How many components? What temporal and spatial resolution? I donšt
recommend regressing the movement parameters out first because correlation
with movement is one of the things FIX uses in the classification.
Peace,
Matt.
On 2/17/16, 10:44 AM, "FSL - FMRIB's Software Library on behalf of Zerbi
Valerio" <[log in to unmask] on behalf of [log in to unmask]>
wrote:
>Dear Greg,
>
>yes, my intent is to regress motion parameters from the timeseries before
>running FIX. At the moment I am working with long rs-fMRI scans
>(2000-4000 scans) and I have the problem that the MELODIC would not
>converge into a finite number of component when using the automatic
>algorithm for component selection. Therefore I am now trying to do some
>pre-processing before running ICA-artifact removal using e.g. FIX.
>It actually works well if I do an initial MELODIC, setting a limited
>number of components myself, and thereafter apply FIX with threshold 0
>and the option -m. Then I run a second melodic, and this time the
>component number converges automatically to a certain number, and based
>on those I can use FIX to remove the other artefacts (eg. CSF or white
>matter, scanner artifacts, vascular related, ...). However I was looking
>for a "smarter" and faster way to do the same motion regression script
>without running two melodic. Another thing I noticed is that
>motion-related components are present also in neural components (visible
>also in the spectra), so if I regress the motion beforehand, the actual
>component is "cleaner" and easier to distinguish.
>
>Best,
>Valerio
>
>____________________
>From: FSL - FMRIB's Software Library [[log in to unmask]] on behalf of
>Greg Burgess [[log in to unmask]]
>Sent: Tuesday, February 16, 2016 5:26 PM
>To: [log in to unmask]
>Subject: Re: [FSL] FIX and Motion Regression
>
>Hi Valerio,
>
>Can you explain a bit more what you're trying to accomplish? Are you
>trying to regress motion parameters from the timeseries BEFORE running
>FIX?
>
>As you say in your original email, "I believe that motion is a strong
>confound in the analysis and in the scoring of the components into good
>or bad." Given that signal related to motion is a strong factor in the
>ability of the FIX classifier to distinguish good and bad components, are
>you sure that you want to risk affecting the performance of the FIX
>classifier by removing those motion-related signals?
>
>--Greg
>
>____________________________________________________________________
>Greg Burgess, Ph.D.
>Staff Scientist, Human Connectome Project
>Washington University School of Medicine
>Department of Neuroscience
>Phone: 314-362-7864
>Email: [log in to unmask]
|