Our plan was to denoise/motion regress prior to a dual-regression analysis.
One other question about the dual-regression method. From reading the
Filippini paper, I was unsure if TFCE was used for clustering/thresholding
the randomise t-maps, or if the gaussian/gamma modeling was used similar to
how the original thresholded melodic IC maps are created. Thanks.
Chris
On Aug 17 2010, Stephen Smith wrote:
>Hi
>
>On 16 Aug 2010, at 20:35, Christopher Bell wrote:
>
>> Hello. I have a couple quick questions.
>>
>> 1)Would it be reasonable to include other regressors in the multiple
>> linear regression along with melodic timecourses? For example, motion
>> parameters? It couldn't hurt I suppose, except for the degrees of
>> freedom.
>
>Sure
>
>> 2)Is there a rule of thumb for how many components you can remove for
>> DOF reasons? For example, we have one subject with 260 TRs and 107
>> components. This subject did not move much, which was my first guess.
>> Since, we only expect to see around ~10-18 "real components" that will
>> mean we are removing a large number of components for this subject. Can
>> this become a problem at some point? I was thinking of applying a
>> low-pass filter prior to running the ICA for our subjects, b/c many of
>> these components are high-frequency and present primarily in the
>> inferior regions where I know there are many arteries and veins. My
>> guess is some kind of cardiac source. I notice that low-pass filtering
>> is not generally done though? We currently do smoothing of 6-mm FWHM,
>> and .01Hz high-pass filter. Thanks for any suggestions.
>
> If you remove lots of components before running FEAT then yes the DOF
> will eventually become incorrect. However if you include the MELODIC
> timecourses as confound EVs then that will be correctly dealt with by the
> FEAT stats.
>
> Low pass filtering is sometimes done before running ICA, I don't have a
> strong clear opinion on whether this is a good thing or not. However if
> you're just doing data cleanup before running FEAT, this would probably
> not interact well with the autocorrelation modelling in FEAT.
>
>> 3)Also, is there a difference between fsl_regfilt and fsl_glm? It seems
>> fsl_glm has more output options. Thanks.
>
> You're right - there's some overlap - I would use whichever is more
> convenient for your purposes.
>
>Cheers.
>
>
>>
>> Chris Bell
>> University of Minnesota
>>
>
>
>---------------------------------------------------------------------------
>Stephen M. Smith, Professor of Biomedical Engineering
>Associate Director, Oxford University FMRIB Centre
>
>FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
>+44 (0) 1865 222726 (fax 222717)
>[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
>---------------------------------------------------------------------------
>
>
>
>
|