[Eeglablist] How to use runica for MEG data

Michael Zvyagintsev Mikhail.Zvyagintsev at med.uni-tuebingen.de
Wed Mar 23 06:13:21 PST 2005

Dear EEGLAB users and developers!
I have some questions how to perform ICA analysis on MEG-data by runica 
procedure. Below I try to describe the way which I am trying to use to 
do it...
I load my data (with our own domestic routines). So I get it in 
fentoTesla(it seems important to me, becouse otherwise, this 'runica' 
doesn't work at all!). Then I do baseline correction, then I separate my 
data across conditions. Let's say as result I get matrix like
151channels x 400samples x 100trials - for one condition.
I am trying to put these data to 'runica' procedure. For this purpose, I 
took just matrix like
151channel x (250:300)samples*100trials =
     151 x 5000.
Then it is necessary to adjust some parameters inside of 'runica' (I am 
not sure, but I think so!).
DEFAULT_SPHEREFLAG = 'none' - this one is imortant for 'Perform 
sphering', and only if it is ='none', variable sphere will contain 
plausible numbers, otherwise I always get NaN in it.
Then I run this procedure, actually as a script - to control averything 
what is going on (but it doesn't help!).
And then I get (how it seems to me) three most important variables - 
weghts, sphere, activations (I can also get 'projections', how it points 
in EEGLAB FAQ, but Should I?...).
The main question is - is this way correct? or it is completely wrong 
approach? And which matrix reflect ICA-components, I mean which matrix 
should I visualize with plot utilities to get some kind of 'topoplot' of 
  every component?
Could you help me with solving these questions?...
Have a nice day!
Thank you,

Mikhail Zvyagintsev, MD, Bachelor of Informatic
Universitätsklinikum Tübingen
Zentrum für Neurologie
MEG-Zentrum, Otfried-Müller-Straße 47
72076 Tübingen, Germany
tel +49-7071-29-87717 fax +49-7071-29-5706
Mikhail.Zvyagintsev at Med.Uni-Tuebingen.de

More information about the eeglablist mailing list