Learning an MR acquisition-invariant representation using Siamese neural networks

None, None; None, None; None, None; None, None

Learning an MR acquisition-invariant representation using Siamese neural networks

Conference Paper (2019)

Author(s)

W.M. Kouw (University of Copenhagen, TU Delft - Electrical Engineering, Mathematics and Computer Science)

M. Loog (TU Delft - Electrical Engineering, Mathematics and Computer Science, University of Copenhagen)

L.W. Bartels ( University Medical Centre Utrecht)

A.M. Mendrik ( University Medical Centre Utrecht, Netherlands eScience Center)

Research Group

Pattern Recognition and Bioinformatics

MRI Acquisition-variation Representation-learning Siamese-neural-network

DOI related publication

https://doi.org/10.1109/ISBI.2019.8759281 Final published version

To reference this document use

https://resolver.tudelft.nl/uuid:cbd6ed9e-9df4-4894-bb27-fccfc4f96eb1

More Info

expand_more

Publication Year

2019

Language

English

Research Group

Pattern Recognition and Bioinformatics

Pages (from-to)

364-367

Publisher

IEEE

ISBN (print)

978-1-5386-3642-8

ISBN (electronic)

978-1-5386-3641-1

Event

IEEE International Symposium on Biomedical Imaging, ISBI 2019 (2019-04-08 - 2019-04-11), Venice, Italy

Downloads counter

170

Abstract

Generalization of voxelwise classifiers is hampered by differences between MRI-scanners, e.g. different acquisition protocols and field strengths. To address this limitation, we propose a Siamese neural network (MRAI-net) that extracts acquisition-invariant feature vectors. These can consequently be used by task-specific methods, such as voxelwise classifiers for tissue segmentation. MRAI-net is tested on both simulated and real patient data. Experiments show that MRAI-net outperforms voxelwise classifiers trained on the source or target scanner data when a small number of labeled samples is available.