Hi Rik,
On Thu, Mar 18, 2010 at 11:25 AM, Rik Henson
<[log in to unmask]> wrote:
>
> is there an easy way in SPM8 to
> estimate timecourses for specified dipole locations (even if there are three
> timecourses per possible dipole orientation)?
If you know the locations and orientations you can use a function in
MEEGtools (spm_eeg_dipole_waveforms, if I remember correctly). You
could also add the options to do SVD to get only one waveform per
source to that function. That'd be a helpful improvement.
>
> Or should I be using a beamformer in such cases (eg to get an estimate of
> "raw" timecourses at the four locations of my DCM model, ie with an over-
> rather than under-determined inversion)? If so, what would the difference in
> timecourses be for four locations when 1) modelled by four simultaneous
> dipoles, and 2) modelled by a beamformer - presumably only the latter
> removes "leakage" (cross-talk) from locations outside the four of interest?
> (but doesn't a beamformer, if applied separately to each location, also
> remove cross-talk from one dipole location to another, which is not
> necessarily wanted?).
>
Usually beamformer filters for each location are computed
independently of other locations so it doesn't matter if you extract
time course for each location separately or together. If your source
model is exactly precise and includes all the sources, it will be as
good as beamforming but that never happens. So signal extracted with
inverse of the leadfields will usually be more contaminated by other
sources, but this is not always critical. The disadvantage of
beamforming is that it is more sensitive to having exactly the right
locations and correct head model and coregistration. You can find the
optimal trade-off between these disadvantages by increasing the
regularization in the beamformer.
> Why would one bother applying the montage function to remove spatial
> components identified by ICA as artifacts, if they had been projected out of
> the data already (eg via EEGLAB)? Presumably the real issue here is to
> project these spatial modes out of the leadfield matrix, which the montage
> function doesn't affect in SPM8 (does it? Eg does R matrix in spm_eeg_invert
> get linked to any montage applied?). (Note that, even though our inversion
> methods reduce the dimensionality, isn't it still possible that some of the
> spatial modes identified by ICA are large enough to remain in the reduced
> leadfield space? (even though, in that case, they are probably *not*
> artifacts in the first place!!))
>
The montage function updates the MEG sensor structure, specifically
the 'tra' matrix. This results in the artefact subspace being absent
from the leadfields when those are computed. In particular this
prevents the problematic situation you described.
Vladimir
|