[Eeglablist] What is the best approach to reject bad components when using a 256-channel (EGI) equipment

[Tarik S Bel-Bahar]

Hello Anoop, some notes for you right below, best wishes.

*************** *Overall, one can 1) just remove eye and muscle ICs, and
keep the rest of the ICs to rebuild the EEG data. This is what researchers
do when they are conservative and are only interested in ICA for light
cleaning.
however, researchers that focus on ICA, and that trust ICA more, tend to
remove all but the neural ICs from their data, thus in a sense, they are
more strict and focus on ICs that are really neural.
There is no agreement at this point as to whether removing more ICs removes
useful neural data, it depends on the biases of the researchers.
Besides SASICA, Adjust, and MARA plugins, you may also want to try Luca's
recent IC-classification plugin. Finding the classic artifact ICs is easy
as they are quite stereotypical, spatially speaking..

****************PCA is not recommended before ICA by experts on the
topic: See the below recent article from Artoni et al that suggests that
PCA is not good before ICA.
Artoni, F., Delorme, A., & Makeig, S. (2018). Applying dimension reduction
to EEG data by Principal Component Analysis reduces the quality of its
subsequent Independent Component decomposition. *NeuroImage*, *175*,
176-187. The online suggestions for beginners that you are looking at are
useful, but it's important to verify/explore what steps to take for
yourself, based on published literature.

*************** Usually there are only about ~5 or 15 neural ICs, even with
high-density EEG, so one can just focus on those, and remove the rest of
the ICs. This is also based on what researchers usually do and find in many
past research publications. See at end of email for list of citations about
this topic

*********** If you haven't had a chance to yet, you can learn more about
these topics by reading the sasica artcle, googling eeglab list for your
topic (rejecting ICs), review eeglab tutorials on using ICA and pruning IC
results.

****** In eeglab one can automatically reject outside the head dipoles.
However, one needs to make sure that electrode positions are correct, and
that the dipfit head-to-electrodes setup is correct, as these can impact
the IC dipole results. Further, the residual variance cutoff is also very
useful in quickly pruning ICs (and also available in the eeglab STUDY setup
gui).

************ If you think you have sleep related spindles, There are
several sleep EEG toolboxes that can help you define and confirm sleep
related spindles. I would recommend seeing how the analysis changes with
and without the ICs dominated by these spindles.


**********************Examples of citations about low number of neural ICs
usually found:

1)    a small universe of possible neural/cognitive ICs in EEG datasets:
Makeig S, Onton J. ERP features and EEG dynamics: An ICA perspective. In:
Luck S, Kappenman E, editors. Oxford Handbook of Event-Related Potential
Components. Oxford University Press; 2009.

2)      a common small set of ICs that regularly show up with ICA
decompositions of EEG data: Delorme, A., Palmer, J., Onton, J., Oostenveld,
R., & Makeig, S. (2012). Independent EEG sources are dipolar. PloS
one, 7(2), e30135.

3)     a small number of ICs are kept for analyses: Steele, V. R.,
Anderson, N. E., Claus, E. D., Bernat, E. M., Rao, V., Assaf, M., ... &
Kiehl, K. A. (2016). Neuroimaging measures of error-processing: Extracting
reliable signals from event-related potentials and functional magnetic
resonance imaging. Neuroimage, 132, 247-260.

4)      a small number of ICs are kept for analyses, in this case using
MEG: Urbain, C. M., Pang, E. W., & Taylor, M. J. (2015). Atypical
spatiotemporal signatures of working memory brain processes in
autism. Translational psychiatry, 5(8), e617.

5)     a small number of ICs are kept for analyses of mediofrontal
activity: “The mean number of resulting maximally independent and
localizable EEG components used in subsequent analysis was 15 per subject
(range: 7 to 26)”. Onton, J., Delorme, A., & Makeig, S. (2005). Frontal
midline EEG dynamics during working memory. Neuroimage, 27(2), 341-356.

6)   a small number of ICs are kept for analyses: “6.8±5.5 of the top 30
components were removed from each EEG recording”. Wu, J., Srinivasan, R.,
Kaur, A., & Cramer, S. C. (2014). Resting-state cortical connectivity
predicts motor skill acquisition. NeuroImage, 91, 84-90.

7)     a large number of datasets and examining the reliability of ICA
decompositions, the following paper found only about 15 reliable clusters
of ICs across participants (including about 10 neural IC clusters and about
5 artifactual IC clusters), and a median of 18 to 20 ICs per dataset that
had <5% dipole-fitting residual variance: Artoni, F., Menicucci, D.,
Delorme, A., Makeig, S., & Micera, S. (2014). RELICA: a method for
estimating the reliability of independent components. NeuroImage, 103,
391-400.

8)     a broad range of blind-source separation ICA algorithms, ~5 to 15
reliable ICs were found by each algorithm. Bridwell, D. A., Rachakonda, S.,
Silva, R. F., Pearlson, G. D., & Calhoun, V. D. (2016). Spatiospectral
Decomposition of Multi-subject EEG: Evaluating Blind Source Separation
Algorithms on Real and Realistic Simulated Data. Brain topography, 1-15.







On Wed, Aug 29, 2018 at 11:04 AM anoop jagadeesh <anoop2187 at gmail.com>
wrote:

> Dear all,
>
> We use a 256-channel EGI equipment to measure auditory evoked potentials.
> There are a few questions I have and I hope you can answer them for me
>
> 1. Should I run the ICA for the full 256 channels or using a PCA (as
> suggested in Makoto's preprocessing pipeline)? Till now I have done it
> using the PCA approach. But it would be helpful to know of any other
> approaches to handle the high-density recordings.
>
> 2. Even after running the ICA (RUNICA) with a PCA value of 64, there are
> many components which are very difficult to explain. Normally, we would
> remove only the components which are"outside the head". However, there are
> many components which are clearly 'bad' inside the head. Especially, there
> are components within the 64 which look like a localised dot. Do we remove
> these components? Please note that I have seen instances where there are 10
> (out of 64) or even more such dot components
>
> 3. Another important issue I have encountered is there are many cases
> where 'spindles' (very likely related to sleep) occur. These spindles are
> often seen in the top 5 most robust components. Do we keep them or remove
> them?
>
> Simply put, how strict do we need to be while rejecting the bad
> components. Inputs regarding this are highly appreciated
>
> Regards
>
> --
> Anoop B J
> Junior Research Fellow
> Dept of Audiology,
> All India Institute of Speech and Hearing, Mysuru, India
> _______________________________________________
> 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/20180829/d8c39c7c/attachment.html>