User Tools

Site Tools


mdm.computeglm

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revision Previous revision
mdm.computeglm [2010/11/24 06:13]
jochen updated the help
mdm.computeglm [2011/04/03 03:21] (current)
jochen added individual explanation for options
Line 66: Line 66:
         .shuflab ​  PRT labels (conditions names) to shuffle         .shuflab ​  PRT labels (conditions names) to shuffle
         .shuflabm ​ minimum number of onsets per label (1x1 or 1xL)         .shuflabm ​ minimum number of onsets per label (1x1 or 1xL)
 +        .sngtrial ​ single-trial GLM (only with PRTs, default: false)
 +        .sngtskip ​ condition list to skip during single-trial conversion
         .subsel ​   cell array with subject IDs to work on         .subsel ​   cell array with subject IDs to work on
         .tfilter ​  add filter regressors to SDMs (cut-off in secs)         .tfilter ​  add filter regressors to SDMs (cut-off in secs)
         .tfilttype temporal filter type (either '​dct'​ or {'​fourier'​})         .tfilttype temporal filter type (either '​dct'​ or {'​fourier'​})
 +        .transio ​  ​boolean flag, if true, save GLM and use transio
         .vweight ​  ​combine runs/​studies variance-weighted (default: false)         .vweight ​  ​combine runs/​studies variance-weighted (default: false)
         .xconfound just as motpars, but without restriction on number         .xconfound just as motpars, but without restriction on number
Line 83: Line 86:
        for .ppi to work, the model filenames must be PRTs!        for .ppi to work, the model filenames must be PRTs!
        all additional fields for the call to PRT::​CreateSDM are supported!</​file>​        all additional fields for the call to PRT::​CreateSDM are supported!</​file>​
 +
 +===== Options =====
 +
 +==== .globsigs ====
 +Setting this field with either
 +  * a number (other than 0) will add as many "​global signals"​ to each design matrix before regression, whereas for 1 this will be the overall mean, for 2 the mean represents left and right hemisphere separately, and for 3 or higher a PCA is performed and as many components as requested added to each design matrix
 +  * a single VMR object or a cell array of VMR objects will add as many "​global signals"​ (from all non-0 VMR voxels, sampled at the VTC coordinates) as requested
 +
 +==== .ithresh ====
 +This field controls any explicit masking based on threshold (mean timecourse value). If left empty/​unset,​ no threshold masking occurs (full datasets).
 +
 +==== .loadglm ====
 +If set to ''​true''​ (and a valid GLM filename is provided in ''​.outfile''​!),​ an already existing GLM file is loaded and then additional subjects added (or replaced) as configured.
 +
 +==== .mask ====
 +Either a [[xff - MSK format|MSK object]] fitting the VTC spatial layout or a binary mask sufficing the same condition can be supplied to compute the GLM only within those voxels that are non zero/not false in the mask.
 +
 +==== .motpars ====
 +If set to a cell array of filenames with size ''​mdm.NrOfStudies''​-by-1,​ the regression will contain any motion parameters found in the referenced files. If set to ''​true''​ any files found under ''​mdm.RunTimeVars.MotionParameters''​ will be used instead.
 +
 +==== .motparsd ====
 +If set to true any motion parameters will be added as is and their first derivative will also be part of the model.
 +
 +==== .motparsq ====
 +If set to true any motion parameters will be added as is and their squared (and mean-removed) version will also be part of the model.
 +
 +==== .ndcreg ====
 +If set to an integer number greater than 0, [[xff - PRT format|PRTs]] will be converted to design matrices using deconvolution designs (stick functions for each deconvolution/​volume regressor; field is passed to [[prt.CreateSDM|PRT::​CreateSDM]] under the same fieldname).
 +
 +==== .orthconf ====
 +If set to true, confounds (motion parameters, filters, global timecourses,​ etc.) will be orthogonalized against one another (which doesn'​t change the outcome of the regression equation as far as the betas of interest is concerned, but can improve the behavior of the underlying functions used, when the design matrix is inverted before computing the betas).
 +
 +==== .outfile ====
 +Can be set to a filename under which the GLM will be stored at the end of the computation (and, for robust regression and enabled transio after each subject). If ''​.loadglm''​ is also set to true and file already exists, the GLM will be loaded first, and only selected subjects from the MDM that are not already computed will be added at the end (unless ''​.redo''​ is set to true, in which case those subjects'​ data, if it is already part of the GLM, will be replaced). See also ''​.loadglm''​ and ''​.redo''​ options.
 +
 +==== .ppicond ====
