Dear Noelle / Ji,

My apologies, you did enter the extra regressors in the correct place. The session regressors look as they do because of the high pass filter (which you have set to quite a low value).

 

It may or may not make a difference, but to improve accuracy of the filtering on the concatenated design matrix, you might wish to use the following script to estimate the high-pass filter separately for each session in the concatenated design matrix (thanks to Guillaume Flandin for his help with this). Instructions:

 

1. Create the subject’s concatenated design matrix (SPM.mat)

2. Set the original (not concatenated) number of scans in each session in the script below, and run it while in the same directory as the SPM.mat.

3. Estimate the subject’s SPM.mat.

 

load('SPM.mat');

 

%-Create backup

save('SPM_backup.mat', 'SPM', spm_get_defaults('mat.format'));

%-When concatenating sessions, SPM.nscan contains the total number

% of scans. Replace with array of number of scans per session

SPM.nscan = [84 84 84];

s = cumsum([0 SPM.nscan]);

%-High-pass filter

for i=1:numel(SPM.nscan)

    K(i) = struct('HParam', SPM.xX.K(1).HParam,...

                  'row',    s(i) + (1:SPM.nscan(i)),...

                  'RT',     SPM.xY.RT);

end

SPM.xX.K = spm_filter(K);

%-Temporal non-sphericity

SPM.xVi.Vi   = spm_Ce(SPM.nscan,0.2);

SPM.xVi.form = 'AR(0.2)';

save('SPM.mat', 'SPM', spm_get_defaults('mat.format'));

 

 

If you review the design matrix now, you may get more accurate results.

 

Best,

Peter.

 

 

Dear Noelle,

You were nearly correct. You entered the session regressors into the ‘Conditions’ rather than the ‘Regressors’ part of the batch. As a result, it has convolved each one with the haemodynamic response, which you don’t want.

 

There’s nothing in your design matrix at the moment modelling which trials are grip or tap, and which are left or right hand… Have you already identified your ROIs from another design matrix which does include this information? You’re going to need one more extra columns to create the modulatory regressors for the DCM.

 

Best,

Peter.

 

 

Dear SPM users,

 

We have an fMRI task dataset that consists of 6 sessions (runs), each composed of one of four unique tasks.  These tasks involve movement (grip or tap) of the left or right hand, with only one task per session (i.e. one session involves R hand grip, another involves L hand tap, etc.).  Each session consists of 60 volumes with TR = 3, and we are including the onsets for each task as a single condition (‘Movement’).  We are interested in modeling the effective connectivity between six ROIs (three bilateral regions), incorporating intrinsic connectivity between all regions and modulatory effects of hand movement on certain connections. 

 

After testing various scenarios with respect to the GLM design matrix and the DCM model, we still have several questions/uncertainties:

 

1)      After reading through the SPM email archives for advice on this matter, we found that it is recommended to concatenate sessions into one session in the design matrix for purposes of DCM.  In order to model the mean for each session (except the last session), we followed the procedure for creating 6 regressors outlined here: https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=spm;e9af3cf7.1409.  After estimating the GLM, our design matrix does not appear as expected (see attached).  We are interested in some advice on how to correct this output such that we have blocks of sixty 1s corresponding to each of the first five sessions and a column of 1s corresponding to the grand mean. 

2)      If the above question is unclear, we are essentially interested in any suggestions for the optimal method of constructing the GLM design matrix for purposes of DCM with multi-session data, when each session has only a single task (condition).

 

In the event that any of the above methodology is inaccurate/suboptimal, additional suggestions are welcomed.

 

Thank you,

Noelle and Ji