An Augmentation Strategy to Mimic Multi-Scanner Variability in MRI

Abstract

Most publicly available brain MRI datasets are very homogeneous in terms of scanner and protocols, and it is difficult for models that learn from such data to generalize to multi-center and multi-scanner data. We propose a novel data augmentation approach with the aim of approximating the variability in terms of intensities and contrasts present in real world clinical data. We use a Gaussian Mixture Model based approach to change tissue intensities individually, producing new contrasts while preserving anatomical information. We train a deep learning model on a single scanner dataset and evaluate it on a multi-center and multi-scanner dataset. The proposed approach improves the generalization capability of the model to other scanners not present in the training data.

The future of MRI in radiation therapy: Challenges and opportunities for the MR community

“…and/or use protocol-agnostic networks. 47,48 Autocontouring tools should also be able to handle inter-scanner differences, such as model, field strength, vendor, etc. Such flexibility would allow inter-institutional cooperation and access to vast heterogeneous training data across multiple centers.…”

“…For instance, autocontouring models should be able to rapidly adapt to new imaging protocols, without the need to obtain and annotate a new training set. Future solutions may include generating large sets of synthetic data with the desired contrast for training using generative adversarial networks 41,44–46 and/or use protocol‐agnostic networks 47,48 . Autocontouring tools should also be able to handle inter‐scanner differences, such as model, field strength, vendor, etc.…”