[Mne_analysis] ICA code clean up

Tue Oct 13 07:13:01 EDT 2015

Hi Virginie,

glad you are finally using our ICA ;)
Let me reply inline:

On Tue, Oct 13, 2015 at 11:09 AM, Virginie van Wassenhove <
virginie.van.wassenhove at gmail.com> wrote:

> Hi
>
> some comments on inconsistencies found on
> http://martinos.org/mne/dev/auto_tutorials/plot_ica_from_raw.html also
> observed when actually performing ICA on data.
>
> In your tutorial, you use:  n_max_ecg, n_max_eog = 3,1
>
> Then you effectively find 3 components for ECG (in red in the graphs: IC
> 59, 33 and 32) but  2 components for EOG (IC 39 and 59)in section 2.
> 1. If we understand properly, IC 59 captures both ECG and EOG arfifacts,
> is this correct?
> 2. In section 3, there are four components showing up in read on the top
> left graph and these are IC 33, 34, 40, and 60. There's a problem of
> numbering here? Shift of 1???
>
>
You pick n_max_XXX components effectively, but the algorithm may propose
more above threshold components. You pick the k first biggest, sorted by
score.
Then the index shift is a peculiarity, as long as you are looking at
component plots you see indices.
Once you look at ICA components in a Raw or Epochs object (ica.get_sources
or certain plot methods) the numbers refer to channel names and hence
regular numbers. I know it can be confusing, we will eventually find a
solution.

> In our data:
> we applied ICA to raw data :
> - we used the same codes specifying n_max_ecg, n_max_eog = 3,3
> - 2 components were found for ECG (IC 0 and IC 38 in red with the indexing
> starting at 0)
>
> [image: Inline image 2]
>
>
The SNR of the detection metric looks very good.

> - 2 components are found for EOG (4 in red; 2 for each EOG channels
> (vertical and horizontal, I am assuming the scores below are or each EOG
> channel)
>
>
yes, two channels, you also see that the one component does not score high
on the other channel.
we could improve the labeling here.

> [image: Inline image 3]
>
>
> When using the ica.plot_sources for ECG only, 3 components (ICA 001; 007;
> 039) show up in red even though the ecg_inds only contains 2 components
> (indices 0 and 38).  Thus, 2 problems here: you need fix the indices issues
> (??) and why 3 components and not 2?
>
>
there is no indexing issue. I fear while trying to make this consistent we
stopped somewhere in the middle.

the overlay plots are not selective, they just display all components that
are marked bad.
Currently we do not have an API or data structure that gives you post hoc
reasons for which a component was selected. Could be improved if enough
people say this is important to them.
Apparently the late ECG wave looks like an EOG for this ICA solution, so
you have 3 components, ok?

> [image: Inline image 4]
>
> With this, I am unsure of which ICA are actually being applied in the next
> steps.
>
> For the EOG, the exact same 3 component names are indicated on the figure
> and eog_inds = [6, 38]
>
>
see above

> [image: Inline image 5]
>
>
> Can you please explain what is correct and what is not above? and see in
> the codes where indices get lost etc...
>
>
Let me know if it is clearer now.

Best,
Denis

> Thanks, best wishes,
> Virginie
>
>
>
>
>
>
> --
> Virginie van Wassenhove
> Virginie.van.Wassenhove at gmail.com
> work:    +33(0)1.69.08.1667
> skype:  virginie_vw
> https://sites.google.com/site/virginievanwassenhove/
>
> _______________________________________________
> Mne_analysis mailing list
> Mne_analysis at nmr.mgh.harvard.edu
> https://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis
>
>
> The information in this e-mail is intended only for the person to whom it
> is
> addressed. If you believe this e-mail was sent to you in error and the
> e-mail
> contains patient information, please contact the Partners Compliance
> HelpLine at
> http://www.partners.org/complianceline . If the e-mail was sent to you in
> error
> but does not contain patient information, please contact the sender and
> properly
> dispose of the e-mail.
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20151013/6ea42b2c/attachment-0001.html 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: reconstructed_ECGsources.png
Type: image/png
Size: 66108 bytes
Desc: not available
Url : http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20151013/6ea42b2c/attachment-0004.png 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: reconstructed_E0Gsources.png
Type: image/png
Size: 77794 bytes
Desc: not available
Url : http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20151013/6ea42b2c/attachment-0005.png 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: plot_scores_ECG.png
Type: image/png
Size: 24348 bytes
Desc: not available
Url : http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20151013/6ea42b2c/attachment-0006.png 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: plot_scores_EOG.png
Type: image/png
Size: 33156 bytes
Desc: not available
Url : http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20151013/6ea42b2c/attachment-0007.png 