[Mne_analysis] Time-resolved connectivity in source space with narrow time window of interest

Eric Larson larson.eric.d at gmail.com
Fri Oct 23 14:20:34 EDT 2020
Search archives:

        External Email - Use Caution        

>
> However, my time window of interest in connectivity is shortly after the
> event, specifically from 0.02 s to 0.08 s. Using this time window, and
> comparing wPLI measures in a pre-post fashion, I have a statistical
> significant effect between in the alpha band (9 - 12 Hz).
>

An interval of 0.06 seconds corresponds to 1 full cycle of a ~16 Hz
sinusoid. So this interval contains less than one full cycle of a 12 Hz
sinusoid and closer to just half a cycle of a 9 Hz sinusoid. A general
signal processing rule of thumb is that you should have ~5 cycles for a
stable estimate of frequency content. I would assume 1 cycle would be a
bare minimum for things to have a chance to work.

However, you also used the CWT function, so really the CWT data for the
"0.02 to 0.08" interval will I think be computed using a longer segment of
the epochs (extending beyond these limits) depending on how you set
`cwt_n_cycles` (assuming you didn't actually crop your instance to just
this interval). So when you say "0.02 to 0.08", that's not really the
underlying window -- for example at 9 Hz if you used `cwt_n_cycles` to be
7, then the Morlet wavelet will span 7 cycles of 9 Hz, or ~0.77 sec,
meaning your interval is really -0.37 to 0.47 seconds, weighted toward the
center of this interval because of the Morlet kernel shape.

So I would say: keep in mind what's actually being used in the computation
based on the parameters you passed, and interpret accordingly. If you look
at the cwt TFR output, you should be able to see some smoothness as a
function of time that will help confirm your interpretations.

Ideally, I would like to know the source-level connectivity information of
> this significant finding.
>

Generally speaking, whatever connectivity measures you compute in sensor
space can be computed in source space, and vice-versa (assuming your
sensor-to-source transformation stays linear). The interpretation is
usually more straightforward in source space because the source time
courses are meant to be more properly resolved to neural sources, as
opposed to being mixed together by the brain-to-sensor physical propagation
and source additivity. (This is one idealized interpretation of source
space data anyway.)

My 2c,
Eric
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20201023/b73212e3/attachment.html 


More information about the Mne_analysis mailing list