Hi - thanks, maybe I'll try the power-spectrum approach. I currently model
the noise artefact in the way you proposed. I wrote a script to denoise the
data in a way that involves multiplying the noise model with the
corresponding PEs and subtracting the result from the original data.
However, since the 0.41 Hz noise artefact has amplitude modulation, this
approach never really gets rid of all the noise in the data (since the noise
model has standard amplitude). As a workaround, I use the cosine noise model
to "capture" (or preselect) the real noise, then use the timeseries of the
peak voxel for "real noise" as a seed in a subsequent model-based analyis to
get a better noise model fit. I then run the same procedure using the
denoise-script to subtract the real noise from the data. Of course, I use
high-pass filtering in such a way that I maximize a priori chances of
isolating only noise-related data, so that I do not subtract physiological
(resting state) data from the dataset. I then do the model-free analysis on
the denoised dataset. Does that seem to be a valid approach to you?
Best regards, Rutger.
-----Original Message-----
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]]On
Behalf Of Christian F. Beckmann
Sent: Tuesday, July 08, 2008 11:19 AM
To: [log in to unmask]
Subject: Re: [FSL] TICA methods question
Hi Rutger,
nope, not easily - unless you're willing to perform the entire
analysis on the power spectra rather than the original time courses.
You could create a cosine basis set of the given frequency and add
these as confound regressors in the GLM, though.
hth
Christian
______
On 28 Jun 2008, at 10:10, Rutger Goekoop wrote:
> Dear Christian,
>
> Would it be possible to use an fslscript to filter out a particular
> frequency from the power spectrum of a 4D timeseries, say, a 0.41 Hz
> noise
> artefact?
>
> Cheers, Rutger.
>
> -----Original Message-----
> From: FSL - FMRIB's Software Library [mailto:[log in to unmask]]On
> Behalf Of Christian F. Beckmann
> Sent: Friday, June 27, 2008 10:51 PM
> To: [log in to unmask]
> Subject: Re: [FSL] TICA methods question
>
>
> Hi Christina,
>
> In that paper the analysis was performed on the power-specral density
> of the FMRI time series. You can try using fslpspec to do the
> transformation, though this might be very slow on your system as the
> current version is based on very un-efficient code. We've fixed this
> issue for the next release version, so it might be worth waiting for
> this instead. Alternatively, you can do this in matlab.
> hth
> Christian
>
> On 27 Jun 2008, at 15:57, Christina Hugenschmidt wrote:
>
>> Hi All,
>>
>> I posted this last week, but was hoping someone could help me out. I
>> am
>> interested in doing a TICA analysis like the one in the Cerebral
>> Cortex
>> Damoiseaux 2007 paper. It is my understanding from reading the paper
>> and
>> the supporting material online that the data need first to be Fourier
>> transformed, and that this was not automatically done as part of the
>> TICA
>> process. I was wondering if someone could please tell me the
>> methodology
>> for doing that, such as what software program was used to transform
>> the
>> data and what files were then input into the analysis? I apologize
>> if this
>> is an obvious sort of question - it is just something I have not done
>> before and I would like to make sure it is done correctly.
>>
>> Thanks for any help!
>> Christina
>
>
> --
> Mijn Postvak In wordt beschermd door SPAMfighter.
> 1588 spam-mails zijn er tot op heden geblokkeerd.
> Download de gratis SPAMfighter via deze link:
http://www.spamfighter.com/lnl
--
Mijn Postvak In wordt beschermd door SPAMfighter.
1610 spam-mails zijn er tot op heden geblokkeerd.
Download de gratis SPAMfighter via deze link: http://www.spamfighter.com/lnl
|