Hi,
I think the challenge here is figuring out the magnitude of the
effect/concern raised by Carp 2013 when using the particular high-pass
filter implemented in FSL (i.e., a locally-Gaussian-weighted straight line
fit). But yes, in theory, the issue raised by Carp is in play.
cheers,
-MH
--
Michael Harms, Ph.D.
-----------------------------------------------------------
Conte Center for the Neuroscience of Mental Disorders
Washington University School of Medicine
Department of Psychiatry, Box 8134
660 South Euclid Ave.Tel: 314-747-6173
St. Louis, MO 63110Email: [log in to unmask]
On 7/6/16, 1:07 PM, "FSL - FMRIB's Software Library on behalf of Katherine
Lawrence" <[log in to unmask] on behalf of [log in to unmask]>
wrote:
Hello!
I thought I would try re-posting my question in case someone might be able
to provide some insight. The full background/content of my question is
below, but briefly I was hoping to gain some insight as to why it seems to
be okay to use fsl_motion_outliers after you've done temporal filtering on
a timeseries which includes motion-contaminated timepoints, when from my
understanding doing temporal filtering on a timeseries with
motion-contaminated time points could lead to ringing artifacts in the
data which fsl_motion_outliers wouldn't correct per se.
Thank you for your time,
Katherine
---------- Forwarded message ----------
From: Katherine Lawrence
Date: 2016-06-22 16:15 GMT-07:00
Subject: Temporal Filtering and fsl_motion_outliers
To: [log in to unmask]
Hello!
I have a question about using fsl_motion_outliers to censor
motion-contaminated timepoints from resting-state data. Here's my
understanding of how to do this so far, followed by my question:
-FSL recommends using confound EVs from fsl_motion_outliers to take care
of motion-contaminated time points instead of deleting the
motion-contaminated time points.
-Based on the May 2014 thread, it seems like for seed-based analyses one
should include the confound EVs from fsl_motion_outliers not in any
preprocessing GLMs where we're regressing out the nusiance regressors to
obtain residuals. Instead, we should include them in the same GLM as where
we're including the seed's timeseries (and therefore the GLM where we are
actually getting our betas of interest).
-So it sounds like preprocessing, including temporal filtering, should
have already been run by the time the confound EVs from
fsl_motion_outliers are included in the seed-based GLM above.
-However, Carp 2013 (NeuroImage) said that if you bandpass filter while
you still have the 'bad' timepoints in your timeseries, the filtering may
introduce ringing artifacts in the data, spreading the artifacts across
time.
-I would therefore think that waiting to include regressors from
fsl_motion_outliers until the seed-based GLM (and after bandpass
filtering) would result in the motion-contaminated time points spreading
their artifact to other time points during the bandpass filtering step.
Therefore, even if the motion-contaminated time points themselves would be
essentially censored by including a confound EV for them, there could
still be motion-related noise in other time points.
I haven't been able to find any mentions of this sort of potential
problem, though, which makes me think I might be missing something. Is my
understanding of this whole situation correct? If not, what am I missing?
If so, is there a reason to not be concerned about potential ringing from
bandpass filtering the data? Just let me know if I haven't phrased
anything clearly enough.
Thank you in advance,
Katherine
________________________________
The materials in this message are private and may contain Protected Healthcare Information or other information of a sensitive nature. If you are not the intended recipient, be advised that any unauthorized use, disclosure, copying or the taking of any action in reliance on the contents of this information is strictly prohibited. If you have received this email in error, please immediately notify the sender via telephone or return mail.
|