Rethinking Uncertainty Quantification and Entanglement in Image Segmentation

2026-03-19

AI Summary

In image segmentation, aleatoric and epistemic uncertainties are often highly entangled, limiting model interpretability and practical deployment. This work proposes the first metric to quantify the degree of such uncertainty entanglement and systematically evaluates combinations of uncertainty estimation methods—including Probabilistic UNet, diffusion models, ensembles, Monte Carlo Dropout, Softmax-based approaches, and SSN—across diverse benchmarks. Experiments reveal that ensemble methods achieve the lowest entanglement and best performance in out-of-distribution detection, while Softmax ensembles demonstrate consistently robust results across tasks. Notably, the optimal strategy varies depending on the dataset. By introducing a principled framework for quantifying and analyzing uncertainty disentanglement, this study provides both theoretical insights and practical guidance for improving reliability in medical and general-purpose image segmentation.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Uncertainty Representations
• Computer Vision: Segmentation

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Security and Privacy: Large-scale security measurements
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics

Abstract

Uncertainty quantification (UQ) is crucial in safety-critical applications such as medical image segmentation. Total uncertainty is typically decomposed into data-related aleatoric uncertainty (AU) and model-related epistemic uncertainty (EU). Many methods exist for modeling AU (such as Probabilistic UNet, Diffusion) and EU (such as ensembles, MC Dropout), but it is unclear how they interact when combined. Additionally, recent work has revealed substantial entanglement between AU and EU, undermining the interpretability and practical usefulness of the decomposition. We present a comprehensive empirical study covering a broad range of AU-EU model combinations, propose a metric to quantify uncertainty entanglement, and evaluate both across downstream UQ tasks. For out-of-distribution detection, ensembles exhibit consistently lower entanglement and superior performance. For ambiguity modeling and calibration the best models are dataset-dependent, with softmax/SSN-based methods performing well and Probabilistic UNets being less entangled. A softmax ensemble fares remarkably well on all tasks. Finally, we analyze potential sources of uncertainty entanglement and outline directions for mitigating this effect.

Problem

• Uncertainty Quantification
• Aleatoric Uncertainty
• Epistemic Uncertainty
• Uncertainty Entanglement
• Image Segmentation

Innovation

• uncertainty quantification
• aleatoric uncertainty
• epistemic uncertainty
• uncertainty entanglement
• image segmentation