[Eeglablist] ICA question

[Makoto Miyakoshi]

Dear Jumana,

> For data rank reasons, I interpolated and average referenced after ICA,
using runica. When re-referencing I deleted the ICA activation matrix and
the ICA weights, because of course I now have interpolated channels which
do not have ICA weights associated with them.

Average referencing is just to subtract a fixed value from all
channels/ICs, so I thought it is harmless. You don't need to delete
ICA-calculated items.

> I only wanted to do ICA for eye blink rejection, and it worked perfectly
and I did not have any ICA corruption (trade offs or rank deficiency). I
compared the data before and after ICA and the level of noise was the same.

Do you mean that if you don't reject ICA-calculated items, it destroys
data? I'm interested in testing it. It's more likely it does not happen. If
you say you reject channels after ICA, it's more likely to result in more
complicated situation, depending on the path you follow using EEGLAB.

By the way what do you mean trade-off?

> I am working with an extremely large dataset and do not ideally want to
re-do these steps, especially because I am mainly working with raw channel
ERP data.

It's a good opportunity for you to start using batch code. If you follow my
instructions in my wiki pages, 100-200 datasets are nothing. I created a
STUDY with nearly 1,000 datasets with no problem, so I can guarantee up to
1,000. If you eventually want to go back to channels, then after ICA
selections using std_selectICsByCluster(), perform channel statistics. In
this case, I think interpolation will definitely help you because it'll
eliminate missing value problems reasonably.

> this time on the Chanel data where ICA blink components have been removed

Even if you run ICA on IC-rejected clean data, you won't obtain cleaner
data. What happens is that you'll get exactly the same decomposition. Try
it with one subject to see it.

> Identify noisy channels
> Run ICA on clean channels
> Interpolate electrodes to give the same 60 electrodes per person
> Average reference to the 60 electrodes
> Compute ERPs on channel data
> Then re-run ICA, and do any further analysis on component data.

Yeah it works. Most likely, the last ICA will produce the same results
except post average-reference components show zero-mean scalp topos.

Makoto




On Fri, Jan 13, 2017 at 6:20 AM, Ahmad, Jumana <jumana.ahmad at kcl.ac.uk>
wrote:

> Dear EEGlab.
>
> For data rank reasons, I interpolated and average referenced after ICA,
> using runica. When re-referencing I deleted the ICA activation matrix and
> the ICA weights, because of course I now have interpolated channels which
> do not have ICA weights associated with them.
>
>
> I only wanted to do ICA for eye blink rejection, and it worked perfectly
> and I did not have any ICA corruption (trade offs or rank deficiency). I
> compared the data before and after ICA and the level of noise was the same.
>
>
> I am working with an extremely large dataset and do not ideally want to
> re-do these steps, especially because I am mainly working with raw channel
> ERP data.
>
>
> However, for future analysis, if I did want to work with component data,
> could I re-run ICA for a second time, this time on the Chanel data where
> ICA blink components have been removed, and where the interpolated
> electrodes and the new reference implemented (this data should already be
> clean from blinks etc).
>
>
> Overal this would look like:
>
>
> Identify noisy channels
>
> Run ICA on clean channels
>
> Interpolate electrodes to give the same 60 electrodes per person
>
> Average reference to the 60 electrodes
>
> Compute ERPs on channel data
>
> Then re-run ICA, and do any further analysis on component data.
>
>
>
> Please let me know if you can see any issues with this. I very much
> appreciate any advice.
>
> Best wishes,
>
> Jumana
>
>
> *------------------------------------------*
> *Jumana Ahmad*
> Post-Doctoral Research Worker in Cognitive Neuroscience
> *EU-AIMS Longitudinal European Autism Project (LEAP) & SynaG Study*
> Room M1.26.Department of Forensic and Neurodevelopmental Sciences (PO 23)
> | Institute of Psychiatry, Psychology & Neuroscience | King’s College
> London | 16 De Crespigny Park | London SE5 8AF
>
> *Phone:* 0207 848 5359| *Email:* jumana.ahmad at kcl.ac.uk
> <antonia.sanjose at kcl.ac.uk> | *Website:* www.eu-aims.eu | *Facebook:*
> www.facebook.com/euaims
>
>
> _______________________________________________
> 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/20170121/7ecc1a9a/attachment.html>