Print

Print

What kind of analysis are you doing?  For example, you could implement a seed-based approach in which you include the seed time series as one EV, and the matrix returned by 'fsl_motion_outliers' as a set of "confound" EVs in the same GLM.  In that case, the ensuing beta for the seed time series EV represents the slope of the regression line while effectively ignoring the confound time points.  And if you aren't doing any prewhitening, this beta will be the same as what you would get if you actually masked out the bad time points and then computed the regression using the remaining time points.  So, in this mode of operation (i.e., no prewhitening), the two approaches (regression with confound matrix and formal "scrubbing") are the same.  If you are prewhitening, that correspondence no longer holds exactly.

Further, the statistics (z-stats) in the regression approach (i.e., seed EV + confound matrix in a single model) are appropriate, because the dof's are accounted for properly.

What you can't do is a two-step approach in which you first residualize the time series using just the confound matrix, and then use the output of that for a seed-based analysis (at least not without properly accounting for the dof's).

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  63110 Email: [log in to unmask]

From: "David V. Smith" <[log in to unmask]>
Reply-To: FSL - FMRIB's Software Library <[log in to unmask]>
Date: Thursday, May 22, 2014 11:44 AM
To: FSL - FMRIB's Software Library <[log in to unmask]>
Subject: Re: [FSL] fsl_motion_outliers question.

Hi Michael,

Thanks for the explanation here. I have a related followup question regarding "scrubbing" vs including the output of fsl_motion_outliers as a set of confound EVs in my analysis. Would there be any advantage of "scrubbing" (i.e., physically removing or masking) those outlier time points instead of having regressors (i.e., the output of fsl_motion_outliers) to account for them? I think the fsl_motion_outliers documentation still recommends the regression approach over the "scrubbing" approach, but I still see reviewers claiming that outlier time points can only be removed via scrubbing (and not via regression).

Thanks,
David


On May 22, 2014, at 10:04 AM, Harms, Michael wrote:


Hi,
I believe that using the confound matrix from fsl_motion_outliers in fsl_regfilt would return a "residualized" time series, with the "mean" added back in, so that the "scrubbed/censored" time points as coded in the confound matrix are replaced by the "mean" of the time series (i.e., their residual is 0).  I put "mean" in quotation marks, because I'm not sure what exactly is added back in -- whether it is the mean of the full time series, which would include the contribution of the time points that you actually want to ignore, or if what is added back in is actually the intercept of the GLM fit (i.e., the mean excluding the time points that you want to ignore).

The question becomes what do you intend to do with this output.  For example, you can't just go and do a seed-based correlation analysis with it.  That is not "scrubbing", because you would be pretending that the time points as identified by fsl_motion_outliers are now somehow fully legitimate time points, which isn't the case (i.e., they would contribute to the calculation of the correlation, whereas in the Power et al. approach to scrubbing, the bad time points are masked out the analysis altogether).

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  63110 Email: [log in to unmask]

From: Andreas Lidström <[log in to unmask]>
Reply-To: FSL - FMRIB's Software Library <[log in to unmask]>
Date: Thursday, May 22, 2014 5:10 AM
To: FSL - FMRIB's Software Library <[log in to unmask]>
Subject: Re: [FSL] fsl_motion_outliers question.

I'm no expert, but I would say no. the output is not a time series but a matrix with one column of zeroes for each outlier image, with the affected image being marked by a one, to be included in the 1:st level GLM.


2014-05-21 19:54 GMT+02:00 Broulidakis M.J. <[log in to unmask]>:

Dear FSL experts.

 

I haven’t found much information on it online. But is it possible to perform motion scrubbing using the output of fsl_motion_outliers with the command fsl_regfilt? I would be grateful for any advice (even if it is just a yes or no).

 

Best wishes,

 

John

--------------------

John Broulidakis
Developmental Brain-Behaviour Laboratory
University of Southampton

 

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.


--------------------------------------------
David V. Smith, Ph.D.
Postdoctoral Fellow, Delgado Lab
Department of Psychology
Rutgers University
Newark, NJ 07102
--------------------------------------------

 

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.