[FieldTrip] Reject components on unfiltered data

[Anne Klepp]

Dear all,

I will jump on the bandwagon here to ask what exactly the difference is 
between using the third input argument or not. With my data, using the 
data from the step right before ICA (so not actual raw/unfiltered data) 
as the third argument changes the output data.grad, but also changes the 
data itself. (This data is from after automatic artifact rejection, 
preprocessed including lp- and hp-filters, and has some repaired 
channels. Data is from a Neuromag306, with the gradiometers rejected 
during preprocessing, so there are 204 MEG channels plus 3 EOG/EMG.)

I tried using fastica as well as specifying cfg.runica.pca=100 with 
runica. 2 components were rejected.
To clarify what I mean by my data being changed I attached some files of 
what data look like in rejectvisual for all channels and in two example 
channels, for both ICA methods.

I'm afraid I don't quite understand the consequences for my data by the 
information given in the FT help "In that case (3 arguments)  
componentanalysis will do a subspace projection of the input data   onto 
the space which is spanned by the topographies in the unmixing matrix in 
comp, after removal of the artifact components.   Please use  this 
option of including data as input, if you wish to use the output   
data.grad in further computation, for example for leadfield 
computation." I suppose it makes sense data are different after 
(subspace) back projection, but what are the implications? If I don't 
want to do sourceanalysis, is there an advantage of sticking with 2 
input arguments?

I'd be thrilled if someone could offer some insight into this,

best,
Anne


Am 25.10.2013 16:31, schrieb Eelke Spaak:
> Dear Carmen,
>
> Yes, that is possible; if you specify your original unfiltered data as
> the third input argument to ft_rejectcomponent this data will be
> projected through the unmixing matrix (and remixed after component
> rejection).
>
> Is that detailed enough to help you?
>
> Best,
> Eelke
>
> On 25 October 2013 16:18, Carmen Kung <c.kung at donders.ru.nl> wrote:
>> Dear all,
>>
>> I wonder if it is possible to first filter the data with a bandpass filter
>> between 0.5 hz and 30 hz (suggested by the ICA tutorial of eeglab), run the
>> ICA on filtered data to look for the components containing artefacts, then
>> use the projection to reject the components in the original (unfiltered)
>> data.
>>
>> Thank you very much in advance,
>> Carmen
>>
>>
>> --
>> Carmen Kung
>> Donders Institute for Brain, Cognition, and Behaviour
>> Radboud University Nijmegen
>> P. O. Box 9104, 6500 HE Nijmegen, The Netherlands
>> Tel: +31 24 36 12634
>>
>> _______________________________________________
>> 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


-- 
Anne Klepp (M.Sc. Psychologie)
Institute of Clinical Neuroscience and Medical Psychology
Heinrich-Heine-University Düsseldorf
Universitätsstraße 1
40225 Düsseldorf, Germany
Email: anne.klepp at hhu.de

-------------- next part --------------
A non-text attachment was scrubbed...
Name: ICA_data_2or3args.7z
Type: application/octet-stream
Size: 378658 bytes
Desc: not available
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20131030/59cf130a/attachment.obj>