ICA running

Discussions
1

Hi!

ICA running is very slow in some of my files. usually it takes about 5 minutes but in some cases it takes more than 10 hours while they are equal in size!
would you please guide me?

Thanks
Shadi

2

Hi Shadi,
I’m sorry, we have very little experience with ICA, I don’t know how to guide you with this issue.
You should refer to the EEG/ICA literature or ask on the EEGLAB email list, the Infomax/runica and Jade options in Brainstorm are directly the EEGLAB implementations.
Francois

3

Hi Shadi:

Could these variations in computation time be due to files of different lengths/sampling rates/number of channels? Also, please note ICA decompositions require sophisticated computations that can be accelerated by indicating to the ICA algorithm that you want to perform a PCA - to reduce the dimensions in the data - before completing the ICA. This is usually a parameter to indicate prior to launching the ICA decomposition.

Hope this helps.

4

Dear Sylvian

Thanks for your response!

the file characteristics are the same.
I think I could find the reason.
I have 6 parts of EEG data (of 6 runs) in each subject and I prefer to run ICA for each of these parts separately. but in each file the ICA would consider the last components computed in last runs. for example I will have two ICA projections when I want to start runing ICA for the 3dr file. and when I try to remove them rom one file I will lose them on all other files. do you have any way to compute ICA for each part separately?

5

Hi Shadi,
All the files that share the same channel file will share the same SSP/ICA projectors.
You can configure your subjects to use one separate channel file for each acquisition run: double-click on the subject and select the option “Channel file: No, use one channel file per run/folder”.
Francois

6

[QUOTE=Francois;9743]Hi Shadi,
All the files that share the same channel file will share the same SSP/ICA projectors.
You can configure your subjects to use one separate channel file for each acquisition run: double-click on the subject and select the option “Channel file: No, use one channel file per run/folder”.
Francois[/QUOTE]

Hi Francois

Thank you
and
Mary Christmas

Shadi

7

Hi again!

as I read previous posts, It seems that brainstorm changes the montage to common average automatically to do source estimation and ICA running. I have recorded my data on linked ear montage and I want to keep this montage for my analysis. what do you suggest? running ICA with average montage and re referencing the data to A1,A2 after removing bad components?

Best
Shadi

8

Hi Shadi,

You do not have to re-reference your data to an average reference before the ICA decomposition. If you want to do so, you have to use the menu “Artifacts > Re-reference EEG” in the Record tab. This process doesn’t let you use a linked ref montage yet, this will be enabled in the future. If you don’t create a projector to re-reference your data, the ICA will use the original signals.

The montage you use for the display of the signals (created in the window “Edit montage”) is not applied to the recordings, it has no effect on the ICA or SSP processes.

You don’t have to worry about the reference for the source estimation. See the following post:
http://neuroimage.usc.edu/forums/showthread.php?1525#post6718

Cheers,
Francois

9

in fact I'm not worried about the ICA decomposition process. It seems that the montage is automatically changed to averaged_ref. for this process.

No, it's not.

in the tutorial I could see that it is possible to change the the reference to A1,A2 in the re-referencing tab. do you mean that it doesn't work yet?

The montage selected in the Record tab and edited with the menu "Edit montages" are for visualization only. They are not applied to the recordings.
If you select a montage and run any process, the selected montage is ignored.
To apply a montage on the recordings, you need to run the process "Re-reference EEG". After that, you should see the re-referencing operator in the window "Select active projectors".

is this the linked Ear montage or this is another kind of referencing?

?

and if yes, is it possible to convert EEG montage from averaged ref. to linked ear?

In Brainstorm, you can easily convert recordings from single reference to linked or averaged reference, but not the contrary.
So no, you cannot easily convert from averaged ref to linked ref with the current interface. You can probably do this manually if needed.
But do you really need it?