Nipype BRAINS Wrapping Demonstration
TODO: What we really need is a fresh machine to do the install on and all of the steps in order. As soon as I can get such a machine and root access, I will document everything needed to go from our default puppet install state to a working pipeline.
Nipype Installation
<< Install dependancies -- add notes later >>
Download latest source code: git clone git://github.com/chrisfilo/nipype.git
In this demonstration, we will wrap a simple BRAINS pipeline BRAINSFit and BRAINSResample into a pipeline over multiple subjects. We will assume for the sake of this demonstration that you have a machine which currently has a working install of nipype.
Task statement: For all subjects in /hjohnson/HDNI/ixi_data/, register the T1 to the space of the file passed to our script on the command line.
1. Wrap BRAINSFit and BRAINSResample into nipype code.
1a. Download generate_classes.py.
https://github.com/chrisfilo/nipype/blob/slicer/nipype/interfaces/slicer/generate_classes.py
Make sure that you download it properly. If you simply copy-paste, it can mess up spacing, which python relies on for control flow.
This script will build a nipype wrapper class for an XML-exporting command-line tool. If you wish to make it executable, add "#!/usr/bin/python" to the header and:
chmod +x generate_classes.py
1b. Run generate_classes.py over BRAINSFit and BRAINSResample
export PATH=${SLICER3_BIN_DIR}:${PATH}
./generate_classes.py
mkdir ~/mypython
mv BRAINS*.py ~/mypython
export PYTHON_PATH=~/mypython
This will also wrap a few other BRAINS commands, such as BRAINSDemonWarp and BRAINSROIAuto. To control what it wraps, simply adjust the module list near the top of the file; it's quite simple, and in my testing seems to work well.
2. Copy the datafiles used for the next test to your working directory, with the name "slicer_tutorial". You can find a copy in ~peasek/slicer_tutorial.
3. Run the attached file hans_nipype.py
The code has been attached to this wiki instead of inlined to prevent formatting issues, and has been documented to describe what is happening. For a related problem - remapping two sets of subjects to a file within each one's subject directory - check out hans_nipype_predoced.py. It documents a slightly wider range of nipype capabilities.
Be sure to note the current limitations of nipype as described in hans_nipype.py:
1) Datasinks currently behave in an undesirable manner.
2) Nipype is not symlink-aware. If your current path contains a symlink, and you try to reference a path further up the tree past your symlink, nipype will break. It converts absolute paths to relative paths using "../", which does not work when you try to go up from a symlink.