Greetings,
Almost all of the research I’ve found on MRI and deep learning deals with structural MRI datasets, not fMRI.
Do you know of some preprocessing steps specific to deep learning on fMRI, recommended network architectures, frameworks or tutorials on classification of an fMRI dataset via neural networks?
For example, would it be viable to first extract time series of fMRI ROI obtained through nilearn and then use that as an input for a recurrent neural network?
Hi,
A quick look on Google Scholar reviews several articles that use deep learning on fMRI images. Here is a list of review articles that may be good for you to look at.
In general, most machine learning studies on fMRI have used resting state data due to their wide availability and generalizability. Very often studies will use whole brain functional connectivity matrices as feature spaces. To match network architecture, these may need to be flattened before they are input to a neural net, but besides that, there isn’t any preprocessing step that is specific to this application.
Best,
Steven
MANAGED BY INCF