Hello
I’m wondering what the best method is for performing group analysis on frequency data at the source level from EEG recordings, similar to the process outlined here: http://neuroimage.usc.edu/brainstorm/Tutorials/RestingOmega#Power_maps.
If you are using the default anatomy, rather than individual anatomy, is it still wise to smooth the data, to allow for individual differences, before performing group analysis?
I saw on this post on the forum (Smoothing frequency data at the source level?) that smoothing is not possible with a mixed model, and that the unconstrained sources are already smooth enough.
If so, when using a mixed model, is it possible to smooth the frequency data for the constrained cortical sources alone e.g. by projecting the results onto the default anatomy surface, and then performing the smoothing? Or would the analysis need to be run with a new head model?
Many thanks
Luli
If you are using the default anatomy, rather than individual anatomy, is it still wise to smooth the data, to allow for individual differences, before performing group analysis?
If you are working with a MNI template for all your subjects, you already accept to give away some accuracy in the source modeling. Therefore, we recommend you use unconstrained source models.
As you said, with unconstrained sources, the results you obtain are smooth enough not to have to smooth again before the group analysis. These guidelines may help with some decisions: http://neuroimage.usc.edu/brainstorm/Tutorials/Workflows#Unconstrained_cortical_sources
when using a mixed model, is it possible to smooth the frequency data for the constrained cortical sources alone e.g. by projecting the results onto the default anatomy surface, and then performing the smoothing? Or would the analysis need to be run with a new head model?
There are currently no functions to convert source maps from "mixed head model" to a "surface head model". So I don't have any solution to offer other than computing a new head model.
Note that for EEG analysis using a MNI template, you can't expect a high spatial accuracy and it is probably not interesting to use these complicated "mixed head models". I'd recommend using only the cortex surface (surface head model, unconstrained sources), or volume head models (http://neuroimage.usc.edu/brainstorm/Tutorials/TutVolSource#Group_analysis).
Francois
Hi Francois
Thanks for your response.
I saw in one tutorial that it was recommended that unconstrained sources should be used when using the default anatomy, and yet in the deep atlas tutorial, it creates constrained cortical sources, so your answer clears that up for me.
With the default MNI anatomy for EEG recordings, what kind of spatial accuracy can I hope for e.g. using a 15000 vertices cortical surfaces model vs. with a 15000 vertices volume head model? Does this accuracy decrease for subcortical structures when using the volume head model?
Thanks very much
Luli
The spacial accuracy is low because EEG source modeling is not an accurate imaging technique.
You would get similar results whether you use a surface or volume head model, but visualized in different ways. Mixed head models would also be very similar, but much more complicated to handle.
If you are interested by the effect of the source model on your results, the best solution is run your analysis in various ways. In the end, ideally your observations should converge.