GLM Options

Predictor scaling

When predictors are defined over short time intervals, the resulting amplitude after application of the hemodynamic response function can be very low. In order to keep the predictors visible, they are now scaled appropriately as default in such a way, that the maximum value of the predictor is mapped to a value of "1.0". The identified scaling parameter for mapping the maximum value is also applied to any other value keeping relative amplitudes intact as without scaling. Automatic predictor scaling can be turned off using the "Scale predictor [0-1]" option in the "HRF and mask functions" of the "Single Study GLM Options" dialog.

Via scripting, the predictor scaling can be set via the property

 

 

The three parameters of the ScalePredictorValues( <parameter 1>, <parameter 2>, <parameter 3>) specify the following settings:

 

1. Name of predictor

2. Maximum value

3. Scale only when values are larger than zero

 

 

 

 

Baseline-Only z-Transformation

When running multi-study GLMs, it is often useful to perform a z-transformation for the voxel time courses from different runs. Such a transformation normalizes the mean of a study time course to zero and the variance to a value of one. This transformation equalizes the potential contribution of each study for explaining the overall variance of a signal time course. One disadvantage of z-normalization is that it diminishes real differences in the effect size of different runs from different subjects. To avoid this effect, a variant of z-normalization is available, which normalizes using only those values constituting the base line condition. This results in a z normalization based solely on the noise variance and leads to effect estimations which are expressed relative to the noise variance. This alternative z-normalization approach is available both for multi-run GLMs (see "Multi-Run GLM Options" dialog) as well as for ROI GLMs (see "ROI GLM Specifications" dialog).

Via scripting, the z-transformation can be set via the document property ZTransformStudies.

 

 

 

For multi-subject studies the ZTransformStudiesBaselineOnly is available.

 

 

 

Percent Signal Change Time Course Normalization

This new option allows to normalize a voxel's or ROI's time course in such a way that the mean signal value will be transformed to a value of 100 and the individual values will be fluctuating around that mean as percent signal deviations. If, for example, the original mean of a time course is 600 and an individual value is 612, the resulting value will be 102. This normalization approach appears to be well suited to integrate fMRI data from different subjects in a multi-subject GLM because it retains better individual effect size differences across subjects than z-Normalization. Percent Signal Change normalization is therefore now the default normalization option for the new RFX GLM, but can be also chosen for the standard multi-study GLM as well as for multi-study ROI analyses. If turned on, percent signal change normalization is applied separately for each run referenced in multi-subject design matrix files.

Via scripting, the percent signal change time course normalization can be set via the document property PSCTransformStudies.

 

Correction for autocorrelation in the residuals

Autocorrelation in the noise can be corrected via the property CorrectForSerialCorrelations. An AR(1) model is then applied to filter the fitted data.