[FieldTrip] ICA - Can I visualize Activity power spectrum of single components?

Max Cantor mcantor at umich.edu
Wed May 28 16:24:38 CEST 2014


What we usually do before ICA is a rough pass removing the most extremely
artifactual trials or channels. For instance, if you see artifacts that are
not systematic and are affecting multiple sensors, it may be something like
a head scratch that is affecting multiple sensors independently, and so
would be turned into multiple ICA components, and so removing these would
be beneficial. That being said, anything that might be, for instance, an
ECG or EOG artifact, is best not cleaned before ICA, because the whole
point of ICA components is to find artifacts that are independent from each
other, but dependent to the source (in other words, a systematic artifact
as the result of a blink) and remove them without having to throw away the
rest of the good data within that trial/channel. In other words, if you're
doing a comprehensive artifact rejection before ICA, it makes sense that
you aren't necessarily picking up artifactual components, since you might
already be removing the data that would be turned into an artifactual
component.


On Wed, May 28, 2014 at 9:54 AM, KatrinH Heimann <katrinheimann at gmail.com>wrote:

> Diego, can I ask you another advice- this time more conceptually? I am
> playing with the ICA now and experience, that I get very different
> components depending on the artifact detection that I do beforehand. So my
> idea (due to the literature) was, that the ICA works better on already
> cleaned data. Now, as I want to do a frequency analysis afterwards, I did
> not want to reject parts of trials. Therefore I deleted a relatively big
> amount of trials in which I found muscle artifacts/drifts (Say 25% of the
> whole amount of trials). However, in fact I see that the ICA then does not
> give me easily to interpret components anymore - actually the results are
> better (that is easier to interpret which is artifact, which not) if I
> include all trials. If I only reject parts of the trials it gets even a bit
> better. So my question is: Shell I a) do the (stricter) artifact detection
> rather after the ICA or b) perform it on data with parts of trials marked
> as bad - in which case I also ask: can I then still reject those components
> in the uncleaned data (saved beforehand) so that I do not get problems with
> the frequency analysis afterwards...?!
> Thanks a lot for your help!!! Katrin
>
>
> 2014-05-28 15:02 GMT+02:00 KatrinH Heimann <katrinheimann at gmail.com>:
>
> Wonderfull! Thanks Diego!!!!
>> Cheers k
>>
>>
>> 2014-05-28 14:24 GMT+02:00 Lozano Soldevilla, D. (Diego) <
>> d.lozanosoldevilla at fcdonders.ru.nl>:
>>
>> Hi Katrin,
>>>
>>> First, to display the independent components:
>>> %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
>>> cfg = [];
>>> cfg.layout = 'CTF151.lay'; % specify the layout file that should be used
>>> for plotting
>>> cfg.viewmode = 'component'
>>> ft_databrowser(cfg, comp)
>>>
>>> Take a look to the following example script:
>>>
>>> http://fieldtrip.fcdonders.nl/example/use_independent_component_analysis_ica_to_remove_eog_artifacts
>>>
>>> Second, you'll have to compute the power spectrum for each independent
>>> component with ft_freqanalaysis:
>>>
>>> %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
>>> cfg              = [];
>>> cfg.output       = 'pow';
>>> cfg.channel      = 'all';%compute the power spectrum in all ICs
>>> cfg.method       = 'mtmfft';
>>> cfg.taper        = 'hanning';
>>> cfg.foi          = 2:2:30;
>>> freq = ft_freqanalysis(cfg, comp);
>>>
>>> And you can plot the spectra:
>>>
>>> %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
>>> nsubplots = 25;
>>> nbyn = sqrt(nsubplots);% sqrt(nsubplots) should not contain decimals,
>>> type doc subplot
>>>
>>> Nfigs = ceil(size(comp.topo,1)/nsubplots);
>>> tot = Nfigs*nsubplots;
>>>
>>> rptvect = 1:size(comp.topo,1);
>>> rptvect = padarray(rptvect, [0 tot-size(comp.topo,1)], 0,'post');
>>> rptvect = reshape(rptvect,nsubplots,Nfigs)';
>>>
>>> for r=1:size(rptvect,1);
>>>   figure;set(gcf,'units','normalized','outerposition',[0 0 1 1]);%full
>>> screen
>>>   k=0;
>>>   for j=1:size(rptvect,2);
>>>     if~(rptvect(r,j)==0);
>>>       k=k+1;
>>>       cfg=[];
>>>       cfg.channel = rptvect(r,j);
>>>       subplot(nbyn,nbyn,k);ft_singleplotER(cfg,freq);
>>>     end
>>>   end
>>> end
>>>
>>> For the IC topos you'll follow the same logic as above but with:
>>>
>>> figure
>>> cfg = [];
>>> cfg.component = [1:20];       % specify the component(s) that should be
>>> plotted
>>> cfg.layout    = 'GSN-HydroCel-129.sfp';
>>> cfg.comment   = 'no';
>>> ft_topoplotIC(cfg, comp)
>>>
>>> I hope it helps
>>>
>>> Diego
>>> ------------------------------
>>>
>>> *From: *"KatrinH Heimann" <katrinheimann at gmail.com>
>>> *To: *"FieldTrip discussion list" <fieldtrip at science.ru.nl>
>>> *Sent: *Wednesday, 28 May, 2014 12:59:38 PM
>>> *Subject: *[FieldTrip] ICA - Can I visualize Activity power spectrum of
>>> single        components?
>>>
>>>
>>> Dear all,
>>> another question:
>>> Is it possible to visualize the Activity power spectrum of the single
>>> components after an ICA by using ft_databrowser (or something else). I like
>>> this feature from EEG lab as I find it easier to detect artifactual
>>> components.
>>> Thanks a lot for your help
>>> Katrin
>>>
>>>
>>> _______________________________________________
>>> fieldtrip mailing list
>>> fieldtrip at donders.ru.nl
>>> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>>>
>>>
>>>
>>>
>>> --
>>> PhD Student
>>> Neuronal Oscillations Group
>>> Donders Institute for Brain, Cognition and Behaviour
>>> Centre for Cognitive Neuroimaging
>>> Radboud University Nijmegen
>>> NL-6525 EN Nijmegen
>>> The Netherlands
>>> http://www.ru.nl/people/donders/lozano-soldevilla-d/
>>>
>>> _______________________________________________
>>> fieldtrip mailing list
>>> fieldtrip at donders.ru.nl
>>> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>>>
>>
>>
>
> _______________________________________________
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>



-- 
Max Cantor
Research Assistant
Computational Neurolinguistics Lab
University of Michigan
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20140528/b431938b/attachment.html>


More information about the fieldtrip mailing list