ANTs cortical thickness pipeline container, contains a template for older individuals.
By default, the trim_neck.sh script is called to crop the input image to
remove the neck region, users may optionally mask the neck region with zeros, or
proceed with the unaltered input image.
The template is provided by Nick Tustison, as described in the longitudinal paper (see references below)
Cortical labels are warped to the subject space after processing, label information and definitions are under template/labels.
A warp between the internal template and the MNI152NLin2009cAsym template is downloaded at build time and included in the container. This is used to warp labels from MNI152NLin2009cAsym space to subject space through the internal template. Users may additionally specify their own labels in the MNI152NLin2009cAsym space.
Download Docker images from DockerHub.
To build a Singularity image,
sudo singularity build antsct-aging-tag.sif docker://cookpa/antsct-aging:tag
where "tag" is the version you want to build, or "latest" to get the latest version.
ANTs cortical thickness is part of ANTs.
The trim_neck.sh is provided by Paul Yushkevich and Sandhitsu Das, and uses c3d.
Please see the individual label directories under the template directory for label information including citations and licensing terms.
Custom labels defined at run time must be aligned to the MNI152NLin2009cAsym template in templateflow v1.4.1.
Cortical labels are propagated to the thickness mask after being warped to the subject space. This tries to ensure that cortical GM is not left unlabeled because of registration error.
The original ANTsCT templates label the amygdala and hippocampus as cortical GM. This
means propagated labels must include these areas. Prior to antsct-aging 0.6.0, the
built-in labels did not include these areas, which leads to inaccurate label propagation
in the temporal lobe. Only "Cortical" labels or labels passed with the
--mni-cortical-labels were affected.
Starting from antsct-aging 0.6.0, the template labels have been updated to temporarily include BrainCOLOR amygdala and hippocampus. This should reduce problems, but it's possible that mislabeling may still occur as a result of registration or segmentation error, so QC of labels is still important.
Thickness in medial temporal labels adjacent to the amygdala and hippocampus, notably parahippocampal and entorhinal cortex, may still be less reliable because even though the label propagation algorithm is fixed, the underlying thickness computation is still the same. We recommend using a dedicated segmentation tool for these areas, such as ASHS.
To assist QC, the "WarpedVsPropagated" statistics have been changed in v0.6.0, we now output "WarpedVsMasked" and "MaskedVsPropagated" statistics. This makes it clearer how much the labels change with each processing step.
The label propagation implicitly includes masking since no labels are propagated outside the thickness mask. The "MaskedVsPropagated" stats therefore show how much the labels change by the propagation adding labels to unlabeled GM voxels. The "WarpedVsMasked" show how much the labels change by removing labeled voxels outside the subject's GM segmentation.
# For cross-sectional processing
docker run --rm -it cookpa/antsct-aging:0.6.0 --help
# For longitudinal processing
docker run --rm -it cookpa/antsct-aging:0.6.0 --help --longitudinal
For ANTs cortical thickness: Tustison, et al 2014.
For ANTs longitudinal cortical thickness: Tustison, et al 2019
For c3d: Yushkevich, et al 2006.