Step by Step Guides
Voxel-Based Morphometry (FSLVBM)
Need T1-weighted acquisitions (note marrow/meninges problems with BET)
- Create directory for analysis and copy images there (using a sensible naming convention will make the GLM designs much easier later on)
Make a template_list file, containing the names of all the structural that should be used to make the template later. Note that this must be balanced across any groups you wish to compare later on.
Make the GLM design.mat and design.con files (e.g. using the Glm GUI). Each row in design.mat corresponds to a structural, in the order as ls command prints the structurals in the analysis directory.
In the analysis directory, run stages of FSLVBM with the three supplied scripts:
- This runs BET on all the .nii.gz files. Use the -b option for default BET, or the -N option if the images contain a lot of neck. View outputs and troubleshoot as appropriate using
<someBrowser> struc/slicesdir/index.html &
This makes the template using all the files in template_list. The -n option specifies nonlinear registration (replacing this with -a gives affine registration, though this is not recommended). Check the alignment using movie mode in fslview:
fslview struc/template_4D_GM &
- This registers all images to the template and then performs modulation and smoothing. Check the alignment using movie mode in fslview:
- This will perform three different levels of smoothing, and these files can be examined with, for example:
- This step also creates the raw t-stat maps using the design.* files. Using these and the images themselves, choose an appropriate level of smoothing.
- Finally, use randomise to generate statistics. A TFCE analysis is recommended, and this can be run using, for example:
cd stats randomise -i GM_mod_merg_s3 -m GM_mask -o fslvbm -d ../design.mat -t design.con -T -n 5000
- Use fslview to examine the 1-p values using, for example:
fslview $FSLDIR/data/standard/MNI152_T1_2mm fslvbm_tfce_corrp_tstat1 -l Red-Yellow -b 0.949,1
Note that the choice of 500 vs 5000 samples in randomise is based on p-value accuracy - see table of p-value errors