Normalization of source activity

Discussions
1

Dear all,

I’d like to examine whether specific motor sources, as defined by MNI coordinates, are active during a behavioural task. I am unsure how to set the thresholds for source activity since the maximum activity is very different for individual subjects. Does anybody know a way to normalize source activity? How can I perform a normalization in Brainstorm?

And another question: if I calculate an average of active sources for several subjects, how does Brainstorm cope with the problem?

Best regards,
Fabienne

2

Hi Fabienne,

Two options for normalizing the source maps: using the dSPM model (it is a normalized version of the wMNE model with respect with the noise level) or computing a Z-score of the source maps with respect with a baseline. Use the “dynamic Z-score” process for shared inverse kernels, and the “static” one for all the other files.

For averaging across subjects: if you have calculated the sources on individual anatomies, you need to project the sources on a template first.
http://neuroimage.usc.edu/brainstorm/Tutorials/CoregisterSubjects
Then you can average directly your normalized source maps across subjects.

Cheers,
Francois

3

Dear Francois,

thank you very much for your fast response! It helps a lot. I will try that.

Best,
Fabienne

4

Dear Francois,

when I use the dSPM model for computing the sources, I get a very large interhemispheric activation that I never observed with wMNE. Does this depend on the model used?

Best,
Fabienne

5

Yes, this is something commonly observed in EEG… This is EEG data, right?
Some people tend to prefer the sLORETA normalization in EEG, it shows smoother maps that are more representative of the reality of the source imaging spatial accuracy. But then you don’t get values that are comparable across subjects…
Alternatively: wMNE + Z-score, this is what we’ve been mainly using at the McGill MEG lab during the past years.

6

Hi Francois,

yes, I’m using EEG data. I’m going the try the z-score now… Just to clarify: by ‘shared inverse kernels’ you mean kernels for group data? For the single subject, I have to use the ‘static’ methods?

Best,
Fabienne

7

Just to clarify: by 'shared inverse kernels' you mean kernels for group data?

Not really, you should never compute one inverse kernel for multiple subjects.
http://neuroimage.usc.edu/brainstorm/Tutorials/CreateProtocol#Create_your_first_protocol
http://neuroimage.usc.edu/brainstorm/Tutorials/TutNoiseCov#Discussion

I was talking about the case where you compute only once the inverse for multiple trials or conditions and it get applied to all of them, showing all those "source links".
http://neuroimage.usc.edu/brainstorm/Tutorials/TutSourceEstimation#Computing_sources_for_multiple_data_files

For shared kernels, the "dynamic" option creates another of those shared source files, instead of one full source matrix per trial or condition. The difference in terms of computation time and storage can be very important.

For the single subject, I have to use the 'static' methods?

The "dynamic" option computes the Z-score on the fly when you open the standardized file, while in the "static" version the standardized values are saved directly in the files. It's just a different way of saving the same information.
For single files, it doesn't make much difference. You could always use the "dynamic" option, to make it simpler.

This distinction is going to disappear soon, and the best option will be selected automatically by the program

8

Ahhh, I was confused. Thank you Francois! I do understand now… :slight_smile:

All the best,
Fabienne

9

Dear Francois,

I have a problem with reducing the vertices for one subject. With 15000 vertices, I’m always getting a memory error. For only one subject, I get the following error message:

** Error: Line 484: Index exceeds matrix dimensions.
**
** Call stack:
** >tess_downsize.m at 484
** >bst_call.m at 28
** >tree_callbacks.m>@(h,ev)bst_call(@tess_downsize,filenameFull,[],[]) at 978

Do you know what is wrong here?

Best,
Fabienne

10

Hi Fabienne,

This is not a memory error, this is an indexing error in one of the structures available in the file (.Reg.Sphere, used for the subject co-registration).
I can’t say if there was an initial problem in the FreeSurfer files, or if something got corrupted in the surface file once in the Brainstorm database.

If you repeat all the operations from the beginning:

  • Right-click on the subject > Import anatomy folder > Selection of the FreeSurfer segmentation folder
  • Right-click on the cortex > Less vertices
    => Do you consistently get the same error?
    => If so, could you zip the FreeSurfer segmentation folder and send it to me? (upload it somewhere and send me the link in a separate email)
    => Please tell me also the numbers of vertices you are selecting at those two steps.

Cheers,
Francois