2. Priors

ANTsPriors creates priors for each of six tissue classes using the individual session aseg images from fMRIPrep according to the following steps:

1. Load the individual session aseg images (fmriprep/sub-*/ses-*/anat/sub-*_ses-*_desc-aseg_dseg.nii.gz) for each subject that is part of the group template and revalue regions such that they correspond to the six tissue classes (Brainstem, Cerebellum, CSF, GMCortical, GMDeep, WMCortical).
2. Add in external CSF to the CSF tissue class by dilating the session mask and calling all voxels that are in that mask but not in the aseg image CSF (i.e., 24).
3. Pad the tissue classification images so that they are in the same space as the padded T1w images.
4. Concatenate the transforms from the session space to the group template space.
5. Warp the tissue classification images from the session space to the group template space using the composite warp from (4).
6. Zero out all voxels with values below .2 in the tissue class images warped to group template space. This is done to prevent voxels that very unlikely to belong to a particular tissue class being given any probability of being so. This is especially important with external CSF voxels, as they are more likely to be partially warped into non-CSF locations (e.g., skull). If they were allowed to stay above 0, and no other tissues had a probability above 0 in these voxels, the voxels would be called CSF by Atropos because the priors are re-normalized so that the probability of being any tissue class is 1 in every voxel within a mask. Considering the mask created is every non-zero voxel for the priors, it is essential not to have low probability voxels on the edges, especially because Atropos partially depends on voxels within a tissue class having similar intensity values.
7. Tissue classification images warped to the group template space are scaled such that every voxel that is non-zero adds up to 1 across the six tissue classes. These are the final tissue classification priors.