Learn to Estimate Labels Uncertainty for Quality Assurance

TL;DR

This paper introduces a method to estimate label uncertainty in deep learning, particularly for medical applications, combining it with epistemic uncertainty to improve quality control and decision referral systems.

Contribution

It proposes a novel approach to model label uncertainty alongside epistemic uncertainty, addressing a gap in existing deep learning uncertainty estimation methods for medical use.

Findings

Modeling label uncertainty improves quality assurance in medical deep learning.

Combining label and epistemic uncertainties enhances decision referral systems.

The approach provides a more comprehensive understanding of model reliability.

Abstract

Deep Learning sets the state-of-the-art in many challenging tasks showing outstanding performance in a broad range of applications. Despite its success, it still lacks robustness hindering its adoption in medical applications. Modeling uncertainty, through Bayesian Inference and Monte-Carlo dropout, has been successfully introduced for better understanding the underlying deep learning models. Yet, another important source of uncertainty, coming from the inter-observer variability, has not been thoroughly addressed in the literature. In this paper, we introduce labels uncertainty which better suits medical applications and show that modeling such uncertainty together with epistemic uncertainty is of high interest for quality control and referral systems.