Contrastive Integrated Gradients: A Feature Attribution-Based Method for Explaining Whole Slide Image Classification

2025-11-11

AI Summary

Existing WSI interpretability methods (e.g., Integrated Gradients) struggle to precisely localize tumor subtype–discriminative regions in high-resolution whole-slide images, undermining the trustworthiness of AI-assisted diagnosis. To address this, we propose Contrastive Integrated Gradients (CIG), which computes gradients in the logit space relative to a contrastive reference class, thereby substantially improving attribution accuracy and visual clarity for subtype-specific regions. We further introduce two weakly supervised evaluation metrics—MIL-AIC and MIL-SIC—specifically designed for multi-instance learning, enabling, for the first time, quantitative assessment of attribution quality under this paradigm. CIG rigorously satisfies the integral attribution axioms, ensuring theoretical consistency. Extensive experiments on CAMELYON16, TCGA-RCC, and TCGA-Lung demonstrate that CIG significantly outperforms baseline methods in both attribution localization accuracy and clinical interpretability.

Technology Category

• Computer Vision: Interpretability, Explainability, and Transparency
• Machine Learning: Multi-instance/Multi-view Learning
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

Interpretability is essential in Whole Slide Image (WSI) analysis for computational pathology, where understanding model predictions helps build trust in AI-assisted diagnostics. While Integrated Gradients (IG) and related attribution methods have shown promise, applying them directly to WSIs introduces challenges due to their high-resolution nature. These methods capture model decision patterns but may overlook class-discriminative signals that are crucial for distinguishing between tumor subtypes. In this work, we introduce Contrastive Integrated Gradients (CIG), a novel attribution method that enhances interpretability by computing contrastive gradients in logit space. First, CIG highlights class-discriminative regions by comparing feature importance relative to a reference class, offering sharper differentiation between tumor and non-tumor areas. Second, CIG satisfies the axioms of integrated attribution, ensuring consistency and theoretical soundness. Third, we propose two attribution quality metrics, MIL-AIC and MIL-SIC, which measure how predictive information and model confidence evolve with access to salient regions, particularly under weak supervision. We validate CIG across three datasets spanning distinct cancer types: CAMELYON16 (breast cancer metastasis in lymph nodes), TCGA-RCC (renal cell carcinoma), and TCGA-Lung (lung cancer). Experimental results demonstrate that CIG yields more informative attributions both quantitatively, using MIL-AIC and MIL-SIC, and qualitatively, through visualizations that align closely with ground truth tumor regions, underscoring its potential for interpretable and trustworthy WSI-based diagnostics

Problem

• Explains WSI classification by highlighting tumor vs non-tumor discriminative regions
• Addresses limitations of attribution methods in high-resolution whole slide images
• Enhances interpretability for AI-assisted cancer diagnostics across multiple cancer types

Innovation

• Computes contrastive gradients in logit space
• Highlights class-discriminative regions using reference comparisons
• Introduces MIL-AIC and MIL-SIC attribution quality metrics