Carlo Hamalainen

MINC interfaces in Nipype


About two years ago I wrote volgenmodel-nipype, a port of Andrew Janke's volgenmodel to the Nipype workflow system. Nipype provides a framework for wrapping legacy command-line tools in a simple to use interface, which also plugs in to a workflow engine that can run jobs on a multicore PC, IPython parallel, SGE/PBS, etc.

Using a workflow that takes care of naming and tracking input/output files is very convenient. To blur an image (using mincblur) one can create a Node with the Blur interface, and then use .connect to send the output of some other node into this node:

blur = pe.Node(interface=Blur(fwhm=step_x*15),
                              name='blur_' + snum_txt)

workflow.connect(norm, 'output_threshold_mask', blur, 'input_file')

When I first developed volgenmodel-nipype I wrote my own Nipype interfaces for quite a few MINC tools. Over the 2015 Xmas holidays I got those interfaces merged into the master branch of Nipype.

I took this opportunity to tidy up volgenmodel-nipype. There are no locally defined MINC interfaces. I added a public dataset, available in a separate repository: Previously this data wasn't publicly available. I also added some Docker scripts to run the whole workflow and compare the result against a known good model output, which I run in a weekly cronjob as a poor-person's continuous integration test suite.

The mouse brain sample data produces a model that looks like this: