Dear all,

Seems I made some mistakes when publishing the message. No text appeared in the forum. Maybe let me describe my problem again here.

I'm now using SPM12 to do spectral DCM with some resting state fMRI data for single subject. I followed the example of chapter 38 in the manual as I didn't have any experience with SPM and DCM. Having managed to generate the DCM results, I found that the fitness of the model (simply revisit the fit of the model by reviewing the "Cross-spectra(BOLD)") was not good. You can have a look at the figure showing the fitness in attachment.

Is this the final result of sDCM in single subject? If it is, can I rely on the result and use it? Seems the predictions are quite far from the measured cross spectra in the figure...

To make sure that I myself have done correctly (actually I'm not sure...), I simply clarify the steps I did during the process here:
 
1. As I have processed the data in AFNI, I then specify the fMRI model and estimate using the processed data to get the SPM.mat and skipped the regression part of WM and CSF, which were the steps in the manual.
2. However, error occurred when estimating the model saying that something wrong with the spm_est_non_sphericity(v6913) (line 208) "Please check your data: There are no significant voxels." I searched the SPM list and found some solutions to the problem, which may be related to the analysis mask . Therefore, I specified a brain mask and disable threshold masking with SPM.xM.TH = -Inf(size(SPM.xM.TH)) (following the answer: https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=spm;12c654c6.1801). After doing that, I then estimated again with the changed SPM.mat but no more fMRI model specification.
3. After estimating, I extracted the time series of 6 regions (I have a question here, so just use the mask.nii generated by the above steps right? not to use the mask that I specified before, following the 38.2.2 in the manual.)
4. The next is to specify and estimate the DCM: I leaved the slice timing as default which is 1, and my TE was 0.03, then defined the fully connected model, chose modulary effects as bilinear, one state region, no stochastic effects and no centre input and fit CSD. After defining DCM, I just click "estimate (cross-spectra)" to estimate DCM.

However, when I reviewed the results after convergence, I just found that Cross-spectra (BOLD) was not good, some were even reversed. I'm not sure if there is anything wrong with my DCM specifications. Or I can rely on the results since they already did the best fitting to get the endogenous couplings?

Can anyone please help solve my puzzle? Many thanks!


Best regards,
Tracy