gaetan yvert wrote:
> Dear SPM list,
>
> I am working with sources reconstruction on eeg evoked potentials
> thanks to the greatfull toolbox of SPM.
> I have some questions to improve my study. I am working on a paradigm
> with 3 conditions and I would like to compare
> the sources creating the evoked potentials.
>
> 1) I have two possibilities : either to reconstruct the sources of
> each condition and then apply statistic methods for
> contrasting them or to reconstruct the contrast of my evoked potential.
> What will be the difference between these both approaches?
The two approaches will differ, because the inversion is non-linear (the
hyperparameters have a nonlinear dependency on the data). That is
inv(A-B) ~= inv(A) - inv(B). The pros and cons of each approach are
discussed in Henson et al (2007), Neuroimage.
> Which one will be the most robust?
Not sure this can be simply answered, but inv(A) - inv(B) is often
preferred because statistical tests on the results are simpler, since
inv(A-B) will produce only positive source power (in SPM), testing of
which probably requires non-parametric approaches.
> 2) I would like to extract the temporal activity of my regions of
> interest. I did not find in spm list a tutorial to do this.
> Which function should I use?
There may be a button for this somewhere, but for single voxels, it is
quite simple to find the vertex nearest an MNI coordinate (via
D.inv{val}.mesh.tess_mni.vert) and then extract its corresponding
temporal mode values in D.inv{val}.inverse.J{c} for each condition c,
and post-multiply by those temporal modes themselves in
D.inv{val}.inverse.T (transposed) - see spm_eeg_invert_display.m
> 3) To improve my statistics and the robustness of my study, I will use
> a group reconstruction, I have some questions about it.
> What happen if there is an outlier in the subjects that I put in my
> group reconstruction? How can I detect him?
Not sure how bad effects would be on group reconstruction, but given
that the reconstruction is based only on the data and the forward model,
you can detect outliers in the data in whatever way you would normally
do when reviewing the data (detecting "bad" forward models is more
difficult, but obviously check the meshes and the data registration)
Rik
>
> Thank you very much for your help
>
> Gaetan
>
> ------------------------------------------------------------------------
> Vous cherchez l'intégrale des clips de Michael Jackson ? Bing !
> Trouvez ! <http://www.bing.com/videos/search?q=Michael+Jackson&FORM=MVDE6>
--
-------------------------------------------------------
Dr Richard Henson
MRC Cognition & Brain Sciences Unit
15 Chaucer Road
Cambridge
CB2 7EF, UK
Office: +44 (0)1223 355 294 x522
Mob: +44 (0)794 1377 345
Fax: +44 (0)1223 359 062
http://www.mrc-cbu.cam.ac.uk/people/rik.henson/personal
-------------------------------------------------------
