[FieldTrip] ICA on MEG data

[Stan van Pelt]

Dear Irina,

The number of time points should indeed be much larger for a good ICA
estimate. However, that seems to be the case with your data, so I think
there's no problem there. 

As far as I know, with ICA in fieldtrip you always get the same amount of
output components as input channels. Activity in adjacent MEG channels
will be somewhat correlated indeed, but not nearly as much as with EEG.
Moreover, the origin of your artifacts (EOG, maybe also ECG) will create a
distinct topographic distribution, and the components will pop out clearly
from your ft_componentanalysis, because it explains large amounts of
variance in the data.

Just give it a try, I'd say. You should get similar results as in the Wiki
example.

Best,
Stan


-----Original Message-----
From: fieldtrip-bounces at donders.ru.nl
[mailto:fieldtrip-bounces at donders.ru.nl] On Behalf Of Irina Simanova
Sent: Tuesday, August 30, 2011 9:56 AM
To: Email discussion list for the FieldTrip project
Subject: Re: [FieldTrip] ICA on MEG data

Dear Stan,

Thank you for your reply,

As far as I understand from the tutorials about ICA by Scott Makeig
http://sccn.ucsd.edu/~scott/tutorial/icafaq.html 
, the number of time points should be considerably larger than the  
number of scalp sensors. How critical is this point?

Besides, the sensors are very close to each other, and the registered  
activity in adjacent sensors is highly correlated, so the components  
would be similar, right? Isn't it reasonable to reduce the number of  
output components then?

Thank you in advance,
Irina



On Aug 30, 2011, at 9:14 AM 8/30/11, Stan van Pelt wrote:

> Dear Irina,
>
> There's a page on the Fieldtrip Wiki about using ICA to remove
> EOG-artifacts, see
>
http://fieldtrip.fcdonders.nl/example/use_independent_component_analysis_i
> ca_to_remove_eog_artifacts.
>
> The common approach is to first define your trials, and then run ICA  
> on
> them.
>
> You will get as many output components as you have channels, i.e.  
> 275 in
> your case. The EOG artifacts usually pop up in 1 or 2 components  
> within
> the first 20 largest components.
>
> Best regards,
> Stan
>
> Stan van Pelt, PhD
> Donders Institute for Brain, Cognition and Behaviour, Radboud  
> University
> Nijmegen
> Kapittelweg 29, 6525 EN Nijmegen, Netherlands
> E-mail:  stan.vanpelt at donders.ru.nl
> Website: www.ru.nl/donders/
> Phone:  (+31) (0)24 36 68495
> Fax:    (+31) (0)24 36 10989
>
> -----Original Message-----
> From: fieldtrip-bounces at donders.ru.nl
> [mailto:fieldtrip-bounces at donders.ru.nl] On Behalf Of Irina Simanova
> Sent: Tuesday, August 30, 2011 8:03 AM
> To: Email discussion list for the FieldTrip project
> Subject: [FieldTrip] ICA on MEG data
>
> Dear FieldTripers,
>
> I would like to identify ocular artifacts in an MEG dataset using the
> ICA.
>
> The data consist of 180 trials of variable duration ( the signal from
> stimulus onset to subject's response). The trials' duration vary from
> 400 ms to 3000 ms. The data is sampled at 1.2 kHz, and has 275
> channels. I am new to ICA, and I need a piece of advice. What would be
> a reasonable number of output independent components?
> Should I do the PCA as a pre-processing stage? How do i find an
> optimal number of principal components to extract?
>
> I would appreciate your help,
>
> Kind regards,
>
> Irina Simanova
> PhD student
> Neurobiology of Language Group
> Max-Planck Institute for Psycholinguistics
> Wundtlaan 1
> 6525 XD Nijmegen
> The Netherlands
> e-mail: irina.simanova at mpi.nl
> phone: +31 24 3521541
>
> _______________________________________________
> 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

_______________________________________________
fieldtrip mailing list
fieldtrip at donders.ru.nl
http://mailman.science.ru.nl/mailman/listinfo/fieldtrip