learning (MIL) observes only the patches admitted by its selector. Deployment can alter this selector through compute limits, tissue masking, or regional workflows, even when the patch count is unchanged. We introduce BagShift, a paired protocol that changes the selector for the same case while holding its features and predictor fixed, thereby isolating selector response from case mix. With equal 128-patch budgets, sampling across the tissue or concentrating around one coordinate exposes markedly different evidence: on PANDA, the two views reduce quadratic weighted kappa by 1.57 and 17.96 points, respectively (QWK reported on the $\times100$ scale). On CAMELYON16, lesion annotations withheld from model development show that localized views retain tumor in only 10.0\% of micrometastatic observations, and matched exposure does not consistently recover the loss. The same fixed-count stressor produces a much smaller response on external lung subtyping, although differences in relative coverage make cross-task severity descriptive. When repeated localized observations are available, unioning their patches before one nonlinear MIL pass improves PANDA QWK by 7.87 points over averaging regional predictions. Patch count specifies computation, not observed evidence; deployment evaluations should report both what a selector preserves and how repeated observations are aggregated.

Problem

• Whole-slide MIL
• Patch selection
• BagShift
• Deployment evaluation
• Evidence shift

Innovation

• BagShift
• Whole-Slide MIL
• Patch Selection
• Paired Protocol
• Evidence Aggregation