Explainability of AI Uncertainty: Application to Multiple Sclerosis Lesion Segmentation on MRI

2025-04-07

AI Summary

AI-based uncertainty estimation in MRI cortical lesion segmentation for multiple sclerosis lacks clinical interpretability. Method: We propose the first analytical framework that attributes model prediction uncertainty to clinically interpretable factors—such as lesion size, morphology, and cortical involvement—by integrating deep ensemble-based uncertainty quantification, instance-level uncertainty decomposition, clinical factor association modeling, and out-of-distribution generalization evaluation. Contribution/Results: Our framework establishes statistically significant alignment (p < 0.001) between predicted uncertainty and expert annotator confidence, enabling bidirectional consistency between technical uncertainty and clinical reasoning. Validated on a multicenter dataset comprising 206 patients and nearly 2,000 lesions, it substantially enhances the clinical decision-support value of uncertainty estimates, enabling reliable segmentation both in-distribution and under distributional shift.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Uncertainty Representations
• Multiagent Systems: Multiagent Systems 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
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Trustworthy artificial intelligence (AI) is essential in healthcare, particularly for high-stakes tasks like medical image segmentation. Explainable AI and uncertainty quantification significantly enhance AI reliability by addressing key attributes such as robustness, usability, and explainability. Despite extensive technical advances in uncertainty quantification for medical imaging, understanding the clinical informativeness and interpretability of uncertainty remains limited. This study introduces a novel framework to explain the potential sources of predictive uncertainty, specifically in cortical lesion segmentation in multiple sclerosis using deep ensembles. The proposed analysis shifts the focus from the uncertainty-error relationship towards relevant medical and engineering factors. Our findings reveal that instance-wise uncertainty is strongly related to lesion size, shape, and cortical involvement. Expert rater feedback confirms that similar factors impede annotator confidence. Evaluations conducted on two datasets (206 patients, almost 2000 lesions) under both in-domain and distribution-shift conditions highlight the utility of the framework in different scenarios.

Problem

• Explaining AI uncertainty sources in MS lesion segmentation
• Enhancing clinical interpretability of predictive uncertainty
• Assessing uncertainty impact on lesion size and shape

Innovation

• Deep ensembles explain predictive uncertainty sources
• Focus on medical and engineering factors, not error
• Evaluate lesion size, shape, cortical involvement impact