antsnetct

Cortical thickness with ANTsPyNet

Installation

Install a complete runnable container from Docker Hub.

Non-containerized installation is not recommended, but is available to advanced users. First install the system requirements ANTs, c3d, and trim_neck.sh. These need to be on the system path.

Then install with pip

git clone https://github.com/ftdc-picsl/antsnetct
pip install antsnetct

If you run into errors with the pip install, you may need to pre-install specific versions of ANTsPyNet and / or ANTsPy. Note also that cached ANTsPyNet data in ~/.keras can get out of date and affect results. The docker containers manage the software and data dependencies for you, and are the recommended way to run antsnetct.

Configuration

All users must set the environment variable TEMPLATEFLOW_HOME to a location containing the template to be used. You can use any template as long as it has both a _T1w.nii.gz file and an associated brain mask.

The default is a single thread for all operations. Multi-threading of ITK can be enabled by setting the environment variable ITK_GLOBAL_DEFAULT_NUMBER_OF_THREADS. Set ITK_GLOBAL_DEFAULT_NUMBER_OF_THREADS=0 to use all available cores, or 8 threads, whichever is smaller. Multi-threading of tensorflow operations in ANTsPyNet processes is controlled by the TF_NUM_INTRAOP_THREADS and TF_NUM_INTEROP_THREADS variables.

Configuration for docker

docker run --rm -it -v /path/to/local/templateflow:/opt/templateflow \
  -e TEMPLATEFLOW_HOME=/opt/templateflow \
  -e ITK_GLOBAL_DEFAULT_NUMBER_OF_THREADS=2 \
  antsnetct:latest --help

Configuration for singularity

Because singularity does not allow the container to set its user, additional options are required.

export SINGULARITYENV_ITK_GLOBAL_DEFAULT_NUMBER_OF_THREADS=2
export SINGULARITYENV_TEMPLATEFLOW_HOME=/opt/templateflow
singularity run --cleanenv --no-home --home /home/antspyuser \
  -B /path/to/local/templateflow:/opt/templateflow \
  antsnetct_latest.sif --help

Cross-sectional thickness

Basic pipeline

• Conform to standard orientation.
• Trim neck (optional).
• Brain extraction (done by antspynet, or supply your own brain mask dataset, or have one in the input dataset).
• Segmentation and bias correction. Priors can come from deep_atropos or user-supplied priors. Priors can be used as input to antsAtroposN4.sh, or used directly as the final segmentation.
• Denoise and bias-correct T1w using the segmentation.
• Thickness computation. Same as antsCorticalThickness.sh.
• Warp to template - similar to antsCorticalThickness.sh
• Generate template space derivatives, Jacobian, GMP, thickness, etc.

See antsnetct --help for current usage.

Cross-sectional output

Output is prefixed with the source entities, which uniquely identifies each T1w input. Some files have metadate stored in a JSON sidecar.

File	Description
desc-biascorrbrain_T1w.nii.gz	Bias-corrected and denoised image, masked with the brain mask.
desc-biascorr_T1w.nii.gz	Bias-corrected and denoised image, with the skull on.
desc-brain_mask.nii.gz	Binary brain mask for the T1w.
desc-preproc_T1w.nii.gz	The input T1w after preprocessing but before any antsnetct processing. Currently, the only preprocessing is neck trimming.
from-{template}_to-T1w_mode-image_xfm.h5	Composite transform for warping images from the template space to the space of the T1w.
from-T1w_to-{template}_mode-image_xfm.h5	Composite transform for warping images from the T1w space to the template.
seg-antsnetct_desc-thickness.nii.gz	Cortical thickness in mm.
seg-antsnetct_dseg.nii.gz	Six-class brain segmentation, using BIDS common derived segmentation labels.
seg-antsnetct_label-{label}_probseg.nii.gz	Probability image for each segmentation label.
space-{template}[_res-{template_res}]_desc-biascorrbrain_T1w.nii.gz	Bias-corrected, denoised image in the template space.
space-{template}[_res-{template_res}]_desc-logjacobian.nii.gz	Log jacobian of the nonlinear component of the warp to the template, for analysis of nonlinear volume change.
space-{template}[_res-{template_res}]_desc-thickness.nii.gz	Cortical thickness in the template space.
space-ADNINormalAgingANTs_res-01_label-CGM_probseg.nii.gz	Cortical gray matter probability in the template space, for VBM.

Longitudinal thickness

First run the cross-sectional (cx) pipeline on all time points. To make this faster, you can use the --longitudinal-preproc option, which will skip thickness and group template normalization.

The longitudinal pipeline defaults to running all session T1w images, but user can input a custom list.

The SST is built from the processed T1w images (ie, desc-biascorr) but the longitudinal segmentation is done with the preprocessed T1w images (ie, desc-preproc), which is oriented and optionally neck-trimmed, but not denoised or bias-corrected.

Pipeline overview:

• Build SST from the cx-processed T1w
• Define a common brain mask from the cx sessions (basically the union of the session masks in SST space)
• Segment SST with ANTsPyNet
• Register SST to group template (optional)

For each session image:

• Register to SST
• Warp SST segmentation to each session space
• Run denoising / N4 / Atropos on session T1w, using SST segmentation as priors.
• Compute cortical thickness on session T1w
• Warp derivatives to SST / group template space

See antsnetct --longitudinal --help for current usage.

Longitudinal output

Subject-level output

Under sub-subjeclabel/anat, subject-level output files are prefixed with sub-JP01. The T1w image in this folder is the single-subject template (SST).

File	Description
desc-brain_mask.nii.gz	Brain mask for the SST.
File	Description
desc-brain_T1w.nii.gz	Brain-extracted SST.
from-{template}_to-T1w_mode-image_xfm.h5	Transform from the group template to the SST.
from-T1w_to-{template}_mode-image_xfm.h5	Transform from the SST to the group template.
seg-antsnetct_dseg.nii.gz	Segmentatation of the SST.
seg-antsnetct_label-{label}_probseg.nii.gz	Probability images from the segmentation. These are used as priors for session-level segmentation.
T1w.nii.gz	The SST image.

Session-level output

Session-level output is similar to the cross-sectional output, except that the transforms map to the SST space, not the group template space. To warp images to the group template, combine the warps, for example:

antsApplyTransforms \
    -d 3 \
    -i sub-01_ses-01_desc-biascorrbrain_T1w.nii.gz \
    -r tpl-group_T1w.nii.gz \
    -o sub-01_ses-01_space-group_desc-biascorrbrain_T1w.nii.gz \
    -t sub-01_from-T1w_to-group_template_mode-image_xfm.h5  sub-01_ses-01_from-T1w_to-sst_mode-image_xfm.h5