[Eeglablist] fastICA and spatial correlation

[Makoto Miyakoshi]

Dear Norman,

It's my turn to answer here in July. I'm an oldbie.

> My aim is to run ICA on a corrected EEG data set (data(1:64;1:30000))
using fastICA (I guess fastICA was the only function enabling me to adjust
the number of IC’s (?))  - computing only 10 components – in order to
remove eye movements.



This is wrong.

runica() default algorithm informax can be used with an option 'pca' which
is to reduce data dimension to an arbitrary number such as 10 (Note that
'pca' does NOT mean to run PCA instead of ICA!)


> The output I get from fastICA are: icasig (10 x 30000 matrix), A (63 x 10
matrix) and W (10 x 63 matrix). Now I want to reject components that
contain artifacts by finding a spatial correlation between the components
from fastICA results and a template (1x63 matrix) and if the max cross
correlation value is > 0.7 it should reject the components and reconstruct
the data.

My problem is that I have absolutely no idea how to start and proceed with
the matrices I have.

Through EEGLAB GUI, you can reject components while visually examining
their scalp maps, power spectral density, ERPimage, etc.
If you choose the first component to reject, it means you are rejecting the
first raw of EEG.icawinv (i.e., A). You can also perform regression with
the first IC's activation against whatever time series (ground truth of
blink, such as EOG channel activity I guess?) to test > 0.7 correlation to
make the decision--but you have to do this calculation outside EEGLAB.

Makoto


On Mon, Jul 2, 2018 at 2:10 PM Norman Sinnigen <
norman.sinnigen at student.uni-tuebingen.de> wrote:

> Hi,
>
>
>
> I am sorry if my question is weird or not 100% clear, but I am a newbie
> with EEGLAB and hope to get some hints.
>
>
>
> My aim is to run ICA on a corrected EEG data set (data(1:64;1:30000))
> using fastICA (I guess fastICA was the only function enabling me to adjust
> the number of IC’s (?))  - computing only 10 components – in order to
> remove eye movements.
>
>
>
> The output I get from fastICA are: icasig (10 x 30000 matrix), A (63 x 10
> matrix) and W (10 x 63 matrix). Now I want to reject components that
> contain artifacts by finding a spatial correlation between the components
> from fastICA results and a template (1x63 matrix) and if the max cross
> correlation value is > 0.7 it should reject the components and reconstruct
> the data.
>
>
>
> My problem is that I have absolutely no idea how to start and proceed with
> the matrices I have.
>
>
>
> Would someone have any hints for me?
>
>
>
> Norman
> _______________________________________________
> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
> To unsubscribe, send an empty email to
> eeglablist-unsubscribe at sccn.ucsd.edu
> For digest mode, send an email with the subject "set digest mime" to
> eeglablist-request at sccn.ucsd.edu



-- 
Makoto Miyakoshi
Swartz Center for Computational Neuroscience
Institute for Neural Computation, University of California San Diego
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20180702/0d4a5bdb/attachment.html>