[Eeglablist] How to do data decomposition with ICA?

[Pascale Sandmann]

Dear Hannu,
Based on our experience with cochlear-implant (CI) artefacts, I can also
recommend you to do the ICA-based artefact reduction for all events at
once (i.e., to remove the CI-artefact components which are calculated for
all events of one dataset). In our last experiment (Sandmann et al., Clin
Neurophysiol 2010) we used 13 different events and we computed ICA for all
the events at once. Using this procedure, we could identify several
components representing the CI artefacts (typically around 8 components
out of 60 components) and CI artefacts could be successfully reduced.
Pascale


> Thanks for your anserw Scott
>
> By CI I ment cochlear implant.
>
> Hannu
>
> On Wed, Jul 28, 2010 at 8:02 PM, Scott Makeig <smakeig at gmail.com> wrote:
>
>> Hannu  -  The  'CI' ( = ??) artifact has the same scalp projection in
>> all
>> conditions, then ICA will separate it into a single component (assuming
>> its
>> activity is not linearly related to other, spatially-varying phenomena).
>> To most cleanly separate the artifact from the rest of the data, it is
>> best
>> to decompose all the data at once. Of course, if the other sources in
>> the
>> different experimental blocks are quite different (say, as an extreme
>> example, if the subject in the three blocks was respectively awake,
>> asleep,
>> and having an epileptic seizure), then separate decompositions followed
>> by
>> artifact component comparison and matching might be more effective.
>>
>> Note that ICA (instantaneous ICA, including infomax, runica, binica,
>> Amica)
>> does not pay attention to the time waveforms at all, just the collection
>> of
>> scalp maps for all the time points (in no particular order).
>>
>> Scott Makeig
>>
>> On Wed, Jul 28, 2010 at 12:09 AM, Hannu Loimo
>> <hannu.loimo at helsinki.fi>wrote:
>>
>>>
>>> Hi,
>>>
>>> We are using ICA to remove CI-artefact from our data. ICA seems to find
>>> pretty easily at least some of the artefact induced by CI-devise. There
>>> are
>>> 8 events in our experiment and two of those events (duration and gap
>>> deviants) produce different form of CI-artefact in time domain. All
>>> events
>>> are made out of syllable sequence ta-ta-ta. Gap deviant means that
>>> second
>>> syllable starts 100ms later than in other events. Duration deviant
>>> means
>>> that the whole ta-ta-ta sequence lasts 50ms longer. Presumably
>>> CI-artefacts
>>> rising from different events have same kind of spatial distribution and
>>> presumably most significant difference is how these artefact peaks are
>>> located in time. So basically there are three differend shapes of
>>> CI-artefact present in time domain: one for gap deviant, one for
>>> duration
>>> deviant and one for other six stimuli. After doing ICA for whole
>>> dataset
>>> (all events) CI-artefacts emerging from different events seem to be
>>> contained in one component and not many. My question is:
>>>
>>> Is it a problem to do ICA for whole dataset or shoud ICA be done for
>>> individual events separatelly? Size of dataset is 2000 epochs and
>>> separating
>>> events into different dataset would end up datasets as small as 150-200
>>> epochs (too few for ICA?). So far we have been doing ICA for whole
>>> datasets,
>>> then removed CI-artefact components and after that separated data into
>>> different events. What I'm asking is does it affect the final outcome
>>> of our
>>> separate event ERP composition if you remove CI-artefact components
>>> that are
>>> calculated according to all events or is it same if you do ICA for
>>> single
>>> events and then remove components?
>>>
>>> thanks,
>>>
>>> Hannu
>>>
>>>
>>> _______________________________________________
>>> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
>>> To unsubscribe, send an empty email to
>>> eeglablist-unsubscribe at sccn.ucsd.edu
>>> For digest mode, send an email with the subject "set digest mime" to
>>> eeglablist-request at sccn.ucsd.edu
>>>
>>
>>
>>
>> --
>> Scott Makeig, Research Scientist and Director, Swartz Center for
>> Computational Neuroscience, Institute for Neural Computation, University
>> of
>> California San Diego, La Jolla CA 92093-0961,
>> http://sccn.ucsd.edu/~scott<http://sccn.ucsd.edu/%7Escott>
>>
> _______________________________________________
> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
> To unsubscribe, send an empty email to
> eeglablist-unsubscribe at sccn.ucsd.edu
> For digest mode, send an email with the subject "set digest mime" to
> eeglablist-request at sccn.ucsd.edu


-- 
Dr. Pascale Sandmann
Department of Psychology
Neuropsychology Lab
Carl von Ossietzky University of Oldenburg
D-26111 Oldenburg
Germany

Office: A7 0-047a
Phone: +49-441-798-4945
Fax:   +49-441-798-5522
Email: pascale.sandmann at uni-oldenburg.de
http://www.psychologie.uni-oldenburg.de/46667.html