 +Can be set to a cell array with condition and/or contrast combinations,​ which will be used by adding an interaction term between an extracted timecourse (from a PPI seed region) and a (contrast of) regressor(s). To specify a contrast, condition names need to be put into a string like this:
 +  * '''​condition_A + condition_B'''​ (sum of two conditions)
 +  * '''​condition_A > condition_B'''​ (difference of two conditions)
 +  * '''​condition_A + condition_B > condition_C + condition_D'''​ (difference of sums, can be used for a condition interaction term)
 +See also ''​.ppivoi''​ and ''​.ppivoiidx''​ options. ​
 +
 +==== .ppirob ====
 +**This option is experimental and not fully tested!**
 +
 +Can be used to remove outliers of the PPI timecourse (which then improves RFX properties of beta estimates associated with the PPI regressors).
 +
 +==== .ppivoi ====
 +Must be set to a valid [[xff - VOI format|VOI object]] for PPI to be used.
 +
 +==== .ppivoiidx ====
 +Can be set to an index to select a specific VOI from the VOI object/file in option ''​.ppivoi''​ (default is 1). Timecourses will be extracted from this selected VOI.
 +
 +==== .prtpnorm ====
 +If set to true, normalize and parameters in PRT files before computing the parametric conditional regressors (is passed as ''​.pnorm''​ into the [[prt.CreateSDM|PRT::​CreateSDM]] method).
 +
 +==== .redo ====
 +If set to ''​true''​ and ''​.loadglm''​ is also set to ''​true''​ **and** an existing GLM file is provided in the ''​.outfile''​ option, any selected subjects (see ''​.subsel''​ option) will be overwritten in the GLM file.
 +
 +==== .regdiff ====
 +**This option is experimental and not fully tested!**
 +
 +If set to ''​true''​ design matrices and data are first passed through ''​diff''​ before regression occurs (in effect regression the first derivative of data on the first derivative of the design).
 +
 +==== .restcond ====
 +This option is passed as ''​.rcond''​ into [[prt.CreateSDM|PRT::​CreateSDM]] calls and defaults to empty (don't remove any condition as rest).
 +
 +==== .robust ====
 +If set to ''​true''​ performs robust regression on all design matrices (**very slow!**) using the [[fitrobustbisquare_img]] function.
 +
 +==== .savesdms ====
 +Can be used to store created/​altered [[xff - SDM format|SDM files]], and must be set to a non-empty string containing '''​.sdm'''​ at its end.
 +
 +==== .showsdms ====
 +If set to either '''​image''',​ '''​ortho''',​ or '''​plot'''​ calls [[sdm.ShowDesign|SDM::​ShowDesign]] before each run's regression starts (visualizing the design matrix of the currently regressed run in the desired way).
 +
 +==== .shuflab ====
 +**This option is experimental and not fully tested!**
 +
 +Allows to re-generate the GLM under the null hypothesis of equal effect of stimuli across conditions by shuffling the labels (onsets across conditions).
 +
 +==== .shuflabm ====
 +Sub-option for ''​.shuflab'',​ ensures that each condition with onsets retains at least this number of onsets.
 +
 +==== .sngtrial ====
 +If set to ''​true''​ (and the MDM object references PRT objects), calls [[prt.ConvertToSingleTrial|PRT::​ConvertToSingleTrial]] prior to converting protocols into design matrices. Uses the ''​.sngtskip''​ option.
 +
 +==== .sngtskip ====
 +Instructs the [[prt.ConvertToSingleTrial|PRT::​ConvertToSingleTrial]] method which conditions are to be left unchanged (e.g. onsets of cues, fixation points, active baseline stimuli, etc.)
 +
 +==== .subsel ====
 +Cell array of subject IDs, sub-selects subjects from those referenced by the MDM file (see [[#​requirements|Requirements section]] above for more info).
 +
 +==== .tfilter ====
 +Sets the temporal filtering cutoff in seconds, so that filtering regressors will be added to the design matrix.
 +
 +==== .tfilttype ====
 +Sets the temporal filtering type, one of '''​dct'''​ (for Discrete Cosine Transform) or '''​fourier'''​ (Fourier basis set).
 +
 +==== .transio ====
 +If set to ''​true''​ and ''​.outfile''​ is set to a valid GLM file, will use [[@transio]] access to store data into the output file. This is necessary in case the output file is very large (bigger than 2GB, for instance), so that no out-of-memory error occurs.
 +
 +==== .vweight ====
 +If set to ''​true''​ beta weights of identically named conditions across runs of the same subject will not simply be combined as a sum but will be weighted by the square root of the variance estimate of the design (reciprocal value of the diagonal values of the inverted design matrix), which is roughly equivalent to computing a single regression over all of one subject'​s runs.
 +
 +==== .xconfound ====
 +Can be set to a cell array with filenames of size ''​mdm.NrOfStudies''​-by-1,​ in which case additional regressors of no interest (confounds) are read from those files and added to the design matrix (e.g. heart-rate data, eye-tracking output, etc.). The files' content must match the number of volumes of the timecourse files!
  
 ===== Usage examples ===== ===== Usage examples =====
mdm.computeglm.txt ยท Last modified: 2011/04/03 03:21 by jochen