LISA second-level GLM (vlisa_2ndlevel)

The program vlisa_2ndlevel performs a 2nd-level (group level) inference using the general linear model including a correction for multiple comparisons using the LISA algorithm [2018_Lohmann]. The input is a list of 3D images and a 2nd-level design matrix. The output is a map thresholded such that FDR < alpha for every voxel. The default is alpha=0.05. The resulting image shows (1-FDR) so that larger values indicate higher significance.

The design file is a txt-file that follows standard conventions for 2nd-level analyses. It may for instance be generated using the tool GLM of the FSL software package, alternatively the design textfile can also be written manually. The number of rows in the design file must match the number of input images. For a detailed description see designformat_higher.

It is important to make sure that the order in which the input images are specified on the command line matches the order of the rows in the design file.

In some cases - for example repeated measures designs - permutations must be constrained to ensure exchangeability. These constraints can be specified using a file containing group IDs (option ‘-grp’). Each input image is represented by one row of this file and contains an integer ID (>=1) that represents the group to which the image belongs. Images within the same group are assumed to be exchangeable. This file corresponds to the file “.grp” of FSL-Glm. If no group file is specified then it is assumed that no exchangeability constraints exist.

Optionally, nuisance regressors can be supplied in a separate txt-file using the option ‘-nuisance’. Each column in this file corresponds to one regressor. The number of rows must match the number of input images (scans). The nuisance regressors are exempt from random permutations, and they are not included in the contrast vector. By default, the mean of each regressor is subtracted (‘-demean true’).

The parameter ‘-contrast’ is used to specify the contrast vector. The length of the contrast vector must match the number of columns of the design file. Each element of the contrast vector corresponds to a column in the design file.

Note that a region-of-interest mask is required. The mask should exclude non-brain voxels, and it may cover the entire brain. In the example below, the mask is in the file “braimmask.nii”.

Example:

vlisa_2ndlevel -in images_*.nii.gz -out result.v -mask brainmask.nii -design design.mat
     -grp design.grp -contrast 1 0 0 0 -nuisance regressors.txt

Parameters of ‘vlisa_2ndlevel’:

-help

Prints usage information.

-in

Input files.

-out

Output file.

-design

Design file.

-grp

Group Ids needed for exchangeability (optional).

-contrast

Contrast vector.

-nuisance

Nuisance regressors (optional).

-demean

Whether to subtract mean in nuisance regressors. Default: true

-alpha

FDR significance level. Default: 0.05

-perm

Number of permutations. Default: 5000

-mask

Region of interest mask.

-seed

Seed for random number generation. Default: 99402622

-radius

Bilateral parameter (radius in voxels). Default: 2

-rvar

Bilateral parameter (radiometric). Default: 2.0

-svar

Bilateral parameter (spatial). Default: 2

-filteriterations

Bilateral parameter (number of iterations). Default: 2

-cleanup

Whether to delete isolated voxels. Default: true

-j

Number of processors to use, ‘0’ to use all. Default: 0

References

[2018_Lohmann]

Lohmann G., Stelzer J., Lacosse E., Kumar V.J., Mueller K., Kuehn E., Grodd W., Scheffler K. (2018). LISA improves statistical analysis for fMRI. Nature Communications 9:4014. (link)