BrainPrint

This is the brainprint python package, a derivative of the original BrainPrint-legacy scripts, with the primary goal to provide a Python-only version, to integrate the LaPy package, and to remove dependencies on third-party software (shapeDNA-* binaries, gmsh, meshfix). As a result, some functionality of the original BrainPrint-legacy scripts is no longer maintained (currently no support of tetrahedral meshes and no support of cortical parcellations or label files).

Installation

Use the following code to install the latest release into your local Python package directory:

python3 -m pip install brainprint

This will also install the necessary dependencies, e.g. the LaPy package. You may need to add your local Python package directory to your $PATH in order to run the scripts.

Usage

Command Line Interface (CLI)

Once installed, the package provides a brainprint executable which can be run from the command line.

The brainprint CLI enables per-subject computation of the individual brainprint descriptors. Its usage and options are summarized below; detailed info is available by calling the script without any arguments from the command line.

brainprint --sdir <directory> --sid <SubjectID>  [--num <num>] [--evec] [--skipcortex] [--norm <surface|volume|geometry|none> ] [--reweight] [--asymmetry] [--outdir <directory>] [--help] [--more-help]

Options:
  --help           Show this help message and exit
  --more-help      Show extensive help message and exit

Required options:
  --sid <SubjectID>
                   Subject ID (FreeSurfer-processed directory inside the
                   subjects directory)
  --sdir <directory>
                   FreeSurfer subjects directory

Processing directives:
  --num <num>      Number of eigenvalues/vectors to compute (default: 50)
  --evec           Switch on eigenvector computation (default: off)
  --skipcortex     Skip cortical surfaces (default: off)
  --norm <surface|volume|geometry|none>
                   Switch on eigenvalue normalization; will be either surface,
                   volume, or determined by the geometry of the object. Use
                   "none" or leave out entirely to skip normalization.
  --reweight       Switch on eigenvalue reweighting (default: off)
  --asymmetry      Perform left-right asymmetry calculation (default: off)
  --cholmod        Switch on use of (faster) Cholesky decomposition instead
                   of (slower) LU decomposition (default: off). May require 
                   manual install of scikit-sparse package. 

Output parameters:
  --outdir=OUTDIR  Output directory (default: <sdir>/<sid>/brainprint)
  --keep-temp      Whether to keep the temporary files directory or not
                   by default False

Python Package

brainprint can also be run within a pure Python environment, i.e. installed and imported as a Python package. E.g.:

>>> from brainprint import Brainprint

>>> subjects_dir = "/path/to/freesurfer/subjects_dir/"
>>> subject_id = "42"

>>> bp = Brainprint(subjects_dir=subjects_dir, asymmetry=True, keep_eigenvectors=True)
>>> results = bp.run(subject_id=subject_id)
>>> results
{"eigenvalues": PosixPath("/path/to/freesurfer/subjects_dir/subject_id/brainprint/subject_id.brainprint.csv"), "eigenvectors": PosixPath("/path/to/freesurfer/subjects_dir/subject_id/brainprint/eigenvectors"), "distances": PosixPath("/path/to/freesurfer/subjects_dir/subject_id/brainprint/subject_id.brainprint.asymmetry.csv")}

Output

The script will create an output directory that contains a CSV table with values (in that order) for the area, volume, and first n eigenvalues per each FreeSurfer structure. An additional output file will be created if the asymmetry calculation is performed and/or for the eigenvectors (CLI --evecs flag or keep_eigenvectors on class initialization).

Changes

Since version 0.5.0, some changes break compatibility with earlier versions (0.4.0 and lower) as well as the original BrainPrint. These changes include:

• for the creation of surfaces from voxel-based segmentations, we have replaced FreeSurfer's marching cube algorithm by scikit-image's marching cube algorithm. Similarly, other FreeSurfer binaries have been replaced by custom Python functions. As a result, a parallel FreeSurfer installation is no longer a requirement for running the brainprint software.
• we have changed / removed the following composite structures from the brainprint shape descriptor: the left and right striatum (composite of caudate, putamen, and nucleus accumbens) and the left and right ventricles (composite of lateral, inferior lateral, 3rd ventricle, choroid plexus, and CSF) have been removed; the left and right cerebellum-white-matter and cerebellum-cortex have been merged into left and right cerebellum.

As a result of these changes, numerical values for the brainprint shape descriptor that are obtained from version 0.5.0 and higher are expected to differ from earlier versions when applied to the same data, but should remain highly correlated with earlier results.

There are some changes in version 0.4.0 (and lower) in functionality in comparison to the original BrainPrint scripts:

• currently no support for tetrahedral meshes
• currently no support for analyses of cortical parcellation or label files
• no more Python 2.x compatibility

API Documentation

The API Documentation can be found at https://deep-mi.org/BrainPrint .

References

If you use this software for a publication please cite:

[1] BrainPrint: a discriminative characterization of brain morphology. Wachinger C, Golland P, Kremen W, Fischl B, Reuter M. Neuroimage. 2015;109:232-48. http://dx.doi.org/10.1016/j.neuroimage.2015.01.032 http://www.ncbi.nlm.nih.gov/pubmed/25613439

[2] Laplace-Beltrami spectra as 'Shape-DNA' of surfaces and solids. Reuter M, Wolter F-E, Peinecke N Computer-Aided Design. 2006;38:342-366. http://dx.doi.org/10.1016/j.cad.2005.10.011