I’m working on running FSL GLM analyses on fMRIprep preprocessed data with AROMA. I’ve added the non-steady-state volume confounds to my first-level confound matrix to avoid the issues associated with including those volumes in analysis, but was hoping to gain a better understanding of why the number of non-steady-state volumes varies from 0-3 between subjects/scans despite using the same scanner protocols. What methods / metrics is fMRIprep using to determine whether or not a volume is classified as non-steady-state?
FMRIPREP uses a simple image intensity outlier detection heuristic implemented in Nipype. More details and references here https://github.com/nipy/nipype/blob/d2a48e1eb/nipype/algorithms/confounds.py#L974-L995
A lot of these processes are stochastic and depend on environment (for example temperature) so it’s possible that steady state is reached at different times from one scan to another.
Great, thank you! Is the global signal value used for this calculation the same as that reported in the global signal column of the confounds-regressors tsv output of fMRIprep?
I’m not 100% sure but I think global signal uses brain mask. Non-steady state detector averages values over the whole image.