Hi Timothy,
it's not quite that simple I'm afraid. The dim-est is based on a
global estimate (from the data covriance matrix) that basically
checks for how many leading eigenvalues need to be ignored in order
for the remaining minor eigenspetrum
to follow a certain form. Once you've regressed out part of the
decomposition the denoised data will in general not have a data
covariance structure which simple corresponds to the old
eigenspectrum with the major values removed.
Instead, the entire spectrum will change.
Essentially what all these techniques do is to look for a first
significant step in the eigenvalues. After a first denoising run you
might get a dim estimate the second time around which is even bigger
or close to the previous estimate since now smaller steps, which were
previously masked due to the presence of the large part of the
eigenspectrum, become important.
Also note that melodic tries to be on the conservative side, i.e. I
much prefer over-estimating the dimensionality then under-estimating
it, the former is a bit annoying as you get a few more spurious
components but the latter might get really poor estimates for the
signals of interest.
hope this helps
chriatian
On 20 Mar 2007, at 17:47, Timothy Laumann wrote:
> Hi Steve,
>
> I did. The procedure I followed was as follows (I was working from the
> command line).
>
> /betfunc {file}.hdr b{file} %Where {file} has already been
> preprocessed by
> other means (motion correction, smoothing, normalization, etc.)
>
> /ip b{file}.hdr fb{file} 10 -s 1.273 -t 100.0 -1
>
> /melodic -i fb{file}.hdr -o fb{file}.ica -v -bgthreshold=10 --tr=1
> -mmthresh=0.5 --report --0all %Where fb{file} is the filtered, brain
> extracted, preprocessed file
>
> I then use the filter option as described below with the fb
> {file}.hdr file
> as input and filtered_fb{file}.ica as output with the appropriate
> components
> to be filtered. I expected for the filtered_fb{file}.img to simply
> be a
> recomposition from the mixing matrix minus the selected components,
> but when
> I perform melodic on the filter output file I get many more
> components than
> I thought I would. Is this because there is unaccounted for variation
> during the original ICA operation that is now coming into play?
> Thanks for your help!
>
> Tim
>
>
>
>
> On Mon, 19 Mar 2007 15:10:50 +0000, Steve Smith
> <[log in to unmask]> wrote:
>
>> Hi,
>>
>> Did you apply the cleanup to the data after it had been through the
>> other preprocessing steps (motion correction etc.)?
>>
>> If you used the Melodic GUI then there should be a file called
>> filtered_func_data in the original .ica output directory, and you
>> should be feeding THAT preprocessed image into the cleanup.
>>
>> Does this help resolve things?
>>
>> Cheers, Steve.
>>
>>
>>
>>
>> On 15 Mar 2007, at 15:12, Timothy Laumann wrote:
>>
>>> Dear FSLlist,
>>>
>>> I have been trying to use the filter option to remove noise from my
>>> data and
>>> I am wondering how exactly the function works. I thought that the
>>> filter
>>> function was merely removing the specified components from the
>>> mixing matrix
>>> and rebuilding the data sans those components. Just to see what
>>> would
>>> happen, I filtered 25 of the 50 components generated from a 200
>>> image dataset,
>>>
>>> melodic -i data.hdr -v -o filtered_data.ica --mix=melodic_mix -f
>>> "2,3,.."
>>>
>>> and then performed a melodic ICA on the filtered_data.img that is
>>> created.
>>> I expected to get the 25 remaining components only, but instead
>>> still got
>>> close to 50 components. Clearly there is something about this
>>> process that
>>> I do not understand.
>>> Thanks for the help!
>>>
>>> Tim
>>
>>
>> ---------------------------------------------------------------------
>> ---
>> ---
>> 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
>> ---------------------------------------------------------------------
>> ---
>> ---
>> =====================================================================
>> ====
____
Christian F. Beckmann
University Research Lecturer
Oxford University Centre for Functional MRI of the Brain (FMRIB)
John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK.
[log in to unmask] http://www.fmrib.ox.ac.uk/~beckmann
tel: +44 1865 222551 fax: +44 1865 222717
|