[Eeglablist] Pipeline questions
alexandre.lehmann at gmail.com
Tue Nov 17 09:03:44 PST 2015
Is anyone aware of a toolbox / software to compute phase lag index and
related measures ? I haven't seen that in EEGLAB ?
Thank you for this insightful discussion, is the paper you mention ?
Numan, T., Stam, C. J., Slooter, A. J. C., & van Dellen, E. (2015). Being
Conscious of Methodological Pitfalls in Functional Brain Network Analysis:
*Anesthesiology*, *123*(2), 484‑485.
On Thu, Oct 22, 2015 at 6:50 PM, Tarik S Bel-Bahar <tarikbelbahar at gmail.com>
> Hello Amanda, a few notes below, hope they are useful ! Cheers!
> For beta band coherence and the length of epochs, you may want to review
> the methods from several recent articles you can find on Google Scholar in
> major journals. I think a rule of thumb is you would want at least 3 or
> more cycles of beta within your time period/epoch/segment. half a second
> seems to . There may be some assumptions about stationarity that need to be
> considered here too. See a recent review from this year of methods and
> pitfalls in connectivity from Stamm and colleagues.
> Overall, the settings on artifact detection need to be played with, and
> you need to double check that the settings are working as expected for you.
> Try exploring other eeglab-related tools for more "automatic" or different
> methods for cleaning up your data, of which there are more and more of in
> the field. You can also check the settings in published reports, but I
> don't think many articles include specifics at that level of detail,
> unfortunately. you might especially like ASR and PREP tools from Kothe and
> colleagues for noisy channel detection.See also SCADS and fieldtrip based
> cleaning techniques.
> Average reference on 64 channels is ok. you may want to remove channels on
> the face, neck, from the average reference. Some groups average reference
> after ICA. See past eeglablist discussions regarding not leaving bad,
> weird, or uninformative channels in for re-referencing.
> Your're fine if you don't include non-EEG channels for ICA, it will pick
> up eye movement and muscle artifacts anyway
> Try SASICA and other IC rejection toolboxes too, competitive runoffs
> between these toolboxes have not yet been done. there are ICs that are not
> pure artifact, and not pure brain dynamics that should likely not be
> I don't think you need to run ICA again, unless you're getting much better
> results from re-running ICA, which should not be the case. Read Onton &
> Makeig chapter in Luck Handbook of ERP components. See also ICA video
> tutorials at EEGLAB summer school online.
> I'd recommend you stay with the ~75% data you have, and not re-apply it to
> the raw data. You're actually not dropping a lot, so it seems like you have
> clean data. I would double-check that you are cleaning well-enough. If you
> IC decompositions have several known ICs, and few or none that are
> dominated by single-trial activity, then you've likely cleaned enough.
> It won't hurt to apply the ICs to the raw data and have a look at things
> that way too. this would allow for near-continuous analysis of the ICs at
> least over the whole session, although your periods of interest are likely
> only at specific times or trials.
> I think there are two camps (at least) in EEG-ICA land. One camp rejects
> just eye-artifacts and perhaps muscle artifacts. The other camp removes
> everything except the really cognitive-brain ICs.
> From the perspective of EEGLAB, look at it this way. You've decomposed the
> data into ICs, which should reflect discrete brain activity with distinct
> spatial-electrode maps.
> Other camps choose to recombine their ICs after a little or a lot of
> cleaning, and they analyze this reconstituted and very clean (perhaps too
> clean sometimes) data.
> Other camps argue that decomposition techniques are the proper way to
> analyze EEG data, and the correct data to analyze is the ICs over time.
> From this view, these are the "real" components, the ICs.
> In other words, why not take your good ICs and do beta-band coherence
> between the ICs themselves.
> On Thu, Oct 22, 2015 at 9:58 AM, Armand Hoxha - Volunteer <
> AHoxha at kesslerfoundation.org> wrote:
>> Dear EEGLAB community,
>> I am currently working on a dataset from which I need to get Beta band
>> coherence (motor cortex to muscle coupling). My processing pipeline for the
>> dataset so far has been:
>> 1- Import data
>> 2- Add channels, Optimize Center, Set channel types
>> 3- Cleanline (default settings, to EEG and EMG)
>> 4- FIR 1-250Hz (pop_eegfiltnew, to EEG and EMG)
>> 5- Segment data in epochs (0.5 seconds of epochs, for coherence
>> purposes I noticed in most studies the general number of windows is usually
>> above 170, thus to achieve the same statistical relevance I needed to have
>> windows of 0.5s rather than 1second. I am a little confused about which
>> window length should be preferred for beta band coherence)
>> 6- Remove noisy channel (if more than 10% of data is “bad”, then I
>> remove channel)
>> 6a)pop_eegthresh on EEG channels, with limits of -50 to 50 uV
>> 6b)pop_jointprob on EEG channels, single-channel std:6, All-channel std:2
>> if a channel is generally responsible for about 10% of rejected epochs, I
>> reject channel
>> 7- Interpolate rejected channels, pop_interp ;Spherical method
>> 8- Average reference EEG channels(for ICA and coherence
>> calculations, our actual reference is between FCZ and CZ)
>> 9- Run ICA on EEG channels only ( I have read that EOG and ECG
>> channels can be included, but they are bipolar readings with a different
>> reference from EEG, should I include them regardless?
>> 10- Use MARA toolbox to get a general sense of how I should feel about
>> some IC’s , also because I prefer the MARA spectrum plot over the default
>> 11- Reject IC components related to EMG artifacts (facial movements,
>> neck, jaw, eyes), from 64 components I reject ~20
>> 12- Reject epochs based on the Components (jointprob 6,2)
>> 13- Run ICA again, reject EMG components again
>> 14- Most of the time by this point I have retained ~75-85% of dataset,
>> and I use this dataset to do coherence analysis. I have read that some have
>> suggested instead to export the ICA matrix of the processed data, and apply
>> it to the raw data?
>> Am I rejecting too many IC’s from the data, or would this data produce
>> reliable results?
>> Thanks in advance,
>> Armand Hoxha
>> Biomedical Engineer, HPEL
>> Kessler Foundation
>> 1199 Pleasant Valley Way, West Orange, NJ 07052
>> [image: Kessler Foundation Signature]
>> [image: Kessler Foundation Facebook Page]
>> <http://www.facebook.com/?ref=home#!/pages/Kessler-Foundation/123330414345741?sk=info> [image:
>> Kessler Foundation You Tube]
>> <http://www.youtube.com/user/KesslerFoundation> [image: Kessler
>> Foundation Research Twitter Page] <http://www.twitter.com/KesslerFound>
>> The information in this transmission is intended for official use of the
>> Kessler Foundation. It is intended for the exclusive use of the persons or
>> entities to which it is addressed. If you are not an intended recipient or
>> the employee or agent responsible for delivering this transmission to an
>> intended recipient, be aware that any disclosure, dissemination,
>> distribution or copying of this communication, or the use of its contents,
>> is strictly prohibited. If you received this transmission in error, please
>> notify the sender by return e-mail and delete the material from any
>> 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
> 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...
More information about the eeglablist