On the notion of missingness for path attribution explainability methods in medical settings: Guiding the selection of medically meaningful baselines

2025-08-20

AI Summary

Medical deep learning interpretability is hindered by semantic distortion in path attribution caused by fixed baselines (e.g., all-zero inputs) that lack clinical meaning. To address this, we propose a counterfactual-guided adaptive baseline selection framework: “missingness” is redefined using clinically plausible counterfactual samples—normal and semantically proximal to the input—thereby overcoming the semantic limitations of conventional baselines. Our method employs a variational autoencoder to generate individualized counterfactual baselines and integrates them with path-integrated gradients to yield medically grounded attributions. The framework is model-agnostic and technically extensible. Experiments across three medical datasets demonstrate that our approach significantly improves attribution faithfulness and clinical relevance compared to standard baselines, yielding more interpretable and trustworthy feature importance distributions. To our knowledge, this is the first work to systematically reconstruct the notion of “missingness” in path attribution from a clinical semantic perspective.

Technology Category

• Computer Vision: Other Foundations of Computer Vision
• Machine Learning: Other Foundations of Machine Learning
• Reasoning under Uncertainty: Other Foundations of Reasoning under Uncertainty

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

The explainability of deep learning models remains a significant challenge, particularly in the medical domain where interpretable outputs are critical for clinical trust and transparency. Path attribution methods such as Integrated Gradients rely on a baseline input representing the absence of relevant features ("missingness"). Commonly used baselines, such as all-zero inputs, are often semantically meaningless, especially in medical contexts where missingness can itself be informative. While alternative baseline choices have been explored, existing methods lack a principled approach to dynamically select baselines tailored to each input. In this work, we examine the notion of missingness in the medical setting, analyze its implications for baseline selection, and introduce a counterfactual-guided approach to address the limitations of conventional baselines. We argue that a clinically normal but input-close counterfactual represents a more accurate representation of a meaningful absence of features in medical data. To implement this, we use a Variational Autoencoder to generate counterfactual baselines, though our concept is generative-model-agnostic and can be applied with any suitable counterfactual method. We evaluate the approach on three distinct medical data sets and empirically demonstrate that counterfactual baselines yield more faithful and medically relevant attributions compared to standard baseline choices.

Problem

• Defining meaningful feature absence in medical AI explainability
• Addressing semantically meaningless baselines in path attribution methods
• Dynamically selecting clinically relevant counterfactual baselines per input

Innovation

• Counterfactual-guided baseline selection
• Variational Autoencoder generates baselines
• Medically meaningful feature absence representation