On Tue, May 31, 2011 at 1:19 PM, Tayaranian Hosseini P.
<[log in to unmask]> wrote:
> Dear SPMers,
> I am going to do some source reconstruction of EEG but I need some of my questions to be answered first. I have an epoched EEG file with 27 trials which are all for just one condition. I would like to reconstruct the source for each trial first and I need to see all these reconstructed images separately. Then, I will average the reconstructed images and would like to see the final reconstructed image.
> 1. When I press the invert button, it starts to reconstruct the sources without asking me if I want to average or not. Does it first average the epochs and then reconstruct the sources or first reconstruct the sources for each trial and then averages the images?
What the routine does is slightly more complicated than that. It
computes data covariance matrix based on all the trials and uses that
matrix to calculate the MAP projector which basically encodes the
information about which sources are part of the solution. Then this
projector can be used to compute the source currents from the data. If
your data is epoched the covariance matrix will be based on single
trials and might be slightly different from when your data is
> 2. In the next step (WINDOW), the question of power pops up. If I choose for example "induced", does it start the reconstruction from the beginning? If so, why this question is not asked in the "invert" step?
No, it will use the pre-computed MAP projector to compute activations
for each trial and then apply wavelet analysis to those.
> 3. In "window" step, even when I choose "trials" for power, I can not see different reconstructions and I will just see "first trial". How can I see the reconstruction of other trials?
You can see them in the images that are exported.
> 4. In "image" step, again, how is the final image computed? EEG epoch averaging or image averaging?
In the Image step, whatever was computed in the 'Window' step is saved
as an image. So if you used the 'evoked' option that would be an image
for the average ERP, if you used the 'induced' option that would be an
average of induced power over trials.
Practically, you have two options:
1) If you want to compute a reconstruction for each trial separately,
you should assign a different condition label to each trial and then
when as at the beginning of the 'Invert' step what condition to
select, you select only that trial. That way the results will not be
influenced by other trials, but if these are really single trials the
data will probably be too noisy for any meaningful source
reconstruction. Also if you want to do statistics across trials (e.g.
regression) it will work very poorly because the sources will not be
constrained to be similar. You can compute the average image and see
what that looks like, but it'd be really a suboptimal way of doing
2) You can keep the same condition label and use the 'trials' option.
Then you'll get an image per trial but the sources will be the same
for all trials. That's the option to use if you want to do some
statistical analysis across trials assuming common sources, but that'd
not be the right thing to do if you think the activated sources will
be different for each trial.