Dear SPM experts,
I am struggling with Bayesian belief updating between MEG and fMRI data recently. I have already known that the paper "Bayesian fusion and multimodal DCM for EEG and fMRI" suggested "...replaced the uninformative neuronal priors (in DCM of fMRI) with the posterior distribution over neuronal parameters (i.e., Ep and Cp) from the above EEG inversion", my question is how to do it properly? I tried adding Ep and Cp got from the DCM of MEG to the DCM structure of fMRI before performing 'spm_dcm_fMRI_nmm', unfortunately it did not work. Should I change the DCM.option.P? If so, may I ask how should I do it?
Another related question is that there is only one B matrix in the estimated DCM of MEG (ERP. TFM), however, in the model of fMRI there are two B matrix (forward and backward connection, superficial and deep pyramidal cells). Should I specify the DCM of MEG the same with the DCM of fMRI? If so, how should do it? It seems that the default set of spm_dcm_erp only support one B matrix.
The third question is more about Bayesian model selection and reduction. In conventional unimodality DCM we can first specify a full model and then apply BMC or BMR across all the subjects to get the Ep and Cp. However, in the multimodality DCM, should I specify a full DCM model of MEG and use BMR to get the Ep and Cp first, then feed the Ep and Cp to the full model of fMRI and use BMR again to the submodels of fMRI?
Many thanks for any suggestion.
Best,
Li Zhi
|