Dear Bernhard,
The answer to your last question is yes. The idea is that what SPM
really uses for source reconstruction is the crosscorrelation matrix
of channels. This matrix is computed based on all the data you
provide, be it ERPs, single trials, one condition or more. SPM also
uses the whole time period (including the baseline) for computing the
matrix. It is possible in script to limit the time period but usually
this is not necessary (unless you have something like a stimulus
artifact somewhere). The idea is that by looking at that matrix SPM
can figure out what sources are involved in the activity being looked
at. Then SPM computes the activity of those sources to fit each
specific trial. This is the output of the inversion. Given a
timecourse for each 'voxel' you can then ask different questions such
as: 'what was the power around 10 Hz in this activity between 800 and
1000 ms'? or 'was this activity larger in amplitude in condition 1
than in condition 2?' To answer these questions you don't need to do
new inversion. You just need to go to the activity which is already
computed and look at it in some particular way. This is what 'Window'
button does. Including some time segments which are not actually of
interest may be beneficial because it enables SPM to 'see' some
confounding sources which are also active in your period of interest
and thus better separate their activity from the sources you are
interested in.
As I said in my previous e-mail what we call 'induced' power
corresponds to what you call 'total' power. The concept of pure
induced power in my personal opinion is not very meaningful because
for most activities there are different degrees of phase locking to a
stimulus so there is usually no pure evoked and pure induced.
I hope this answers your questions.
Best,
Vladimir
On Fri, Aug 1, 2008 at 2:41 PM, Bernhard Spitzer
<[log in to unmask]> wrote:
> Dear SPM (and M/EEG) experts,
>
> I came across a number of comprehension questions regarding the 3D source
> reconstruction of EEG data using SPM5. I want to use SPM to reconstruct
> induced EEG Power within specified time-frequency windows for two different
> conditions. Therefore, I feed epoched single trial data(say 80 Trials for
> Cond1, 80 Trials for Cond2) into source reconstruction. After meshing coreg,
> forward model computation and model inversion, I can define TF-windows,
> chose between induced/evoked activity, and finally get niftis for further
> analysis - everything works fine and looks good.
> However, it is not fully clear to me what the resulting solutions actually
> depict. For example, when I'm interested in the sources of say, induced
> alpha 800-1200 ms after stim onset, I can define this contrast of interest
> only after all model inversion, ReML estimation and so on is already
> completed. Critically, shouldn't such contrasts of interest be specified
> before the model is actually inverted? (Given that the solution should
> specifically describe activity in this TF window). Or, put differently, what
> data is actually used for the (generic?) model inversion? Does it just
> somehow use all data, including all trials, conditions, samples, prestimulus
> interval and so on? Or, how can a single model inversion later account for
> evoked as well as for induced activity in full TF space, seperately for
> different conditions? In the MIP window, only the inverse solution's average
> (i.e. evoked) response is depicted. How can this inverse solution then
> contain (or preserve) induced activity, which I can define later on? I
> realize that when I select an induced contrast (after inversion), the
> program appears to work through each single trial (again?). What happens in
> this step? Is the model re-estimated? Or does SPM just use the initial
> inversion solution (which is somehow based on the raw signal - or on the
> average signal only?) to account for the specified TF activity in the single
> trials? Finally, does this procedure in fact image induced power only (i.e.
> total power - evoked power), or does it actually image total power, or
> anything completely different?
>
> Sorry for so many questions but manual and mailing list remain relatively
> silent about these points. All my questions may be subsumed under:
> Can I really get the specific sources of e.g., induced alpha power 800-1200
> ms after stim onset, using the source reconstruction currently implemented
> in SPM?
>
> best regards
> Bernhard Spitzer
>
|