[Eeglablist] Lower Number of ICs than others

[Makoto Miyakoshi]

Dear Hamed,

> So, based on your experience the number of ICs (between 15 to 20 per
subject) could be rational.

Yes, empirically speaking the number seems valid.
It's really too bad that we don't have a publication to justify it...

> Also, by changing the high pass filter from1 Hz to 2 Hz the number of ICs
increased in some subjects by 10% or even 30%.

Yes, I'm glad to hear that consistent result from you. I described the
topic in 'Makoto's preprocessing pipeline'. Fortunately, there is a paper
to justify it.

Makoto

On Tue, Jul 10, 2018 at 3:22 PM Hamed Taheri <hamedtaheri at yahoo.com> wrote:

> Dear Makoto,
>
> Thanks a lot for your reply and detailed information.
> So, based on your experience the number of ICs (between 15 to 20 per
> subject) could be rational.
> Also, by changing the high pass filter from1 Hz to 2 Hz the number of ICs
> increased in some subjects by 10% or even 30%.
>
> Hamed
>
>
>
>
>
>
> On Tuesday, July 10, 2018, 1:42:48 AM GMT+2, Makoto Miyakoshi <
> mmiyakoshi at ucsd.edu> wrote:
>
>
> Dear Hamed,
>
> 20 x 64 = 1280 ICs you had in the beginning, but you have in the end 305.
> 305/20 = 15 ICs per subject, which does not seem very strange for me.
>
> There are several empirical facts.
>
>    1. More channels ~= More ICs.
>    2. Number of ICs to reject ~= amount of variance to reject
>
> About 1, regardless of the number of channels you have, to certain extent,
> the final number of 'brain ICs' are consistent to 10-20 per subject. There
> is no dedicated study on this (I know we SCCN are responsible for
> investigating this kind of ICA problems, sorry!), but empirically this is
> the case. I even think that the true degrees of freedom on scalp EEG
> recording is limited from the beginning.
>
> About 2, ICs are sorted by variance (i.e., power). Even if you reject
> 80/100 ICs, if the remaining ICs are the first 10 ICs, namely IC1, IC2,
> ..., IC10, you probably keep 80-90% of original variance. If you want to
> report exactly how much data rejection was performed, the variance rejected
> should be reported (which you can't do from GUI, sorry... but what you need
> to do is subtract EEG.data before and after IC rejection, and compute
> 1-100*mean(var(data_beforeRejection)-var(data_afterRejection))/mean(var(data_beforeRejection))
> see evntopo() function for this calculation, or Lee et al. 2014 IEEE conf
> proc) rather than the number of ICs since the latter is poor estimator of
> the former. A single blink IC may even account for 30-40% of variance
> easily!
>
> Makoto
>
>
> On Mon, Jul 9, 2018 at 11:10 AM Hamed Taheri <hamedtaheri at yahoo.com>
> wrote:
>
> Hello all,
>
> I have EEG data from 20 subjects which were recorded using 64 channels EEG
> cap. I've used Mokoto's preprocessing pipeline. When I make a STUDY and
> remove the dipoles with < 15% residual variance and the dipoles outside
> the brain, the number of remained ICs compared to other studies are very
> lower.
> For instance, a Paper with 21 Subjects and the same recording system (64
> Channels) reached 897 ICs while my ICs are just 305.
> Based on your experience do you think what is my mistake?
> Do you think it can be related to the quality of my recording system
> because our cap is a bit old or I have a problem with my processing?
> I would be so grateful if you could help me with this issue.
>
> Best Regards,
> Hamed
> _______________________________________________
> 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
>


-- 
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/20180711/538c5755/attachment.html>