Here is an answer from Gareth.
Best,
Vladimir
> From: Barnes, Gareth
> Sent: 20 June 2012 18:28
> To: John D. Griffiths; [log in to unmask]
> Subject: RE: [SPM] extracting dipole waveforms: where in the D struct are the fitted location and orientation params?
>
> Dear John
> The Bayesian dipole fit in SPM requires prior information about the source moments. That means that it fits an instant in time or an average over a time window. IN contrast, many other fitting algorithms fit the location (3 parameters) and then estimate the moments (without any prior constraints on the magnitude) over time through the pseudo inverse.
> I have some code that I can send you to extract moments over time, assuming that the dipole model remains valid outside the window you fit. (as it will be an extraploation)
> But if you really want to look at the time series you might try a DCM for evoked analysis where the time series is fit (based on priors on the temporal behaviour); or revert to a simpler method using the pseudo inverse.
> best
> gareth
>
>
> ________________________________________
> From: SPM (Statistical Parametric Mapping) [[log in to unmask]] on behalf of John D. Griffiths [[log in to unmask]]
> Sent: 20 June 2012 18:19
> To: [log in to unmask]
> Subject: [SPM] extracting dipole waveforms: where in the D struct are the fitted location and orientation params?
>
> Dear list,
>
> I would like to extract time series from fitted dipole models.
>
> Please correct my if I'm wrong but I cannot find a way of doing this **with all the estimated parameters** directly, since the location and orientation parameters need to be re-entered in the .dipoles struct argument to spm_eeg_dipole_waveforms.
>
> So it seems in order to extract waveforms from the fitted dipole models I need to first obtain the estimated orientation and location parameters from the D struct. Where would I find them? (poking around it looks likely to be somewhere around D.inv{1}.inverse...)
>
> (by way of context: I did a dipole fit based on the results of an MSP source analysis, because from previous conversations in this forum on the topic of extracting source time series this seemed like the most sensible way to avoid having to arbitrarily combine the three xyz timeseries for each voxel when the orientations are unknown. Is there anything problematic about this strategy?)
>
>
>
> Thanks,
> John
>
>
|