B. Run Statistical Analysis

FIRST uses the standard FSL "randomise" non parametric statistical engine to calculate statistical values and perform correction for multiple comparisons. Since there are different types of corrections for multiple comparisons, it is recommended to create method-specific subfolders of the design subfolders within which to run randomise and save resutls. 

The shape analysis subfodlers contain a text file with example scripts. These just need to be run from the design subfolders.

The following gives background information.

Running Analysis

Randomize is a combination of "Estimate" and Results" in SPM in that it estimates a model and does a statistical test on a contrast. The inputs (analysis and mask) are the outputs from the previous vertex analysis step. Just as in SPM "Results" where we put in statistical tests and thresholds, we do the same with randomise (see here for initial details). The approach is to to look at bi-directional effects (--fonly), after de-meaning the data (-D). See guide (see Running vertex analysis) for details. 

The "-o" indicates the output file name, so best practice is to label this with the test parameters (e.g., -o L_Hipp_vertexMNI_2sample_rand_F3 ). However, the example scripts only use the region and vertex type in the name ("L_Hipp_vertexMNI_rand"), in order to allow the same commands to be used for multiple designs.

Step 1: create a correction-specific subfolder, i.e., one of vox, tfce, clustere, or clusterm.

Step 2: (in Linux from here on) open a terminal in that subfolder

Step 3: run the F-test; this will perform the correction for multiple comparisons.

[user@localhost vox]$ randomise -i ../R_Amyg_vertexMNI.nii.gz -m ../R_Amyg_vertexMNI_mask.nii.gz -o R_Amyg_vertexMNI_randN5000 -d ../design.mat -t ../design.con -f ../design.fts --fonly -n 5000 -D -x

Options: -i input data from all subjects, -m mask, -o output name; Note use "N5000" to distinguish this from the t test (next). Always need -d design matrix, t contrast, and for F test the f contrasts, --fonly does only F test, -D de-means, and -x does voxel correction.

[user@localhost vox]$ randomise -i ../R_Amyg_vertexMNI.nii.gz -m ../R_Amyg_vertexMNI_mask.nii.gz -o R_Amyg_vertexMNI_randN1 -d ../design.mat -t ../design.con -n 1 --glm_output -D -x

This can be repeated for the three other correction methods; note that the clustere and clusterm need F and t contrast thresholds specified.

Output Multiple comparison options: (see table in randomise User Guide):

Possible options include -x, --T2, -F <F threshold*>, -S <F threshold>, -c <t threshold>

* At some point, this could also run without a threshold; however, it usually gives an error ("F missing non-optional argument" or similar).

Older examples

Refer to table in randomise guide for files that are created (randomise User Guide).

[user@localhost temp]$ randomise -i L_Hipp_vertexMNI_2sample.nii.gz -m L_Hipp_vertexMNI_2sample_mask.nii.gz -o L_Hipp_vertexMNI_2sample_rand -d des_2sample.mat -t des_2sample.con -f des_2sample.fts --fonly -D -F 3