[FieldTrip] Fwd: head motion regression with ft_regressconfound on a continuous MEG data

[McGowin, Inna]

Hello,
I am trying to apply head motion correction to a continuous  MEG recording
with
Ft_regressconfound.

Bcs the output of the Ft_regressconfound function is in the trial format ,
technically, I can proceed with the head motion regression method on
the continuous
data by dividing the data into smaller segments; which I did. I have 120
sec of MEG data (SR=600 Hz) that were divided into 10, or 20, or 30 sec
trials. I run  ft_regressconfound on these trials. No matter what the trial
length  or the total number of trials are, I get out of memory message in
Matlab.

Matlab error message:
??? Error using==>mldivide
Out of memory.

Error in ==> ft_regressconfound at 213
beta = regr\dat; % B= X\Y

The error appears in the part of the code where the linear equation is
solved directly.
I tried to only work with a subset of all trials (half for example) with no
luck.

I rum MatlabR2010a on Windows XP 32-bit with the following memory
allocation in Matlab:
Max possible array: 851 Mb
Memory available for all array: 1703 Mb
Memory used by Matlab: 850 Mb
RAM: 3317 Mb

Were anybody else able to run this motion regression algorithm in Matlab
without out of memory? Is there anything I can do to avoid the problem? I
can share the code if needed.

Thanks in advance,
Inna









Inna McGowin


On Mon, Nov 18, 2013 at 10:42 AM, Arjen Stolk <a.stolk8 at gmail.com> wrote:

> Hi Inna,
> Ft_regressconfound will remove trial by trial variability of signal
> amplitude or frequency power that can be explained by variability in head
> position over those trials. The input, whether sensor or source level,
> should contain a trial field (keep trials in the previous step - eg at
> timelockanalysis). Also variance due to head position can only be reliably
> estimated when the neural signal is consistent over trials (as with an
> event related modulation or potential). I would not know how to segment a
> resting state into trials (for further analysis, given that those analyses
> typically involve connectivity measures over the resting period) but if you
> an idea on how to do this, you could give ft_regressconfound a try. Note
> that the output will also be in the trials format. Hope this has made its
> use a bit more clear. Yours, arjen
>
> Op 13 nov. 2013 11:50 schreef "McGowin, Inna" <mcgoiv0 at wfu.edu> het
> volgende:
>
> >
> > Hello everybody,
> >
> > I would like to use the  ft_regressconfound function to remove/reduce
> the head motion in a continuous MEG data.
> > It's a resting state data and no averaging can be applied. Is there a
> way to apply the correction without using the
> >
> > ft_timelockanalysis(cfg, data) function?
> >
> > I understand that the ft_regressconfound function has limits and caveats
> but I still would like to see the results of its correction.
> >
> >
> > Another way to ask this question is:
> > what is the format/structure of the second input (timelock) in the
> > ft_regressconfound(cfg, timelock) function and how it can be created
> without ft_timelockanalysis?
> >
> >
> > Thanks,
> >
> > Inna
> >
> > _______________________________________________
> > fieldtrip mailing list
> > fieldtrip at donders.ru.nl
> > http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>
>
> _______________________________________________
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20131127/ccd04d8c/attachment-0002.html>