[Eeglablist] resting-state artifact rejection after ICA

[Tarik S Bel-Bahar]

Hello Shira, comments below, best wishes.

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

1. 1 hz higpass is usually recommended for great ICA decompositions, and is
usually not necessary to be lower than 1hz for resting-state data, unless
you have specific hypotheses <1hz. Further, after getting a good ICA
decomposition, you can apply those ICs to continuous data that has been
filtered with a lower highpass filter, and then remove some bad ICs, and
then epoch, and then drop epochs that still have clear artifacts.

2. One should reject bad data before ICA. You can leave blinks and muscle
noise in, but big movements, multi-channel artifacts, etc.. should be
removed. See the pre-ICA eeglab tutorials, Makotos' eeglab pipeline
suggestions, and past notes on the eeglablist on this topic. Then, after
ICA, one can epoch, or apply the ICA to the original continuous data, and
clean from there via multiple methods.

3. regarding the hunt for "right criteria". Try playing around with the
various artifact-detection and measures, and don't just use only the
threshold measure. Ideally you can use a combination of eeglab settings,
and have different artifact measures converge on particular epochs to drop.
If you do this in eeglab, you will see some epochs marked with multiple
colors, which signifies that multiple measures are catching that epoch. If
you're running artifact measures after ICA, one can also choose to use the
ICA-based artifactual epoch rejection measures rather than the
channel-based ones. Overall, a good rule of thumb is that you should be
able to keep at least 50% of the recorded data. If not, there is something
wrong with the artifact-detection and/or with the general quality of the
eeg data.

4. See also the "reject continuous" data tool in the eeglab gui, which
works quite well and works at the frequency level, looking for unique noise
in several frequency ranges.  Another option is to use the settings you see
in a published articles that uses a similar paradigm and recording system.
There are also other options within eeglab, such as Kothe's clean data
tools available in the gui and the PREP toolbox. See also TAPEEG and other
solutions for automatic cleaning of data.

5. EEG cleaning is still an art, and it depends on how clean your data is
to begin with, how many channels you have, and what tools you are using to
find bad periods. Since your data is not very long I recommend visual
cleaning by human expert, unless you have hundreds or thousands of sessions
to deal with.

6. If you are a beginner with eeg or with cleaning EEG data, one good thing
to do is have an expert show you what really clean / good data looks like,
and this can be a reference for further cleaning attempts. Another option
is to identify the cleanest time periods within one subject and use that
information to clean the rest of the subject's data. EEG data often varies
in quality across subjects and across recording sessions.










On Sat, Aug 5, 2017 at 10:49 AM, shira frances <shirafrn at gmail.com> wrote:

> hello all,
> i am analyizing 5 min closed-eyes resting-state data (bdf file) 32
> electrodes. i am intersted on the alpha power band.   i read old mails and
> answers on eeglablist and found (thank you!!!) the script for frequncy
> analysis (https://sccn.ucsd.edu/wiki/Makoto%27s_useful_EEGLAB_code#
> How_to_extract_EEG_power_of_frequency_bands .)
> the plan i am trying to do now is this: import the raw data, define
> channel location, reference to CZ, FIR high-pass 0.5, lowpass 45, import
> events(from a file i created, an event evey second, only for the epoching,
> the events has no meaning), after that i run the ICA and removing what
> looks like eye movement (the oarticipants moved their eyes altough it was
> eyes-closed condition) , then i epoch the data to 1s epochs (from -500ms to
> 500ms). now i want to reject artifacts by epochs, i used like i read - "all
> methods" and then reject by the first option (abnormal values) - i wrote
> +-100. the problem is it reject about half of the data... i tried it again
> this time with +-150 and it rejected 13% - is this the right criteria for
> rejection? (i was instructed to try this because we want to see alpha)
> may i use other criteria?
> and generally how can i know what is the right criteria?
>
> thank you vert much
>
> shira frances.
> tel-aviv university
>
>
>
> <http://www.avg.com/email-signature?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=webmail> Virus-free.
> www.avg.com
> <http://www.avg.com/email-signature?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=webmail>
> <#m_3007892038193664287_DAB4FAD8-2DD7-40BB-A1B8-4E2AA1F9FDF2>
>
> _______________________________________________
> 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/20170809/0bbc4bbf/attachment.html>