Principle-Guided Supervision for Interpretable Uncertainty in Medical Image Segmentation

2026-05-09

AI Summary

This work addresses the limited semantic interpretability of uncertainty estimation in medical image segmentation, which often fails to link spatial uncertainty patterns with underlying image ambiguity sources. To bridge this gap, the authors propose PriUS—a principle-guided uncertainty supervision framework—that, for the first time, incorporates human perceptual principles—such as inter-structure contrast, image degradation level, and anatomical geometric complexity—into uncertainty modeling. By leveraging evidential learning, PriUS explicitly constrains the spatial distribution of uncertainty to align with these perceptual cues. The framework introduces interpretable supervision objectives and consistency metrics that ensure uncertainty maps faithfully reflect image ambiguity characteristics. Experiments on the ACDC, ISIC, and WHS datasets demonstrate that PriUS yields uncertainty estimates more consistent with human perceptual principles while maintaining state-of-the-art segmentation performance.

Abstract

Uncertainty quantification complements model predictions by characterizing their reliability, which is essential for high-stakes decision making such as medical image segmentation. However, most existing methods reduce uncertainty to a scalar confidence estimate, leaving its spatial distribution semantically underconstrained. In this work, we focus on uncertainty interpretability, namely, whether estimated uncertainty behaves in a human-understandable manner with respect to sources of ambiguity. We identify three perception-aligned principles requiring the spatial distribution of uncertainty to reflect: (1) image contrast between structures, (2) severity of image corruption, and (3) geometric complexity in anatomical structures. Accordingly, we develop a principle-guided uncertainty supervision framework (PriUS) based on evidential learning, in which the corresponding supervision objectives are explicitly enforced during training. We further introduce quantitative metrics to measure the consistency between predicted uncertainty and image attributes that induce ambiguity. Experiments on ACDC, ISIC, and WHS datasets showed that, compared with state-of-the-art methods, PriUS produced more consistent uncertainty estimates while maintaining competitive segmentation performance.

Problem

• uncertainty interpretability
• medical image segmentation
• spatial uncertainty distribution
• perception-aligned principles
• ambiguity sources

Innovation

• uncertainty interpretability
• principle-guided supervision
• evidential learning
• medical image segmentation
• spatial uncertainty