 |
 |
Dear Tammar Sorry I missed this email. I am afraid my wording was confusing, so just to make sure before I proceed: Each of our subjects has 4 sessions (runs): 2 sessions in scan 1 (pre-treatment) and two sessions in scan 2 (post-treatment). I saw the option of the PEB-of-PEBs model in Wikibooks regarding pre- and post-treatment scans. However, 1. I am not sure how to treat every two sessions of the same scan, if not concatenated. The solution you offered refers also to this? You would have one DCM per run – so four DCMs in total. Then if you pursue the PEB-of-PEBs approach, you would have a PEB model per subject. This would be fitted to the parameters of their four DCMs. I assume the difference between runs within-session is not interesting? In which case, your design matrix will have two regressors, one for the mean over runs and one for the difference between runs due to treatment: X= [1 -1 1 -1 1 1 1 1]; If you then look in the matrix PEB.Ep for any subject, you’ll have a matrix of estimated parameters, of dimension [C x 2], with one row for each of C connections. The first column is the mean connectivity over runs, and the second column is the difference in connectivity due to treatment. You can then form a (column) cell array containing all N subjects’ PEB models, and give this as the input to spm_dcm_peb, where the design matrix is simply a vector of ones of length N: PEB2 = spm_dcm_peb(PEBs); If you look in the matrix PEB2.Ep, you should find two columns of parameters. The first is the commonalities across subjects (i.e., the average connectivity across runs), and the second is the average effect of treatment across subjects. You can view these effects more easily using the GUI: spm_dcm_peb_review(PEB2); 2. Is the command "spm_dcm_average" relevant here to average two DCMs of the same scan to one DCM? and then build Peb of PEbs model to compare pre and post-treatment results pre and post across subjects? No, I expect the approach I described above is better, because it will properly separate variance across runs and variance across subjects. The function spm_dcm_average is a fixed effects average – which doesn’t distinguish different sources of variance. 3. Seghier et al. (2010) didn't concatenate the sessions in the first-level analysis, instead "extracted ROI time series were concatenated over the 2 sessions and incorporated in the DCM model". Are you familiar with this solution? I didn't understand it. I think the timeseries were extracted and concatenated, rather than concatenating the onsets, which is more typical. You can read our approach to concatenating sessions here - https://en.wikibooks.org/wiki/SPM/Concatenation . Let me know if anything is unclear. Best Peter On Tue, Apr 7, 2020 at 10:06 AM Zeidman, Peter <[log in to unmask]<mailto:[log in to unmask]>> wrote: Dear Tammar > First, thank you for the DCM PEB manual. It is very detailed and easy to follow. You’re welcome! > Our questions regard multiple-session fMRI: We have 4 runs per subject (two pre-treatment, and two post-treatment). We didn't concatenate them for good reasons. > 1. what would be the best way to treat every two runs (pre and post) as one session in the DCM analysis? > 2. If the answer goes to averaging - BMA, can we use later the PEB method at the group level? If you don’t want to concatenate the runs, that means you have separate DCMs for each one. You can have two options. A) build a PEB model with a regressor for pre- vs- post treatment, however the model won’t know that some runs are from the same subject. B) build a PEB-of-PEBs model, i.e., one PEB model per subject, and then a PEB-of-PEBs across subjects. You can read more about these two options at: https://en.wikibooks.org/wiki/SPM/Parametric_Empirical_Bayes_(PEB)#Hierarchical_experimental_designs<https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fen.wikibooks.org%2Fwiki%2FSPM%2FParametric_Empirical_Bayes_(PEB)%23Hierarchical_experimental_designs&data=02%7C01%7C%7C7a5f3dd2cb694976f1c208d7e18be925%7C1faf88fea9984c5b93c9210a11d9a5c2%7C0%7C0%7C637225863335916596&sdata=90OOHVIEgV0aHIWJweFOBWieUhtoVkHA1OMsU6P%2FVfs%3D&reserved=0> > 3. what is the minimal group size we can execute DCM PEB? There isn’t a minimum – but the precision with which you can estimate parameters will decrease as you have fewer subjects. Best Peter -- תמר טרוזמן קלינאית תקשורת, MA מ.ר. 13-138103 -- תמר טרוזמן קלינאית תקשורת, MA מ.ר. 13-138103