ICA Analysis

✉️ Mailing List Archive (read-only)
1

External Email - Use Caution

MNE Team,

First and foremost we?d like to thank you for the responses and suggestions to our last email. As we get closer to finalizing our Py script, you have all certainly given us some food for thought.

We now have a new concern relating to the ICA analysis. We have been consulting the documentation and have followed the tutorial for ICA analysis:
https://mne.tools/stable/auto_tutorials/preprocessing/plot_artifacts_correction_ica.html

However, our component scores seem to be all over the place. It?s also saying that there are three EOG channels in the latent sources plot, when there should only be HEO and VEO. Historically, we have used the Semlitsch algorithm [Semlitsch HV, Anderer P, Schuster P, Presslich O., (1986) A solution for reliable and valid reduction of ocular artifacts, applied to the P300 ERP, Psychophysiology,23(6):695-703] with no issues. Is there a way to use this method in MNE-Python instead? We realize that the Semlitsch algorithm may be outdated, is there a reason/reference ICA may stand apart from this algorithm?

As always thank you for your assistance,

UNLV PEPLab
Bianca Islas
Research Assistant

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20190813/ee1cb7b0/attachment-0001.html
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Comp_scores.PNG
Type: image/png
Size: 22616 bytes
Desc: not available
Url : http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20190813/ee1cb7b0/attachment-0002.png
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Lat_Sources.PNG
Type: image/png
Size: 64498 bytes
Desc: not available
Url : http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20190813/ee1cb7b0/attachment-0003.png

2

External Email - Use Caution

Hi,

it does not look so bad to me.
I would not call that ?all over the place?.
Perhaps there is a bit of noise but you seem to capture the main components. How many components it takes to describe your artefact not only depends on the number of channels but also on how long the recording is, if there were head movements and on your filter settings. You can for example lowpass filter your data at 2hz at fit-time or fit ICA run by run
to mitigate issues with non-stationarity that may drive up your component numbers.
Anyways, your plot does look rather encouraging to me, having 2-3 more components than expected is not a catastrophe if the results are ok.
Did you look at the overlay plots to see how the average EOG looks after rejection?

Hope that helps,
Denis

3

External Email - Use Caution

Hi!

I haven?t looked at your results in detail, but maybe my blog posts about removing ocular artifacts via ICA (https://cbrnr.github.io/2018/01/29/removing-eog-ica/) or via regression (https://cbrnr.github.io/2017/10/20/removing-eog-regression/) might be helpful.

Clemens

4

External Email - Use Caution

Once again, thank you MNE team!

The suggestions you gave us, helped out immensely. Mr. Clemens Brunner, your blog on ICA and regression analysis were real eye-openers!
Following your lead on creating a copy, and making some tweaks here and there to fit our needs, it would appear that ICA000, as we were hoping/expecting is the component we will be continually removing. This certainly streamlines the process of running our script on all of our CNT files.

Once again thanks your for your time, suggestions, and resources!

Best,
Bianca Islas
UNLV PEPLab Research Assistant