Recalibration of Aleatoric and EpistemicRegression Uncertainty in Medical Imaging

Max-Heinrich LavesInstitute of Medical Technology and Intelligent Systems, Hamburg University of Technology
Institute of Mechatronic Systems, Leibniz Universit ̈at Hannover
, Sontje IhlerInstitute of Mechatronic Systems, Leibniz Universit ̈at Hannover, Jacob F. FastInstitute of Mechatronic Systems, Leibniz Universit ̈at Hannover
Hannover Medical School
, Lüder A. KahrsCentre for Image Guided Innovation and Therapeutic Intervention, The Hospital for Sick Children, Toronto
Department of Mathematical and Computational Sciences, University of Toronto Mississauga
, Tobias OrtmaierInstitute of Mechatronic Systems, Leibniz Universit ̈at Hannover
MIDL 2020 special issue
Publication date: 2021/04/28
PDF · arXiv · Video · Code

Abstract

The consideration of predictive uncertainty in medical imaging with deep learning is of utmost importance. We apply estimation of both aleatoric and epistemic uncertainty by variational Bayesian inference with Monte Carlo dropout to regression tasks and show that predictive uncertainty is systematically underestimated. We apply sigma scaling with a single scalar value; a simple, yet effective calibration method for both types of uncertainty. The performance of our approach is evaluated on a variety of common medical regression data sets using different state-of-the-art convolutional network architectures. In our experiments, sigma scaling is able to reliably recalibrate predictive uncertainty. It is easy to implement and maintains the accuracy. Well-calibrated uncertainty in regression allows robust rejection of unreliable predictions or detection of out-of-distribution samples. Our source code is available at: https://github.com/mlaves/well-calibrated-regression-uncertainty

Keywords

bayesian approximation · variational inference

Bibtex @article{melba:2021:008:laves, title = "Recalibration of Aleatoric and EpistemicRegression Uncertainty in Medical Imaging", authors = "Laves, Max-Heinrich and Ihler, Sontje and Fast, Jacob F. and Kahrs, Lüder A. and Ortmaier, Tobias", journal = "Machine Learning for Biomedical Imaging", volume = "1", issue = "MIDL 2020 special issue", year = "2021" }

2021:008 cover