table of contents
LIPSIA     Performing a general linear regression using whitening
vwhiteglm

The program 'vwhiteglm' performs a general linear regression with pre-whitening (see literature below). It fits a linear model to fMRI time series. In a first stage, autocorrelation parameters are estimated from the least squares residuals using the Yule-Walker equations. Then, the autocorrelation parameters are used to `whiten' the data and the design matrix. In a second stage, the linear model is re-estimated using least squares on the whitened data to produce estimates of effects and their standard errors.

As input, 'vwhiteglm' requires fMRI raw data which may have been pre-processed (using the programs 'vmovcorrection', 'vslicetime', 'vfunctrans', and 'vpreprocess').

In addition, 'vwhiteglm' requires a design matrix as generated by the programs 'vgendesign', 'vcatdesign', 'vaddcovariates', or 'vgetcovariates'.

Examples for calling 'vwhiteglm'
Example 1:

vgendesign -in design.txt -out design.v -tr 2 -ntimesteps 648 -deriv 0

vwhiteglm -in ntbsm_data.v -out conimg.v -design design.v -contrast 1 -1 0

This sequence of program calls first generates a design matrix from a text file containing the design specification. Then 'vwhiteglm' computes a contrast image using the general linear model. The length of the contrast vector must coincide with the number of covariates in the design matrix. This can be checked using the program 'vcheckdesign'.

Example 2:

vgendesign -in design1.txt -out design1.v -tr 2 -ntimesteps 648 -deriv 0

vgendesign -in design2.txt -out design2.v -tr 2 -ntimesteps 648 -deriv 0

vcatdesign -in design1.v design2.v -out design.v

vwhiteglm -in ntbsm_session1.v ntbsm_session2.v -out conimg.v -design design.v -contrast 1 -1 0 1 -1 0

This sequence of program calls first generates two design matrices from two text files containing design specifications of two sessions. These two design matrices are then concatenated by 'vcatdesign' to allow for a fixed-effects analysis across sessions. Then the program 'vwhiteglm' computes a contrast image using two fMRI data files. The length of the contrast vector must coincide with the number of covariates in the design matrix 'design.v'. This can be checked using the program 'vcheckdesign'.

Further options of vwhiteglm

The default output of 'vwhiteglm' is a contrast image. A set of contrast images can be further processed in a second/third level analysis using, e.g. 'vonesample_ttest' or 'v2ndlevel'). The type of output can be changed using the option '-type'. To generate a zmap instead of a conimg, specify '-type zmap'.

vwhiteglm -in ntbsm_data.v -out zmap.v -design design.v -contrast 1 -1 0 -type zmap

The option '-order' specifies the order p of the autoregressive model. In most cases, an AR(1) model should be adequate for the data. Note that 'vwhiteglm' might be very slow for p>1.

vwhiteglm -in ntbsm_data.v -out conimg.v -design design.v -contrast 1 -1 0 -order 4

The option '-numcon' can be used for computing several contrasts in 'vwhiteglm'. For compuing the contrasts '1 -1 0', '1 0 0', and '0 1 0', 'vwhiteglm' can be called in the following way:

vwhiteglm -in ntbsm_data.v -out conimg.v -design design.v -contrast 1 -1 0 1 0 0 0 1 0 -numcon 3

In this case, the output of 'vwhiteglm' are 3 files.

The contrasts can also be specified using a text-file which containes all contrasts. In our example, this textfile could have the following form:

1 -1 0
1 0 0
0 1 0

Using this file 'contrasts.txt', 'vwhiteglm' can bge called as follows:

vwhiteglm -in ntbsm_data.v -out conimg.v -design design.v -confile contrasts.txt -numcon 3

Warning

The goodness of the estimation of the AR coefficients depends on the lengths of the fMRI time series, i.e., the goodness depends on the number of timesteps. If there are not enough timesteps to provide a good estimation of the AR coefficients, 'vwhiteglm' gives a warning message.

Warning: critical number of timesteps!

In that case, the computed statistical values are too conservative. Then, the resulting values can be even lower than the values computed with 'vcolorglm'.

Parameters of 'vwhiteglm':
-help
Prints usage information.
-in
One or more fMRT data file(s). Default: (none)
-out
Output beta file. Default: (none)
-design
File containing design matrix. Required.
-type [ conimg | tmap | zmap | Fmap ]
Type of output. Default: conimg
-numcon
Number of contrasts. Default: 1
-contrast
Contrast vector.
-confile
Contrast file.
-order
Order of AR model. Default: 1
Literature
Worsley, K.J., Liao, C., Aston, J., Petre, V., Duncan, G.H., Morales, F., Evans, A.C. (2002). "A general statistical analysis for fMRI data." NeuroImage,15,1-15.



Max Planck Institute for Human Cognitive and Brain Sciences. Further Information: lipsia@cbs.mpg.de
Copyright © 2007 Max Planck Institute for Human Cognitive and Brain Sciences. All rights reserved.