Abstract

It has become obvious that machine learning, and within machine learning deep learning [1], is providing both a powerful and ubiquitous set of technologies able to answer a much wider range of questions than initially forseen. These techniques are transforming the way we think of experiments in particular because they often require much more data than the traditional data analysis techniques. They also have a potential for disruption of the current research framework [2]. In this context, I will review first the type of machine learning techniques available and their specific applications Magnetic Resonance in Medicine. I will emphasize what is needed for these techniques to be efficiently working especially in functional and anatomical brain imaging applications and their caviats and limitations. I will consider how machine learning advances are solving for some but not all of the reproducibility issues that have recently been demonstrated in MR brain imaging as well as more generally in biomedical sciences. I will conclude by a list of actionable recommendations for technical and cultural changes necessitated by the rise of machine learning technologies.

Acknowledgements

J.-B.P. was partially funded by NIH-NIBIB P41 EB019936 (ReproNim) NIH-NIMH R01 MH083320 (CANDIShare) and NIH 5U24 DA039832 (NIF), as well as the Canada First Research Excellence Fund, awarded to McGill University for the Healthy Brains for Healthy Lives initiative.