3D ReX: Causal Explanations in 3D Neuroimaging Classification

TL;DR

3D ReX is a novel causality-based explainability tool for 3D medical imaging models, helping clinicians understand model decisions by highlighting crucial regions, demonstrated on stroke detection.

Contribution

Introduces 3D ReX, the first causality-based post-hoc explainability method for 3D models in medical imaging, enhancing interpretability.

Findings

Effectively highlights key regions in 3D images for stroke detection

Provides insights into spatial feature importance

Improves trust in AI medical diagnostics

Abstract

Explainability remains a significant problem for AI models in medical imaging, making it challenging for clinicians to trust AI-driven predictions. We introduce 3D ReX, the first causality-based post-hoc explainability tool for 3D models. 3D ReX uses the theory of actual causality to generate responsibility maps which highlight the regions most crucial to the model's decision. We test 3D ReX on a stroke detection model, providing insight into the spatial distribution of features relevant to stroke.