Hi François,
I’d like to do export cortical surface maps (projected to std space) for a surface-based group analysis (in SPM).
Looking at the tutorial example maps, I am wondering if sLORETA might be a better solution than MNE for this since it smooths over gyri and sulci (and therefore people’s active areas are more likely to align to some degree over highly folded areas). Is this a reasonable thing to do?
Emily
Ref:
http://neuroimage.usc.edu/brainstorm/Tutorials/TutSourceEstimation#Minimum_norm_options
Hi Emily,
I’m still not very clear on this topic. I’m still waiting for instructions from the project’s PIs on how to consider these different normalizations.
I think they tend to prefer an explicit Z-transformation wrt a baseline rather than dSPM or sLORETA maps.
If you think the maps might not align after projection on the template, you can smooth them a bit.
But if these are already unconstrained source maps, you don’t have to worry about this.
Francois
I played a little and it seems that unconstrained sLORETA might give me a slightly better localization than smoothed (5mm) unconstained wMNE (at least in this brain) for auditory cortex, but since I’m fuzzy on the model differences and interpretation too, maybe I’ll stick with MNE!
If I want to extract an interpretable time domain signal should I be using smoothed constrained wMNE models? (I had trouble getting time-domain signals for our earlier work, but that was quite a different signal… so I’m wondering if other people have been extracting ROI timeseries with unconstrained models or not for cortical signals!)
-Emily
Just for interest's sake...
Left: wMNE, right: sLORETA (both unconstr) - single subject
Effect of 5mm smoothing kernel on wMNE model (centre: no smoothing, right: 5mm) - single subject
First I want to insist on the fact that sLORETA is a normalized MNE solution. So it is not "MNE" vs "sLORETA", all these solutions are minimum norm solutions.
What changes is the normalization you apply on your MNE current density maps: sLORETA or dSPM or Z-transformation.
Tutorials/SourceEstimation - Brainstorm
If your goal is to do statistics across subject, it's easier to justify the use of the Z-score version.
Constrained vs unconstrained:
Constrained maps are lot easier to manipulate. Unconstrained ones are smoother and work better if the orientation of the normal to the cortex is not matching the orientation of the real currents.
I'm not sure what recommendations to give you, you should ask Sylvain for these kinds of directions.
Francois
Hi François,
Okay thanks for the model clarification and reference. I do want to do cross-subject stats so I’ll use MNE with z-score.
Based on that, I will try unconstrained for surface-based group analyses and constrained when I need to pull out time courses. (and then check with SB)
Thanks again,
Emily
