pydra-tasks-ants

Pydra tasks for ANTs

Pydra is a dataflow engine which provides a set of lightweight abstractions for DAG construction, manipulation, and distributed execution.

ANTs is a toolbox for multi-variate image registration, segmentation and statistical analysis.

Table of Contents

• Available Tasks
• Installation
• Development
• License

Available Tasks

• ApplyTransforms
• CreateJacobianDeterminantImage
• N4BiasFieldCorrection
• Registration, registration_syn, registration_syn_quick

Installation

pip install pydra-ants

A separate installation of ANTs is required to use this package.

An official conda package is available through conda-forge:

conda install -c conda-forge ants

Automatic Conversion

Automatically generated tasks can be found in the pydra.tasks.ants.auto sub-package. These interfaces should be treated with caution as they likely do not pass testing. Generated tasks that have been edited and pass testing will be imported into one or more of the pydra.tasks.ants.v* sub-packages (e.g. pydra.tasks.ants.v7_4) corresponding to the version of the ants toolkit they are designed for.

Continuous integration

This template uses GitHub Actions to run tests and deploy packages to PYPI. New packages are built and uploaded when releases are created on GitHub, or new releases of Nipype or the Nipype2Pydra conversion tool are released. Releases triggered by updates to Nipype or Nipype2Pydra are signified by the postN suffix where N = <nipype-version><nipype2pydra-version> with the '.'s stripped, e.g. v0.2.3post185010 corresponds to the v0.2.3 tag of this repository with auto-generated packages from Nipype 1.8.5 using Nipype2Pydra 0.1.0.

Development

Methodology

The development of this package is expected to have two phases

1. Where the corresponding Nipype interfaces are considered to be the ground truth, and the Pydra tasks are generated from them
2. When the Pydra tasks are considered be mature and they are edited by hand

Different tasks will probably mature at different times so there will probably be an intermediate phase between 1 and 2.

Developer installation

First install the requirements for running the auto-conversion script and generate the Pydra task interfaces from their Nipype counterparts

pip install -r nipype-auto-conv/requirements.txt

The run the conversion script to convert Nipype interfaces to Pydra

nipype-auto-conv/generate

Install repo in developer mode from the source directory and install pre-commit to ensure consistent code-style and quality.

pip install -e .[test,dev]
pre-commit install

Auto-conversion phase

The auto-converted Pydra tasks are generated from their corresponding Nipype interface in combination with "conversion hints" contained in YAML specs located in nipype-auto-conv/specs/. The self-documented conversion specs are to be edited by hand in order to assist the auto-converter produce valid pydra tasks. After editing one or more conversion specs the pydra.tasks.ants.auto package should be regenerated by running

nipype-auto-conv/generate

The tests should be run on the auto-generated tasks to see if they are valid

pytest pydra/tasks/ants/auto/tests/test_<the-name-of-the-task-you-edited>.py

If the test passes you should then edit the pydra/tasks/ants/v*/__init__.py file to import the auto-generated task interface to signify that it has been validated and is ready for use, where v* corresponds to the version of ANTs that you have tested it against e.g.

from pydra.tasks.ants.auto import <the-task-you-have-validated>

and copy the test file pydra/tasks/ants/auto/tests/test_<validated-task>.py into pydra/tasks/ants/v*/tests.

File-formats and sample test data

The automatically generated tests will attempt to provided the task instance to be tested with sensible default values based on the type of the field and any constraints it has on it. However, these will often need to be manually overridden after consulting the underlying tool's documentation.

For file-based data, automatically generated file-system objects will be created for selected format types, e.g. Nifti, Dicom. Therefore, it is important to specify the format of the file using the "mime-like" string corresponding to a fileformats class in the inputs > types and outputs > types dicts of the YAML spec.

If the required file-type is not found implemented within fileformats, please see the fileformats docs [https://arcanaframework.github.io/fileformats/developer.html] for instructions on how to define new fileformat types, and see fileformats-medimage-extras for an example on how to implement methods to generate sample data for them. Implementations of new fileformats that are specific to ANTs, and functions to generate sample data for them, should be defined in related-packages/fileformats and related-packages/fileformats-extras, respectively.

License

This project is distributed under the terms of the Apache License, Version 2.0.