[FieldTrip] regressconfound

[McGowin, Inna]

Hello Raghavan,
I am actually looking at GLM head motion correction method (FieldTrip
regressconfound) with some phantom MEG data and found a few things that
might be of interest to you:
- the method is best for Evoked Response analysis as it requires trial
averaging at the end of the correction (you want to average motion
corrected data not data that has motion); the method works for the resting
state too but with an error linearly increasing with the amount of motion
-  you should see differences in before and after regressconfound
correction as the method re-aligns the waveforms for one (initial=averaged)
head position; you will even see the differences if comparing true
recordings (no motions) to recordings with the motion (which I did with the
phantom) as the method is not perfect and can't possibly remove all motion
confounds/effects completely
- your question: "What do I do if I am comparing a condition across
different time points, rather than 2 conditions in one time point?", my
answer: I would use one (initial) averaged head position (for 1st trail for
example) to compare, because you want to convert/correct your measurements
for *relative comparison* and they should represent the same head position
(for better statistical sensitivity).


Arjen,
Please correct me if I am mistaking.

Cheers,
Inna
Wake Forest Baptist Health
MEG lab



Inna McGowin


On Mon, Apr 7, 2014 at 12:59 PM, Raghavan Gopalakrishnan <
gopalar.ccf at gmail.com> wrote:

> Arjen,
>
> Thanks for all your previous help on using regressconfound.
> It would be great if you can shed light on some questions I asked earlier
> (See below).
> Also, when applying regressconfound to time-frequency (TF) data,
> regressconfound removes variance from TF from individual trials. So,
> utimately regressconfound helps to improve the induced TF responses rather
> than evoked responses. (By induced, I mean calculating TF on trial by trial
> basis and then computing average, But evoked I mean calculating TF on
> averaged data)
> If one is interested in only evoked, then regressconfound would not be of
> much help. Is that correct?
>
> Thanks,
> Raghavan
>
>
>
>
>
>
> Date: Wed, 26 Feb 2014 11:38:23 -0500
> From: Raghavan Gopalakrishnan <gopalar.ccf at gmail.com>
> To: fieldtrip at science.ru.nl
> Subject: [FieldTrip] regressconfound and statistics
> Message-ID: <A36DB35E-1C8A-4C8A-A0F3-083B8E1131C1 at gmail.com>
> Content-Type: text/plain; charset=windows-1252
>
> Arjen,
>
> I agree with your steps and your assumptions about my data, though I am
> only interested in evoked activity. I am indeed doing the steps 1 and 2 in
> the correct order. I just kept the 4 blocks and confounds associated with
> them separate from each other (for book keeping purposes), but I used the
> mean head position of all 4 blocks to demean the translations and rotations
> in each block. Technically, I guess this is the same as appending the
> blocks prior to running regress confound.
>
> When I said, "However, the problem is, whatever significance I found
> earlier (i.e. by comparing means rather than t-statistic) doesn?t test
> significant now.? I was comparing the means of the data that has been run
> through regress confound. When I plot (sum square of all gradiometers in
> each subject in each condition)  before and after regress confound, I see
> some differences. Should I not see any differences at all?
>
>  But there is one catch. What do I do if I am comparing a condition across
> different time points, rather than 2 conditions in one time point? MEG
> collected at two different time points will have different average head
> positions. In that case, should I use one average head position (computed
> from time point - 1) and use that to demean the translations and rotations
> in rest of the time points?
>
>
> Thanks for your support.
> Raghavan
> --
>
>
> _______________________________________________
> 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/20140408/c0c1079e/attachment-0001.html>