Creating a group average mask in source space

I have 64 channel eeg data for subjects with individual MRIs. I want to find the sources of the erp data on certain trials but some subjects have few trials - only 9 or 10. So I thought I could restrict the search for individual subject's sources by creating a mask in source space from the group average and then identify individual's sources within this mask. Can anyone advise how I might do this given the headmodels etc are based on individual's MRIs?

Restricting the source reconstruction is not advised, especially if you data is noisy (few trials in your averages). If you limit your source space to your regions of interest, the minimum norm model will try to explain all the signal components in the EEG recordings with only your ROIs, not only the signals actually from your ROI...

I recommend you try the following:

  • use the same forward and inverse models for all the subjects (this way you also avoid introducing additional bias between subjects),
  • project all the sources to the ICBM152 default anatomy,
  • compute the grand averages in source space
  • design scouts based on the grand averages source maps
  • project these scouts back to the individual brains
  • use these scouts for analyzing the data of each individual subjects

Hi Francois,
Thanks so much for the fast reply!
Is there a tutorial on how to do this?
I found this one which looks relevant
Thanks again!

There is no specific tutorial for this particular sequence of operations, but it should not be too complicated.

  1. For producing group averages in source space, you can find some inspiration in the group analysis tutorial:

  2. Design your ROIs on the group average of the source maps:

  3. Project the group scouts on the individual brains:
    in the Scout tab, menu Scout > Project to... > Target subject > Target cortex surface
    (if you need to script this for tens or hundreds of subjects, you can use the function bst_project_scouts.m directly, as called by the menu "Project to":

  4. Compute the ROI-based measures you need at the individual level. And probably run some non-parametric statistic on them?