Learning an MR acquisition-invariant representation using Siamese neural networks
This addresses a domain-specific problem for medical imaging researchers by improving tissue segmentation in MRI, but it is incremental as it builds on existing Siamese networks and voxelwise classifiers.
The paper tackled the problem of MRI-scanner differences hampering voxelwise classifier generalization by proposing MRAI-NET, a Siamese neural network that extracts acquisition-invariant feature vectors, and experiments showed it outperforms voxelwise classifiers trained on source or target scanner data with limited labeled samples.
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.