[Eeglablist] Ghost ICs on EEG data

[Scott Makeig]

Maitane -

See the paper linked below for why performing PCA to reduce the data
dimension to less than its inherent rank is not advised before ICA
decomposition.

Artoni F, Delorme A, S Makeig. Applying dimension reduction to EEG data by
principal component analysis reduces the quality of its subsequent
independent component decomposition.
<https://sccn.ucsd.edu/~scott/pdf/Artoni_PCAICA18.pdf> *NeuroImage*
175:176–187,
2018. https://urldefense.com/v3/__https://doi.org/10.1016/j.neuroimage.2018.03.016__;!!Mih3wA!RMjgZqLUIWol0JEI_rw1FzRIfSQth2BVphKLh2QIp6UOBIzjd6U_oXksaRAEvTb_eJqL_Q$ 
<https://urldefense.com/v3/__https://www.sciencedirect.com/science/article/pii/S1053811918302143__;!!Mih3wA!RMjgZqLUIWol0JEI_rw1FzRIfSQth2BVphKLh2QIp6UOBIzjd6U_oXksaRAEvTYr_pQITA$ >

The k=30 timepoints/ICA_weight suggestion is not meant to be a fixedrule -
definitely first try applying ICA decomposition (after careful cleaning of
'non-stereotyped' noisy data periods) to your whole data. Its length
(k=15.5) may prove to give quite meaningful source information.

If not, then consider removing a spatially dispersed subset of electrodes
before decomposition OR (more complicated and sophisticated), you might try
performing repeated decompositions removing different spatially dispersed
subsets of electrodes, then cluster and merge components in common from the
distributions, akin to the approach used in the paper below (using time
point  rather than channel bootstrapping)...

*Artoni, F., Menicucci, D., Delorme, A., S Makeig, Micera, S. RELICA: a
method for estimating the reliability of independent components.
<https://sccn.ucsd.edu/~scott/pdf/Artoni_RELICA_NeuroImage14.pdf>NeuroImage,
103:391-400 (2014)*

Scott Makeig

On Fri, May 15, 2020 at 3:26 AM Maitane Barrenetxea Carrasco <
mbarrenetxea at mondragon.edu> wrote:

> Hi all,
>
> I am working on a dataset that has 128 EEG channels and 4 EOG channels (2
> vertical + 2 horizontal) plus a nose-tip reference. Hence, 133 channels in
> total.
> After ICA decomposition I have noticed that there are 2 ghost ICs in one
> the datasets (the 13 files that I have preprocessed before were apparently
> fine). These ghost ICs appear in first two positions of the ICs and exhibit
> inverted time-course and activation maps.
>
> This seems to be a problem of low-rank data before ICA. The thing is that
> in this case my data is full rank when entering ICA (128 channels and rank
> 128 as this subject doesn't have any interpolated channels). Additionally,
> I am only computing 92 components (I use PCA for dimensionality reduction)
> as the EEG data length doesn't allow for the calculation of at least 30
> data points per ICA weight if all 128 components are to be estimated (data
> length=254458, 30*92^2=253920).
>
> So, is this a matter of runica() instability or is there something wrong in
> my pipeline? Here is the code I use to preprocess the EEG data:
>
> %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
> %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
> %% STEP 2: Filtering  %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
> ...
> %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
> %% STEP 3: Remove initial and final segments of the data
> %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
>     ....
> %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
> %% STEP 4: Import channel info
> %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
>
> if  there are bad channels then
>         %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
>         %% STEP 5: Reject bad channels
>         %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
>
>         EEG = pop_select(EEG,'nochannel',toremove);
>
>         %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
>         %% STEP 6 : Interpolate
>         %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
>         EEG = pop_interp(EEG, originalEEG.chanlocs, 'spherical');
> end
>
>     %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
>     %% STEP 7: Average re-reference --> exclude EOG channels to avoid
> artifact propagation
>     %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
>     EEG = pop_reref( EEG, [],'exclude',[129:132] );
>
>     %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
>     %% STEP 8: ICA --> EOG and reference channel excluded
>     %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
>
>     EEG = pop_runica(EEG, 'pca', 92,
> 'extended',1,'interupt','on','chanind',[1:128]);
>
> %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
>
> Thank you very much in advance,
>
> Maitane
>
> --
> Maitane Barrenechea Carrasco (PhD)
> Biomedikoa - BIO
> Mondragon Unibertsitateko Goi Eskola Politeknikoa
> Loramendi, 4; 20500 Arrasate - Mondragón (Gipuzkoa)
> Tel. : +(34) 647504294 / +(34) 943794700 + Ext. 8162
>
> https://urldefense.com/v3/__http://www.mondragon.edu__;!!Mih3wA!ULzoRJq4QVX4tD2cn-imhLSZRrdTWUjwCJNE0QGEDDphbevgTlbKbGve0CJ3aGjeOsRUiQ$
> _______________________________________________
> 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



-- 
Scott Makeig, Research Scientist and Director, Swartz Center for
Computational Neuroscience, Institute for Neural Computation, University of
California San Diego, La Jolla CA 92093-0559, http://sccn.ucsd.edu/~scott