Dear Daniel,
If your design is not time-resolved (a.k.a. event-related) it doesn't
make sense to use a time frequency methods such as Morlet wavelets. I
would suggest to split your data into 1 sec epochs assigned to
separate conditions for the baseline and different experimental
factors. Then you should use the 'Spectrum' option in the
time-frequency tool to generate scalp x frequency images of power and
you can do stats on those. I'm not sure it's necessary to Z-score the
power across electrodes because when you do the stats you will compare
each band to the same band in another condition so even if the power
has vastly different values, that will be normalised when computing
the statistic. The only reason to Z-score is if you want to visualise
the raw power in all the bands together.
Hope this helps.
Vladimir
On Fri, Jul 22, 2022 at 5:56 PM Daniel Eckhoff
<[log in to unmask]> wrote:
>
> Dear all,
>
> In an experiment, we have a block design that includes a 1 minute baseline followed by a 1 minute experimental condition (2 factors with 2 levels). This is repeated 8 times.
>
> I would like to analyze all 5 EEG bands.
>
> What is the best way for me to perform a spectral decomposition in this case?
>
> I have seen that in some articles spectral decomposition is performed for epochs of 1 minutes length or more. Others split the long epochs into smaller 1s epochs.
>
> I would use Morlet spectral decomposition with a time window of [-60s +60s] and use LogR for the processing including a baseline correction. Does that sound plausabile?
>
> I also found this interesting paper using SPM for their EEG analysis:
> https://www.sciencedirect.com/science/article/pii/S1053811920301191
>
> They split the long epochs into 1s epochs. In addition, they normalize the nifti images using z-scores across the electrodes for each frequency step.
>
> Is this a method to improve inter-subject variability supported by SPM, or are there other methods built into SPM to address this issue?
|