pmacsASHSThickness

ASHS thickness pipeline, for computing thickness on ASHS segmentation results. The script is a wrapper for the PMACS LPC, which calls the containerized thickness pipeline by Long Xie.

The ASHS thickness paper has a lot more detail on the process and the outputs.

Input

The input to the wrapper is a directory containing MTL segmentations from ASHS. Currently, only ASHS T1w output is supported. The script will search the input directory for output files in the fastashs format, as implemented in pmacsASHS, or in the original format from calling ASHS ashs_main.sh directly. Because the script searches the input directory for matching files, the input directory must only contain files from one session.

The pmacsASHS format is the same as the ASHS fastashs wrapper, where segmentations have the suffix _MTLSeg_[left|right].nii.gz. Standard ASHS segmentations have the suffix _[left|right]}_lfseg_heur.nii.gz.

The script does not detect what atlas was used in the ASHS segmentation - users must ensure that input was segmented with the ASHS-T1 pipeline.

Dependencies

Dependencies have been installed for FTDC LPC users.

ASHS thickness container

Build a Singularity image from the Docker source, and edit the path in the run script. The wrapper has been tested with v4, built with

singularity build ashsthk-4.0.sif docker://longxie/ashsthk:v4

ASHS thickness multi-atlas template

The template path can be set at run time with -a. The default template is for ASHS-T1 segmentations only. A template for ASHS-T2 segmentations is under development.

Running the script

By default, both hemispheres are processed, but this can be changes with the -e option. BSC users may want to run both hemispheres separately to avoid run time limitations.

Pipeline stages

The container supports staged execution, by default stages 1 through 5 are run, which produces thickness using a "variant template", selected to best fit the individual anatomy. Stages 6-8 use a "unified template", which fits a common coordinate coordinate system to the segmentation mesh, enabling pointwise thickness comparisons.

Computational requirements

Using the default stages, run time is approximately 7 hours per hemisphere on the FTDC compute nodes. Long Xie recommends 8Gb of RAM.

Multi-threading is handled automatically based on the number of processors reserved for the LSB job. Greedy will use multiple threads but the geodesic shooting runs single threaded - this may be changed in later versions. Therefore, it may be most efficient to run with a single core.

Output

For each hemisphere, there are three outputs when running the default stages:

• [hemisphere]_template_[group]_fitted_mesh.vtk - segmentation mesh with pointwise thickness data.

• [hemisphere]_template_[group]_momenta.vtk - geodesic shooting momenta.

• [hemisphere]_thickness.csv - Regional average thickness and fit quality measures.

The group is an integer group identifier, which identifies the selected variant template.

When running the unified template steps, there are additional output files

• [hemisphere]_UT_template_fitted_mesh.vtk
• [hemisphere]_UT_template_momenta.vtk

and statistics for thickness and fit quality in the unified template space are appended to the [hemisphere]_thickness.csv files.

Citation

If using this software for research, please cite

Long Xie, et al, "Multi-template analysis of human perirhinal cortex in brain MRI: Explicitly accounting for anatomical variability", NeuroImage 2017 Jan 1; 144(Pt A):183-202, PMID: 27702610, PMCID: PMC5183532, DOI: 10.1016/j.neuroimage.2016.09.070.