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.
Original languageEnglish
Title of host publicationProceedings of the IEEE International Symposium on Biomedical Imaging
Publication statusAccepted/In press - 2019
EventIEEE International Symposium on Biomedical Imaging, ISBI 2019 - Venice, Italy
Duration: 8 Apr 201911 Apr 2019


ConferenceIEEE International Symposium on Biomedical Imaging, ISBI 2019
Abbreviated titleISBI'19
Internet address

    Research areas

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

ID: 47954088