I'd like to extract the peak evoked amplitude within a particular ROI
for each of my subjects, then test if this value differs between two
subgroups of the subjects.
I'm vaguely aware that normalization can be important when comparing
certain statistics between subjects, although I'm not entirely certain
when this is appropriate. I have a hunch it may be applicable now, though.
As I understand it, dspm is essentially noise-normalized mne. Is it more
appropriate to use dspm, rather than mne, when comparing peak response
amplitude between subjects?
Hi Matt,
In my opinion, for a fixed ROI, either would be fine if the
distributions are made similar for some baseline period across
subjects. For instance, if you express each subject's evoked response
in the ROI (dSPM or MNE) in terms of z-scores relative to the baseline
period i.e. something to the effect of
z(t) = (x(t) - mean(x(baseline)))/std(x(baseline))
then since all subjects are similar in the baseline period (i.e mean 0,
variance 1), the peaks in the trial period can be compared across groups
of subjects.
The preference for dSPM over MNE in my mind primarily comes from the dSPM
enabling a better ROI selection (if it is based on MEG data, rather than
apriori) since localization is typically less biased.