Learning an MR acquisition-invariant representation using Siamese neural networks

TL;DR

This paper introduces MRAI-NET, a Siamese neural network that learns acquisition-invariant features to improve MRI voxelwise classification across different scanners, especially with limited labeled data.

Contribution

The paper presents a novel Siamese neural network architecture that effectively extracts scanner-invariant features for MRI classification tasks.

Findings

MRAI-NET outperforms traditional classifiers on both simulated and real data.

It achieves better generalization with fewer labeled samples.

The method is effective across different MRI acquisition protocols.

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.