To combine several subjects' data into one compound analysis, it is custom to compute random-effects statistics on the second level. The MDM file format of BrainVoyager QX references the required time-course files (VTC or MTC) and the respective information containing the model (onsets in PRT files or regressors in SDM files).

This method then allows to perform the regression on a session-by-session basis and combine the results in one compound GLM object.

This function is currently implemented for BrainVoyager QX's GLM format only. The following files must be available:

• VTC (or MTC) files for all runs of all subjects (e.g. the ones automatically created by spm5_preprojobs, for how to name files see below)
• PRT (or the derived SDM) files for all runs (if subjects share this information, only one file is needed for each uniquely specified model!)
• optionally the realignment parameter files (supported formats: TXT for SPM rp*.txt files, SDM for BV's motion correction output)

Note: The combination of within-subjects regressors (e.g. same condition across several runs) is done by identifying the subject ID as the particle of the time-course filename before the first underscore character! This means that VTC (or MTC) filenames must be named as follows:

• SUBJ1_RUNorTASK1.vtc
• SUBJ1_RUNorTASK2.vtc
• SUBJ1_ …
• SUBJ2_RUNorTASK1.vtc
• SUBJ2_RUNorTASK2.vtc
• …

The number of runs per subject can be different (although, for the sake of having the error on beta estimates coming from the same distribution, it is suggested to use roughly the same number of volumes–and usually runs–for all subjects), and also the names after the first underscore need not match across subjects (in fact, it needn't be unique, which of course is highly recommended for the user to be able to distinguish time-course files!). The following is thus a valid naming scheme:

jd3512_r48217.vtc
jd3512_r48218.vtc
jd3512_r48219.vtc
pk5199_r89112.vtc
pk5199_r89113.vtc

 MDM::ComputeGLM  - compute a GLM from an MDM file
 
 FORMAT:       glm = mdm.ComputeGLM([options])
 
 Input fields:
 
       options     optional 1x1 struct with fields
        .globsigs  add global signals as confound, one of
                   0 - none
                   1 - entire dataset (above threshold/within mask)
                   2 - two (one per hemisphere, split at BV Z=128)
                   3 or more, perform PCA of time courses and first N
                   xff object(s), extract average time course from masks
        .ithresh   intensity threshold, default: 100
        .loadglm   boolean flag, load GLM file named in .outfile
        .mask      optional masking, default: no mask (for now only VTC)
        .motpars   motion parameters (Sx1 cell array with sdm/txt files)
        .motparsd  also add diff of motion parameters (default: false)
        .motparsq  also add squared motion parameters (default: false)
        .ndcreg    if set > 0, perform deconvolution (only with PRTs!)
        .orthconf  orthogonalize confounds (and motion parameters, true)
        .outfile   output filename of GLM file, default: no saving
        .ppicond   list of regressors (or differences) to interact
        .ppirob    perform robust regression on VOI timecourse and remove
                   outliers from timecourse/model (threshold, default: 0)
        .ppivoi    VOI object used to extract time-course from
        .ppivoiidx intra-VOI-object index (default: 1)
        .prtpnorm  normalize parameters of PRT.Conds (true)
        .redo      selected subjects will be overwritten (default: false)
        .regdiff   flag, regress first discreet derivatives (diff) instead
        .restcond  remove rest condition (rest cond. name, default: '')
        .robust    perform robust instead of OLS regression
        .savesdms  token, if not empty, save on-the-fly SDMs (e.g. '.sdm')
        .showsdms  token, passed to SDM::ShowDesign (if valid)
        .shuflab   PRT labels (conditions names) to shuffle
        .shuflabm  minimum number of onsets per label (1x1 or 1xL)
        .subsel    cell array with subject IDs to work on
        .tfilter   add filter regressors to SDMs (cut-off in secs)
        .tfilttype temporal filter type (either 'dct' or {'fourier'})
        .vweight   combine runs/studies variance-weighted (default: false)
        .xconfound just as motpars, but without restriction on number
 
 Output fields:
 
       glm         GLM object
 
 Note: if RFX flag in MDM is set to true, predictor separation will be
       set to "Subjects".
       if .outfile is given, GLM is saved after each subject for robust
       regression models (to allow later continuation of crashed/broken
       computation using the .subsel field)
       for .ppi to work, the model filenames must be PRTs!
       all additional fields for the call to PRT::CreateSDM are supported!

Starting from scratch, the user has to do the following steps:

• locate the filenames of the VTC (or MTC and SSM/TSM) files to be regressed
• locate the filenames of the PRT (or SDM) files to be used as regressors (model)
• optionally locate the filenames of the realignment parameter files
• make settings in the MDM structure
• pass optional settings to the ComputeGLM method call

mdm_computeglm_sample.m

% locate filenames...
cd /Projects/MAN1/Imaging/Subjects
vtcs = findfiles([pwd '/M*/func'], '*.vtc', 'depth=1');
prts = findfiles([pwd '/M*/model'], '*.prt', 'depth=1');
if numel(vtcs) ~= numel(prts)
    error('Number of VTCs and PRTs must match!');
end
 
% locate realignment parameter files
rps = findfiles([pwd '/M*/func/r*'], 'rp*.txt', 'depth=1');
 
% create MDM
mdm = xff('new:mdm');
mdm.RFX_GLM = 1;
mdm.PSCTransformation = 1;
mdm.NrOfStudies = numel(vtcs);
mdm.XTC_RTC = [vtcs(:), prts(:)];
 
% optionally save MDM, .SaveAs without argument opens a dialog!
% mdm.SaveAs;
 
% ComputeGLM options
opts = struct( ...
    'motpars',   {rps(:)}, ...
    'restcond',  'rest', ...
    'robust',    true, ...
    'tfilter',   160, ...
    'tfilttype', 'fourier');
 
% compute GLM (robust takes a **LONG TIME** ! e.g. for each 200-Volume
% run with 15 regressors between 30 and 60 minutes, depending on the
% hardware used)
glm = mdm.ComputeGLM(opts);
 
% save GLM
glm.SaveAs;