EEG: improving source localization

Discussions
1

Hi to all

I want to do source localization and after that I’d like to do a connectivity analysis with my EEG data. (Recorded with Geodesic 128 Electrodes)
The EEG is from an audio listening task and the data are already pre-processed. I used 109 Channels for the export. The yellow marked channels in the picture below where rejected.

1
1786×795 192 KB

Now I'd like to ask if my processing pipeline was correct.

  1. Loaded the ICBM 152 default anatomy

  2. Imported EGI 128 channel file for ICBM152
    2

  3. Checks MRI edit => Looked fine

    3
    3622×640 90.4 KB
    4
    4589×625 74.6 KB
    5
    5636×628 84.8 KB

  4. Imported 1 average file for every subject (auditory listening condition).

  5. Imported about 150 one second epochs resting state (EO) for the noise covariance matrix for each subject. (I used EO because during the listening task the subjects had their eyes opened as well and locked at a fixation cross).

  6. Calculated noise covariance matrix from EO

    6
    61075×157 5.33 KB

  7. Calculated head model OpenMEEG

  8. Computed source 2018 and used Loreta (Unconstrained, no covariance regularization, full results)

    7
    7657×633 42.2 KB

  9. Avarged all source estimations (arithmetic Average)

    8
    8556×825 48.1 KB

Now I'm struggling a bit with the source localization. Because the activation with the OpenMEEG looks a bit strange with basal activation and activations around the Cz electrode. But as the subjects simply listened to a piano tone, this should be localized in the auditory cortex.

Loretta with OpenMEEG (at 108ms)

1091112

MN with OpenMEEG(at 108ms)

2341

When I do the source localization with the 3-Shell-Sphere it locks quite better.
Loreta with 3-shell sphere (at 108ms)

1718419

MN with 3-shell sphere(at 108ms)

21222423

Do you have any idea what went wrong and how I could fix that and why there are this huge differences between the source models?
Or is it ok, when I'm going with the 3-Shell-Sphere? Because in your tutorial for the „Head modeling“ you recommend the OpenMEEG BEM model for EEG data.

I have also a question concerning the noise covariance matrix. Is it ok to use 100 sec of EO for the calculation or would you recommend something else, likeEC?

Do you have any hints for source localization with a geodesic high density system? Like removing electrodes or stuff like that?

My last question is about the connectivity. I thought about using the „Phase locking Value NxN“. But I’m not sure what kind of connectivity the coherence NxN and correlation NxN are. Is this a plain coherence and correlation matrix? Are there any papers or documentations for that? Or can you recommend another connectivity measurement?

Thanks
Domi

2

Hello,

1 Loaded the ICBM 152 default anatomy
2 Imported EGI 128 channel file for ICBM152
4 Imported 1 average file for every subject (auditory listening condition).

The proper way to import EEG positions for your recordings, in order to avoid possible issues due to varying channel orders across the various files, is to first import your recordings in your Brainstorm database and then use the menu "Add EEG positions", as illustrated in the EEG/Epilepsy tutorial:
http://neuroimage.usc.edu/brainstorm/Tutorials/Epilepsy#Prepare_the_channel_file

sLORETA vs. Minimum norm:

For a fair comparison, you should only be comparing between normalized minimum norm maps. https://neuroimage.usc.edu/brainstorm/Tutorials/SourceEstimation#Standardization_of_source_maps

OpenMEEG vs. spherical model

@Alexandre @John @Sylvain?

I have also a question concerning the noise covariance matrix. Is it ok to use 100 sec of EO for the calculation or would you recommend something else, likeEC?

It would probably be better to use these 100s of resting recordings than no noise covariance at all. If the task is done with eyes opened, then prefer eyes opened recordings to avoid issues with the eye movements you get with closed eyes. See our detailed recommendations:
http://neuroimage.usc.edu/brainstorm/Tutorials/NoiseCovariance#Variations_on_how_to_estimate_sample_noise_covariance

last question is about the connectivity. I thought about using the „Phase locking Value NxN“. But I’m not sure what kind of connectivity the coherence NxN and correlation NxN are. Is this a plain coherence and correlation matrix? Are there any papers or documentations for that? Or can you recommend another connectivity measurement?

@Sylvain @leahy @peterd @hossein27en @Guiomar?

3

Hallo

Thank you for the Help and the Links, I will go again over the Tutorials.

Regards
Dom