Dear Farouk,
There are several steps in the inversion algorithm that change the
scaling of the data which makes it difficult to recover meaningful
units. If you are interested in studying the algorithm there is a
simplified version of the code that does not allow group inversion and
multimodal fusion but retains meaningful scaling. I can put you in
touch with people who are working on it.
Best,
Vladimir
On Sat, Jul 2, 2011 at 1:28 PM, Farouk Nathoo <[log in to unmask]> wrote:
> Hi,
>
> I am new to SPM and am performing a simulation study to examine the bias of
> estimated responses in M/EEG source reconstruction using some of the
> hierarchical Bayes models implemented in SPM. Having computed the leadfield
> matrix "L", I fix the source activity in a matrix "S" (with most rows of S
> being zero) and simulate data according to
>
> Y = L*S + E
>
> where E is simulated Gaussian sensor noise. In my script, I use the SPM
> function spm_eeg_invert(D) to fit the model and then extract the
> reconstructed source estimates using the command:
>
> S.est = D.inv{1}.inverse.J{1}*D.inv{1}.inverse.T'
>
> Comparing the estimates in "S.est" to the true values in "S", they seem to
> capture the temporal dynamics, but it seems as though these estimates are
> computed on a very different scale than the true values I have set in "S". I
> am not sure how to rescale these estimates, so that they will be on the same
> scale as the original "S" values. Any help would be greatly appreciated.
>
> Many thanks.
> Farouk
>
|