[Eeglablist] Transferring ICA weights from pruned data to original (moderately cleaned) dataset results in non-smooth activity power spectrum

[Arnaud Delorme]

Dear Amelie,

Without plotting the erpimage with the same scale, it is hard to have a clear idea of the amount of artifacts in the original dataset vs the aggressively pruned dataset. It is true that you original data seems to have strong artifactual transient spikes of activity, which drive the scale into an extreme range (more than 600 microvolt). Maybe these artifacts data should be removed.

And yes, if you run ICA on the pruned data, and apply the decomposition to the original data, you can expect some strong artifact in the original data to remain.
It is difficult to achieve a perfect balance. I personally do not do it this way: I always process the pruned data. However, sometimes I filter the data above 1 or 2 Hz, apply ICA, and then use the ICA components on a less aggressively filtered dataset (at 0.01 Hz for example or no filtering). This is because ICA quality often dramatically increases when you high pass filtering your data. ICA components may be viewed as spacial filter so it is perfectly valid to so - you simply apply some spatial filters on the unfiltered data.

Best wishes,

Arno

> On Oct 16, 2018, at 8:42 AM, Roi, la, A. <a.la.roi at rug.nl> wrote:
> 
> Dear Arno,
> 
> Thank you very much for taking the time to answer my question. I understand that the visual difference in the plot is caused by the difference in the plots' scale. However, my actual question is whether the visual difference in the plots also signals a failed transfer of ICA weights from a heavily cleaned dataset to a dataset with more noise. Or is it that the plots of the non-cleaned dataset are just different, because this dataset contains more noise which changed the power spectrum of the ICA components? So in short, should I be worried based on these plots or not?
> 
> Kind regards,
> Amélie la Roi
> 
> On Tue, Oct 16, 2018 at 1:06 AM, Arnaud Delorme <arno at ucsd.edu> wrote:
> Dear Amélie,
> 
> Look at your plots and you will see that the scale is very difference which should explain the difference you observe.
> Best wishes,
> 
> Arno
> 
> > On Oct 2, 2018, at 1:27 AM, Roi, la, A. <a.la.roi at rug.nl> wrote:
> > 
> > Dear list,
> > 
> > When preprocessing my EEG data, I run ICA on a heavily cleaned dataset. Subsequently, I transfer the ICA weights to the original dataset, which is only moderately cleaned. The number of channels in the two datasets are the same and both are re-referenced to average reference (therefore, I use the 'pca', [number of channels in dataset - 1] settings when running ica). 
> > 
> > Component 1 and 6 probably reflects blinks and horizontal eye movements, respectively. When I plot the components in the pruned dataset, the activity power spectrum looks fine (see attachment: ptc5_component1_pruneddata.png and ptc5_component6_pruneddata.png). However, when I plot the components in the original dataset, the activity power spectrum of both components contains bumps, while the scalp plot remain the same (see attachment: ptc5_component1_originaldata.png and ptc5_component6_originaldata.png). Does anyone know what's going wrong here? The component activations when plotted in the original dataset still indicate that component 1 and 6 reflect blinks and horizontal eye movements, respectively, although the component activation plot does contain a bit more noise than the component activation plot of the pruned data.
> > 
> > Any comments are welcome!
> > 
> > Kind regards,
> > Amélie la Roi
> > 
> > -- 
> > Amélie la Roi, MA
> > PhD student CLCG, Semantics and Cognition
> > University of Groningen, the Netherlands
> > Harmony building, room 1311.0412 | Phone: +31 50 363 6683
> > 
> > <ptc5_component1_originaldata.png><ptc5_component1_pruneddata.png><ptc5_component6_originaldata.png><ptc5_component6_pruneddata.png>_______________________________________________
> > 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
> 
> 
> 
> -- 
> Amélie la Roi, MA
> PhD student CLCG, Semantics and Cognition
> University of Groningen, the Netherlands
> Harmony building, room 1311.0412 | Phone: +31 50 363 6683
>