Measuring What Matters: A Causal Reliance Score for Histopathology Image Segmentation
Reliable histopathology segmentation requires models that rely on pathology-relevant structures rather than spurious background correlations. In this work, we argue that improving architectures alone…
- 用途
- 原典で確認
- 難易度の目安
- Hard
- 実装コストの目安
- High
概要
Reliable histopathology segmentation requires models that rely on pathology-relevant structures rather than spurious background correlations. In this work, we argue that improving architectures alone is insufficient if models continue to rely on spurious correlations rather than pathology-relevant features. We introduce a causality-inspired evaluation framework based on controlled interventions that explicitly probes model sensitivity to foreground tissue structure and background/domain variation. Building on this, we propose the Causal Reliance Score (CRS), a metric that quantifies the relative sensitivity of model predictions to foreground perturbation versus background/domain variation under controlled interventions. To support this analysis, we design augmentation strategies that act as implicit counterfactual interventions during training, encouraging invariance to nuisance factors while preserving morphology. We demonstrate how these interventions reveal behavioural differences between models that are not captured by standard metrics such as Dice or IoU. While we instantiate our framework using a self-supervised transformer (CAUSE-ViT), the primary contribution of this work is not a new architecture, but a principled intervention-based methodology and evaluation metric for auditing model behaviour in medical image segmentation. Our results suggest that evaluating model behaviour through counterfactual interventions is essential for developing reliable and clinically robust medical image segmentation systems.
参考リンク
リンク先のライセンス・再現性・データ条件をご確認ください。自動分類や実装の目安は保証ではありません。
実装を始めるプロンプト
参考リンクを確認したうえで、次の内容をコピーして利用できます。
# 論文・記事 Measuring What Matters: A Causal Reliance Score for Histopathology Image Segmentation # 概要 Reliable histopathology segmentation requires models that rely on pathology-relevant structures rather than spurious background correlations. In this work, we argue that improving architectures alone… # 原典 https://arxiv.org/abs/2610.41763 上記の手法を実装してください。実行前に依存関係とライセンスを確認してください。