Dear Ed,
This is not a valid way of doing things as your PLV values are highly
non-linear with respect to the original source activities which
grossly violates the assumptions of all the source-localisation
methods. The only way to localise PLV is to get source activity from
the raw data first and then compute PLV of that source activity. In
SPM this can be done using the source extraction functionality but
only for some areas of interest rather than for the whole brain.
Theoretically you could also do it for the whole brain and make an
image but then you'd have to write some custom code.
On the other hand high PLV values by definition result in ERPs so I
wouldn't expect the PLV effects to be expressed at locations different
from your ERP sources. Therefore, using the source extraction strategy
with your ERP sources as ROIs is a well-justified strategy.
Best,
Vladimir
On Wed, Aug 15, 2012 at 1:54 PM, Ediz Sohoglu
<[log in to unmask]> wrote:
> Dear SPMers,
>
> Using SPM’s source reconstruction procedure, I have been able to localize
> time-frequency (power) effects occurring in my sensor-space data.
>
> I also have significant phase locking value (plv) effects (across trials) in
> sensor space that I have tried to source localize. I am aware that SPM has
> not been specifically designed to do this so I am wondering what your
> opinion is of the approach I have taken:
>
> 1) Averaged my mtph_.mat files (there is one for each subject) across the
> frequencies in which I have significant plv effects. The resulting files
> contain the plv across time for each sensor and condition. In terms of
> dimensions, these are exactly the same as evoked response files (i.e.
> sensors X time X conditions).
>
> 2) Submitted the resulting files to spm_eeg_invert using IID and group
> inversion options. For the pre-filtering option, I have specified lpf as 0
> Hz and hpf as 100 Hz (because I have already isolated the desired frequency
> range in step 1).
>
> The results look sensible to me and are comparable with previous source
> inversions I have performed with this dataset. But I am aware this is not a
> conventional approach and would be grateful for your opinion.
>
> Thanks in advance, Ed
|