We donšt generally smooth at all prior to performing ICA and thereby we
get interpretable components that are easy to classify into signal and
noise. Of course this is with very high quality HCP data with high
spatial and temporal resolution and plenty of minutes of fMRI.
Peace,
Matt.
On 2/1/18, 5:23 AM, "FSL - FMRIB's Software Library on behalf of Kostakis
Gkiatis" <[log in to unmask] on behalf of [log in to unmask]> wrote:
>I can tell better than when I was smoothing with 4, 5 or even 6mm
>kernel. I am sure that some of the components are not signal but at
>least now I know that the majority is. I believe that something is wrong
>with the scanner. Hopefully, an upgrade in the scanner is being
>performed now, so if this is true the artefact should disappear.
>In the results some clusters extend to the white matter but they are
>mostly in grey matter in areas that it was expected, so I believe that
>they are correct.
>Do you think that some other parameter could influence the ICA
>components and make them uninterpretable when I smooth with 5mm FWHM
>kernel?
>
>Thank you for your help.
>Yours Sincerely,
>Kostas
>
>---
>Electrical and Computer Engineer
>Phd Candidate in Brain Imaging
>Medical Image Processing, Algorithms and Applications
>National Technical University of Athens
>9,Iroon Polytechneiou str.,
>157 73 Zografou Campus,
>Athens, GREECE
>phone:+30 210 772 3577
>
>
>On , Matt Glasser wrote:
>> With that much smoothing can you even tell what is signal and noise?
>> 8mm
>> FWHM smoothing in the volume will make localization of signal to the
>> greymatter quite challenging.
>>
>> Peace,
>>
>> Matt.
>>
>> On 1/31/18, 6:21 AM, "FSL - FMRIB's Software Library on behalf of
>> Gkiatis
>> Kostakis" <[log in to unmask] on behalf of [log in to unmask]>
>> wrote:
>>
>>> Dear Ludovica,
>>>
>>> Thank you for your quick response. After reading the article you send
>>> me,
>>> I changed the FWHM of the Gaussian kernel to 8mm and the signal has
>>> now
>>> appeared. There are still too many components as the one I attached
>>> but
>>> now many of remaining components can be clearly classified as signal.
>>> My voxel size is 2.5mm, so I thought that 5mm FWHM filter should be
>>> fine.
>>> But it wasn't. Do you have any suggestions why?
>>>
>>> Thank you in advance.
>>> Yours Sincerely,
>>> Kostas
|