[Eeglablist] Details of analysing data

[Makoto Miyakoshi]

Dear Fang-Yu,

> First,
If I would like to apply ICA metrics which is from high-pass at 1 hz to
high-pass at 0.01hz dataset, should the dataset be same condition? In my
situation, before running ICA, should removed bad channels be the same for
both 1 Hz and 0.01 Hz high-pass filtered datasets.

Make sure that the datasets between which ICA weight matrices are passed
have same channels.

> Second,
About creating STUDY. By "Makoto's preprocessing pipeline" reads "Importantly,
creating STUDY means you clean your data, because you can 1) Exclude
dipoles with > 15% residual variance (Artoni et al., 2014), and 2) outside
brain. Don't forget that this is very powerful cleaning, and this is why
you don't need to manually reject ICs at the individual level. Don't forget
to set STUDY.design before you start precompute." My question is can eye
blinks be removed as well during this step?

No, they often slip into the final clusters (EMGs too). Thus I made another
plugin to clean them. http://sccn.ucsd.edu/wiki/Std_selectICsByCluster I
should integrate description about it in the pipeline page.

> Third,
I applied 1 Hz high-pass filter, imported channel location, removed bad
channels, did re-reference to average (discard 1 channel) and then ran ICA.
However, there was a message showing a rank-deficient in the dataset.
Should I continue anyways? Or are there any other ways to solve this kind
of problem?

The most advanced understanding for me is to add zero-filled channel to
end+1 and perform average reference. This is described in my pipeline
website. If it is technically difficult for you... ok follow the
instruction below. Average referencing without including the initial
reference channel (i.e. zero-filled channel) reduces rank by one because A
+ B + C + ... = 0 means the left hand terms are rank deficient by
definition of linear algebra. If you don't mind discarding one channel, you
may do to to make the data full rank. Alternatively, you can use pca option
to reduce the rank by 1 when running runica().

Makoto

On Mon, Jun 13, 2016 at 7:07 AM, Fang-Yu Chang <hardheard101 at gmail.com>
wrote:

> Dear EEGLABlisters,
>
> I would like to ask three questions.
>
> First,
> If I would like to apply ICA metrics which is from high-pass at 1 hz to
> high-pass at 0.01hz dataset, should the dataset be same condition? In my
> situation, before running ICA, should removed bad channels be the same for
> both 1 Hz and 0.01 Hz high-pass filtered datasets.
>
> Second,
> About creating STUDY. By "Makoto's preprocessing pipeline" reads "Importantly,
> creating STUDY means you clean your data, because you can 1) Exclude
> dipoles with > 15% residual variance (Artoni et al., 2014), and 2) outside
> brain. Don't forget that this is very powerful cleaning, and this is why
> you don't need to manually reject ICs at the individual level. Don't forget
> to set STUDY.design before you start precompute." My question is can eye
> blinks be removed as well during this step?
>
> Third,
> I applied 1 Hz high-pass filter, imported channel location, removed bad
> channels, did re-reference to average (discard 1 channel) and then ran ICA.
> However, there was a message showing a rank-deficient in the dataset.
> Should I continue anyways? Or are there any other ways to solve this kind
> of problem?
>
> Thanks your help in advance.
>
> Sincerely,
> Fang-Yu Chang
>
> _______________________________________________
> 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/20160701/75fbda5c/attachment.html>