Hi,
I guess you identify several ICs presumably related to stimulus correlated motion (a likely scenario as Christian just pointed out). If you include all their time courses as covariates in the model then you easily run into problems because these are highly correlated and are primarily seperated in the spatial domain. So just limited yourself to, for example, one or two.
Maybe that helps-
Andreas
________________________________
Von: FSL - FMRIB's Software Library im Auftrag von Richard Albistegui-DuBois
Gesendet: Mo 15.08.2005 23:14
An: [log in to unmask]
Betreff: Re: [FSL] AW: [FSL] MELODIC
We're using a block design. They don't actually seem all that highly
correlated, but that's no problem; I'll try your other suggestions.
Sincere thanks for your prompt reply.
On Aug 15, 2005, at 1:56 PM, Christian Beckmann wrote:
> Hi
>
> On 15 Aug 2005, at 19:24, Richard Albistegui-DuBois wrote:
>
>> Hello--one additional question on this old topic, if you have no
>> objection.
>>
>> Can I assume that this does not work well with stimulus-correlated
>> motion?
>
> We actually find that it often delineates stimulus-correlated motion
> from stimulus quite nicely by virtue of the fact that they are often
> spatially well separated processes. One problem is that ICA only finds
> a linear decomposition - strong head motion will typically result in
> various components showing the spatial characteristics of the motion.
>
>> When I have used MELODIC to identify motion components which are
>> stimulus correlated to some degree, and try to include them as
>> nuisance covariates, I get complaints about orthogonality.
>>
>
> You mean in FEAT? They must be stimulus-correlated to an extremely
> high degree fro this to happen... are you using a single-event design
> or a block design?
>
>> Is there any other way to try to remove stimulus-correlated motion
>> via MELODIC?
>>
>
> You could try and use the filter option in melodic:
>
> i) run melodic
> (ii) identify the effects you wish to remove
> (iii) melodic -i <original data file name> --mix=<estimated
> melodic_mix> --filter="string of component numbers to remove"
>
> this creates an .ica directory with an blah_ICfiltered.nii.gz data set
> in it.
>
> hope this helps
> Christian
>
>> Thanks much.
>>
>> -Richard
>>
>> On Apr 25, 2005, at 9:41 AM, Christian Beckmann wrote:
>>
>>
>>> Hi Julius,
>>>
>>> the report subdir contains a set of tXX.txt files which themselves
>>> contain the time course of component XX as a 1 column text file.
>>> These
>>> you can use as nuisance regressors in Feat - use the "Custom (1 entry
>>> per volume)" basic shape and switch off convolution and temporal
>>> derivative.
>>> best
>>> christian
>>>
>>>
>>>
>>> On 25 Apr 2005, at 17:25, Julius Fridriksson wrote:
>>>
>>>
>>>> Hi Andreas and/or Russ -- Would you mind giving me a little more
>>>> detail
>>>> how you would use the timecourse of a given component in the design
>>>> matrix in FEAT? Thanks very much for taking time to help. Best --
>>>> Julius
>>>>
>>>>
>>>>>>> [log in to unmask] 04/22/05 04:57PM >>>
>>>>>>>
>>>> Hi Julius,
>>>> yes - you may use melodic to identify residual motion in your data.
>>>> Such components are usually characterized by "spikes" in their time
>>>> courses and by edge effects in their spatial maps. Christians little
>>>> fmri shop of horror provides an example
>>>> (http://www.fmrib.ox.ac.uk/~beckmann/homepage/academic/littleshop/
>>>> motion/art1.html).
>>>> However, as in any data-driven analyses it is - in the end - up to
>>>> you
>>>> to interprete the components melodic identifies in a sensible
>>>> manner.
>>>> In
>>>> any case, this may guide you to some extend. You may then use
>>>> melodic
>>>> to
>>>> remove such components from the data (type melodic --help for the
>>>> usage). Alternatively, you can include their time course (without
>>>> further convolution) into the design matrix as a confound of no
>>>> interest.
>>>> Cheers-
>>>> Andreas
>>>>
>>>> ________________________________
>>>>
>>>> Von: FSL - FMRIB's Software Library im Auftrag von Julius
>>>> Fridriksson
>>>> Gesendet: Fr 22.04.2005 21:00
>>>> An: [log in to unmask]
>>>> Betreff: [FSL] MELODIC
>>>>
>>>>
>>>>
>>>> Hi -- At last year's FSL workshop there was some talk about using
>>>> MELODIC to estimate head motion related noise in fMRI datasets (and
>>>> using the resulting data to co-vary in FEAT). Is this correct? If
>>>> so,
>>>> is
>>>> there a paper or tutorial available on this issue? Thanks -- Julius
>>>>
>>>>
>>>>
>>> --
>>> Christian F. Beckmann
>>> Oxford University Centre for Functional
>>> Magnetic Resonance Imaging of the Brain,
>>> John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
>>> Email: [log in to unmask] -
>>> http://www.fmrib.ox.ac.uk/~beckmann/
>>> Phone: +44(0)1865 222782 Fax: +44(0)1865 222717
>>>
>>>
>>>
>> Richard Albistegui-DuBois
>> UCLA NeuroRehab, lab of Bruce Dobkin, MD.
>> Office: 1-132 Reed
>> Phone: 310-825-4016
>> Mobile: 310-774-1305
>> Fax: 310-794-9486
>> AIM: dubistegui
>> email: [log in to unmask]
>>
>
> --
> Christian F. Beckmann
> Oxford University Centre for Functional
> Magnetic Resonance Imaging of the Brain,
> John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
> Email: [log in to unmask] - http://www.fmrib.ox.ac.uk/~beckmann/
> Phone: +44(0)1865 222782 Fax: +44(0)1865 222717
>
>
Richard Albistegui-DuBois
UCLA NeuroRehab, lab of Bruce Dobkin, MD.
Office: 1-132 Reed
Phone: 310-825-4016
Mobile: 310-774-1305
Fax: 310-794-9486
AIM: dubistegui
email: [log in to unmask]
|