Auditing prediction explanations improves reliability in digital pathology

CHARTER: Auditing Reference Substitution in Hierarchical Compact-Evidence Evaluation for Computational Pathology

Computer Vision and Pattern RecognitionMachine Learning

Summary

In digital pathology, machines analyze many tiny image parts to diagnose disease, and people try to explain their decisions with short summaries called compact evidence. The authors found that how these summaries are evaluated can change depending on which part of the prediction is checked, which can confuse comparisons. They created CHARTER, a new way to clearly state and test which prediction is being explained, making results more trustworthy. This helps make sure that improvements in explanation methods are real and not just due to changing what part of the prediction is looked at.

What this means in practice

  • For computational pathology teams: Evaluate the stability and trustworthiness of explanation methods used for pathology image classification models by clearly specifying prediction targets and references.
  • For machine learning engineers: Design testing protocols that detect when explanation gains are artifacts of shifting the model prediction reference rather than real improvements in explanation quality.

Authors

Hyun Do Jung, Jungwon Choi, Soojung Choi, Yujin Oh, Hwiyoung Kim

Abstract

In digital pathology, compact evidence is often used to explain or audit predictions made by whole-slide image multiple instance learning models. In hierarchical compact-evidence pipelines, candidate filtering introduces a strategy-specific candidate-conditioned prediction alongside the original full-bag prediction. If the evaluation reference changes while the intended target remains the original full-bag prediction, however, not only can the measured fidelity of the same compact evidence change, but comparisons between competing candidate strategies can also change. To make this dependence explicit, we introduce CHARTER, a reference-aware evaluation charter that asks researchers to DECLARE the intended target and reference, QUANTIFY candidate-induced prediction shift, and AUDIT the stability of comparative conclusions. Across the 15 comparisons in our main five-seed Random-K audit, 4 showed determinate reversals; in a matched native-ranking stress test, the ACMIL comparison changed from REVERSED to PRESERVED. CHARTER turns otherwise implicit candidate-filtering and reference choices into an auditable evaluation specification, helping distinguish genuine preservation of the intended prediction from apparent gains induced by changing the prediction being explained.