[Eeglablist] eeglablist question

[Tarik S Bel-Bahar]

>
>
> Hello Bert,  notes below.
>

**************************************

> yes generally speaking that's what the rule of thumb for number of
> datapoints for ICA is about. Basically it has a big stomach, and in order
> to derive valid spatial ICA components, it needs enough "information". When
> one does not have enough time-points, downsampling helps, as well as, in
> some cases, reducing total number of channels. However, I've seen
> valid/neural/dipolar ICs from setups similar to the one you describe,
> with only a few minutes of data and 64 or greater channels.
>
> *generally speaking, it is recommended to retain only the relevant
> periods where participants are doing the intended task (and eliminate
> periods before, after, and between such tasks). A period where "nothing is
> happening" and "participant is just waiting" should be removed. Periods
> that are just a few seconds between trials, and which are artifact free,
> can remain. However, if you want to see for yourself, review your ICA
> solution when you A)  leave in all your data (but without the
> artifact/noisy periods), and compare it to the ICA solution when having
> removed all the periods that are not-of-interest (where the participant is
> not engaged in a particular task you have required of them). Note that
> "resting" periods, for example, a few minutes of eyes closed resting, are
> considered "tasks".
>
> *make sure to have removed the "bad periods" so that ICA doesn't get
> confused by them, as per the eeglab tutorials (if you're new to ICA-EEG,
> look at results with and without such cleaning to get a feel for why basic
> cleaning is important for ICA).
>
> *if you haven't had a chance to, review the basic "kinds of valid/neural
> ICs" that can show up, as described in various recent articles you can
> find  via Google Scholar from Makeig/Delorme/Artoni, and also in various
> articles about IC selection methods/plugins (SASICA, ADJUST, MARA, etc..).
>  also see check out tutorial on ICA types
> https://labeling.ucsd.edu/tutorial
>

> *run ICA anyway on a shorter period (you won't break anything) and see if
> you get interpretable and expected/normal neural ICs. Note that many existing
> publications have used quite short periods that don't fit under this
> (ideal) rule about "enough timepoints". However, the validity of their
> ICs should be considered with caution.
>
> *google scholar articles that use ICA with short resting/continuous
> periods, or that use alternative techniques for source decomposition (e.g.,
> variants of ICA, or even PCA -though PCA is not recommended by eeglab
> developers)
>
> *google eeglablist + your topic for previous eeglablist comments related
> to your topics
>
> *if you haven't had a chance to, try running ICA with data prepped in
> different ways to find out for yourself how the ICA results change based on different
> kinds of data prepping
> ********************************
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
> On Thu, Nov 29, 2018 at 2:42 AM Bert Vandenberghe <
> b.vandenberghe at kuleuven.be> wrote:
>
>> Hello
>>
>>
>>
>> For ICA decomposition, I read on your wikipage that you need at least
>> (#channels^2*30) data points to perform ICA on epoched data.
>>
>> If not, ICA on continuous EEG is preferable.
>>
>>
>>
>> In my case, I use 64 electrodes.
>>
>> Does that mean that 4 minutes of epoched data would suffice in order to
>> perform ICA ?
>>
>>
>>
>> I.e., 64^2*30 = 122880 data points
>>
>>
>>
>> If you sample at 512 Hz:
>>
>>
>>
>> 512*240seconds=122880 data points
>>
>>
>>
>> Is this a correct way of reasoning for deciding to perform ICA
>> decomposition on epoched data ?
>>
>>
>>
>> Thank you for your advice
>>
>>
>>
>> Bert
>> _______________________________________________
>> 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
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20181201/f06644f6/attachment.html>