[Eeglablist] Guidance for exploratory study

[Tarik S Bel-Bahar]

Hello Rich, some notes below, best wishes. As a newbie, you should
feel free to play around, can't break anything within matlab/eeglab,
just restart. In general, however, data collection is best with dense
research grade systems, at least 30 minutes of experimental recording
per subject, and with data collectors who've received/done at least a
few hundred hours of collection/experiments training, including eeg
analysis experience.







*****************BEGIN
Yes, great, you can pull out metrics related to the "continuous EEG
features" rather than the event-related features. For example, overall
spectral power at a particular set of channels can be pulled out from
both continuous and "event-related" data. Note however that eeglab is
more geared towards.

Google Makoto's eeglab pipeline recomendations if you haven't.

If you plan to use eeglab's single-subject or STUDY functions, you
will likely need a single .set eeglab file for Each Subject X Each
Session X Each condition. Thus if 10 people each have 3 conditions,
then 30 files will be the final result. Once you're ready go to the
STUDY eeglab tutorial online to familiarize yourself with next steps.
Checking out the online documentation for the eeglab structure and
study structure will be useful too if you haven't had a chance to go
through them. Working with the tutorial eeglab data to start is also a
good idea for newbies, and examining eeglab and matlab while working
with the different files and steps.

If you get your eeg metrics by single-subject and single-condition,
you can grab the output data from the eeglab functions you run and
organize them into a matlab structure. However, STUDY in eeglab does
this for you, and creates new structures with metrics that you can
access with less work.

Regarding artifact rejection, when you mark a period to be removed,
after it is removed....then eeglab does not use the time around that
period when it tries to do epoching. It looks only for unbroken
periods. Thus eeglab does not make epochs from continuous data that
has "breaks or boundaries" in it, becuause that data is not
unbroken/continuous.

In your case, you might be able to achieve a good ICA with some clean
files (don't remove blinks or EMG, just big noise as per the online
eeglab tutorial). Then apply the ICA solution to your original
continuous data, and remove the blink/eye ICs. This would eliminate
most "time-dropping" because of eye-blinks.

If you use an automated method, and you're a newbie, you should
definitely visually compare the results of different cleaning
strategies, and show examples to some experts who can help you make
sure you're in the right direction.

More notes below...


****consider some/all of the following steps
Import each session as one .set file. Treat each file individually.

Place unique events (or get the specific times) of the start and end
of your 30 second periods.

Detect/remove bad channels, reref

Do visual rejection if you wish. Because of your long but few trials
per person, you will have a problem if your data is dirty.

**you may want to do visual rejection at this step. Remember that if a
"period of time" you remove in the continous data is necessary for an
"epoch or trial", then that epoch/block/trial will not be generating
by eeglab's epoching function, because that period is no longer
continuous (as a chunk has been removed).

**for your purposes, for cleaning, you should try clean_rawdata, which
automagically (ASR) removes a lot of dirt from the signal and makes
the data clean enough that there are very few "bad" periods to remove.
See also note at the top about removing artifactual ICs from your
continuous data.

Then Cut into just the 30-second blocks (one .set file per condition,
with each .set file having as many trials as the condition)

Then drop/remove any of the 30 second blocks that have a lot of noise
(as per eeglab artifact detection tutorial).

Then do ICA for further cleaning (note that you can't benefit from ICA
much because of your low channel count, and because of the relatively
short time period of your recording sessions. As mentioned you can
also apply ICA and drop ICs within your original data. However, this
kind of cleaning that results is technically the same thing that
happens with ASR in clean_rawdata (see/try also PREP, which uses ASR I
believe).



****for your purposes you might want to google/try some of the range
of approaches from BCI and consumer EEG research. They often have to
make do with data that is not necessarily ideal length, quality, and
density. See also the openBCI tools, and BCILAB if you've not had a
chance.


*****************END










On Wed, Sep 21, 2016 at 3:15 PM, Ingram, Richard E - ingramre
<ingramre at jmu.edu> wrote:
> Greetings EEGLAB community,
>
> I'm an EEGLAB newbie.  I've done the tutorial and reviewed hundreds of
> Eeglablist entries.  What a wonderfully supportive community.  I wonder if
> you might have a few tips on how best to approach analyzing data from an
> exploratory study?
>
> I have 8 subjects, each of whom has 15 30-second time blocks of continuous
> EEG data (sampled at 128 Hz from 14 channels) reflecting differences in task
> difficulty (3 levels) and task condition (3 conditions).  I wrote software
> that handles the stimuli display and enters markers (generated by
> mouseclick) into the datastream .  For this first look, I'm more interested
> in characterizing the continuous data than event-locked data (although I
> would like to come back to event-locked data later).
>
> How should I organized the data files for import?  Right now I have one data
> file for each subject.  Should I divvy into 120 separate files (8 subjects x
> 15 data blocks) with each block uniquely named and import them separately?
>
> As to artifact cleaning, two questions - 1) If I reject artifacts by eye,
> wouldn't that lead to data blocks of different lengths and so complicate
> analysis?  2) Or should I use one of the automated methods available via
> EEGLAB extensions (eg, clean_rawdata, PREPPipeline, ADJUST, AAR) and if so,
> which do you recommend?
>
> I know these questions may be simplistic but I do appreciate any and all
> guidance you may provide.
>
> Rich
>
>
>
>
> _______________________________________________
> 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