I’m running analyses using fMRIPrep outputs where I’d like to keep the data in the subject’s T1w space until the group level and recently noticed that running fsl_motion_outliers (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FSLMotionOutliers) on the T1w space functionals yields wildly different outlier timepoint estimation than running on the MNI space functionals e.g. for the same run, 4 outliers in T1w space and 48 outliers in MNI space. The fmriprep reports reflect the poor motion and playing movies of both functional images also indicate poor motion, so I’m not sure where the breakdown is happening or whether it has consequences for using the T1w space images in analyses but wanted to raise this issue.
Hi @oesteban, the intention is not to re-estimate motion parameters (although reading fsl_motion_outliers documentation more closely, I see that mcflirt is run), I’ve just been trying to use this to flag time points to be scrubbed for analysis. We’re running a multiecho sequence so it didn’t seem to make sense to run this on any one echo of the raw data. What would be the suggestion for identifying and scrubbing time points corrupted by motion?
As you found out, you’d be re-running MCFLIRT (which is problematic). fMRIPrep provides you with head-motion parameters AND Framewise-Displacement time-series within the confounds files. That’s what you should use for scrubbing.