Enhancing the Reliability of Medical AI through Expert-guided Uncertainty Modeling

2026-04-02

AI Summary

This work addresses the challenge of unreliable aleatoric uncertainty estimation in medical AI systems under data noise and ambiguity, which can lead to uncontrolled risk. The authors propose a novel dual-ensemble approach—along with a lightweight variant—based on total variance decomposition that explicitly disentangles aleatoric and epistemic uncertainties. For the first time, inter-expert disagreement is leveraged as a supervisory signal alongside standard labels. By integrating expert-guided target generation with multi-task learning, the method significantly improves uncertainty estimation quality by 9%–50% across diverse medical tasks, including image classification, segmentation, and visual question answering. These results demonstrate a marked enhancement in system reliability and underscore the critical role of expert knowledge in developing high-assurance medical AI.

Abstract

Artificial intelligence (AI) systems accelerate medical workflows and improve diagnostic accuracy in healthcare, serving as second-opinion systems. However, the unpredictability of AI errors poses a significant challenge, particularly in healthcare contexts, where mistakes can have severe consequences. A widely adopted safeguard is to pair predictions with uncertainty estimation, enabling human experts to focus on high-risk cases while streamlining routine verification. Current uncertainty estimation methods, however, remain limited, particularly in quantifying aleatoric uncertainty, which arises from data ambiguity and noise. To address this, we propose a novel approach that leverages disagreement in expert responses to generate targets for training machine learning models. These targets are used in conjunction with standard data labels to estimate two components of uncertainty separately, as given by the law of total variance, via a two-ensemble approach, as well as its lightweight variant. We validate our method on binary image classification, binary and multi-class image segmentation, and multiple-choice question answering. Our experiments demonstrate that incorporating expert knowledge can enhance uncertainty estimation quality by $9\%$ to $50\%$ depending on the task, making this source of information invaluable for the construction of risk-aware AI systems in healthcare applications.

Problem

• medical AI
• uncertainty estimation
• aleatoric uncertainty
• expert disagreement
• risk-aware AI

Innovation

• expert-guided uncertainty
• aleatoric uncertainty
• two-ensemble learning
• medical AI reliability
• uncertainty estimation