Dear Vladimir,
Would you mind answering me one more question: what is the unit of the "Dipole Intensity" shown in the SPM8: Graphics after running the spm_eeg_dipole_rv.m? Many thanks!
Bests,
--------------
Sun Delin
>Dear Sun,
>
>Your analysis strategy sounds very sensible. The problem is, there is
>no way to positively prove that your dipole solution is the best
>possible one. Each dipole model in VB-ECD is just a hypothesis, you
>can compare those hypotheses to each other and say that one is more
>likely given the data but you cannot say that one is definitely true.
>Reporting free energy for a single solution is not very meaningful or
>informative. I would suggest that you formulate a set of possible
>dipole models (e.g. by starting just with those fMRI peaks that are
>most likely to contribute to the ERP and gradually adding others), do
>model comparison and report what the best model was. It would be more
>convincing if the best model is not the most complex one, i.e. you can
>show that adding some extra dipoles does not make the model better.
>
>The F values are stored in D.inv struct . I think you can also get
>them if you open your dataset in the reviewing tool and go to the inv
>tab. If you fit the models to every subject you can make a subject x
>model table of Fs and load it directly into the Bayesian Model
>Selection tool (accessible via Batch).
>
>Best,
>
>Vladimir
>
>On Fri, Dec 28, 2012 at 8:39 AM, Sun Delin <[log in to unmask]> wrote:
>> Dear SPM experts,
>>
>> I have both fMRI and ERP data from a same group of participants. For the contrast of interest, for example, condition A - condition B, fMRI provided some significant clusters (A>B or A<B); whilst ERP also showed significant difference within some time windows. I recruited the peak locations of the fMRI activation as spatial priors in ERP source localization by using the VB-ECD method. The output after 10 iterations gave an optimal solution which could explain more than 90% of variance. However, I do not know what I should report in an article, percentage of variance explained, goodness of fit, free energy, log evidence or something else? Which is the best index to state that "the brain activation within this time window originates from this brain region"? Moreover, how can I extract these information? The results were shown on the SPM graphy, but they went away when I pressed some other buttons.
>> Could you please give me some advice? Many thanks!
|