TY - GEN
T1 - Quantitative Comparison of Monte-Carlo Dropout Uncertainty Measures for Multi-class Segmentation
AU - Camarasa, Robin
AU - Bos, Daniel
AU - Hendrikse, Jeroen
AU - Nederkoorn, Paul
AU - Kooi, Eline
AU - van der Lugt, Aad
AU - de Bruijne, Marleen
N1 - Funding Information:
Acknowledgments. This work was funded by Netherlands Organisation for Scientific Research (NWO) VICI project VI.C.182.042. The PARISK study was funded within the framework of CTMM, the Center for Translational Molecular Medicine, project PARISK (grant 01C-202), and supported by the Dutch Heart Foundation.
Publisher Copyright:
© 2020, Springer Nature Switzerland AG.
Copyright:
Copyright 2020 Elsevier B.V., All rights reserved.
PY - 2020
Y1 - 2020
N2 - Over the past decade, deep learning has become the gold standard for automatic medical image segmentation. Every segmentation task has an underlying uncertainty due to image resolution, annotation protocol, etc. Therefore, a number of methods and metrics have been proposed to quantify the uncertainty of neural networks mostly based on Bayesian deep learning, ensemble learning methods or output probability calibration. The aim of our research is to assess how reliable the different uncertainty metrics found in the literature are. We propose a quantitative and statistical comparison of uncertainty measures based on the relevance of the uncertainty map to predict misclassification. Four uncertainty metrics were compared over a set of 144 models. The application studied is the segmentation of the lumen and vessel wall of carotid arteries based on multiple sequences of magnetic resonance (MR) images in multi-center data.
AB - Over the past decade, deep learning has become the gold standard for automatic medical image segmentation. Every segmentation task has an underlying uncertainty due to image resolution, annotation protocol, etc. Therefore, a number of methods and metrics have been proposed to quantify the uncertainty of neural networks mostly based on Bayesian deep learning, ensemble learning methods or output probability calibration. The aim of our research is to assess how reliable the different uncertainty metrics found in the literature are. We propose a quantitative and statistical comparison of uncertainty measures based on the relevance of the uncertainty map to predict misclassification. Four uncertainty metrics were compared over a set of 144 models. The application studied is the segmentation of the lumen and vessel wall of carotid arteries based on multiple sequences of magnetic resonance (MR) images in multi-center data.
UR - http://www.scopus.com/inward/record.url?scp=85093121226&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-60365-6_4
DO - 10.1007/978-3-030-60365-6_4
M3 - Conference contribution
AN - SCOPUS:85093121226
SN - 9783030603649
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 32
EP - 41
BT - Uncertainty for Safe Utilization of Machine Learning in Medical Imaging, and Graphs in Biomedical Image Analysis - 2nd International Workshop, UNSURE 2020, and 3rd International Workshop, GRAIL 2020, Held in Conjunction with MICCAI 2020, Proceedings
A2 - Sudre, Carole H.
A2 - Fehri, Hamid
A2 - Arbel, Tal
A2 - Baumgartner, Christian F.
A2 - Dalca, Adrian
A2 - Tanno, Ryutaro
A2 - Van Leemput, Koen
A2 - Wells, William M.
A2 - Sotiras, Aristeidis
A2 - Papiez, Bartlomiej
A2 - Ferrante, Enzo
A2 - Parisot, Sarah
PB - Springer Science and Business Media Deutschland GmbH
T2 - 2nd International Workshop on Uncertainty for Safe Utilization of Machine Learning in Medical Imaging, UNSURE 2020, and the 3rd International Workshop on Graphs in Biomedical Image Analysis, GRAIL 2020, held in conjunction with the 23rd International Conference on Medical Image Computing and Computer-Assisted Intervention, MICCAI 2020
Y2 - 8 October 2020 through 8 October 2020
ER -