# Creating a new combined workflow

A CombinedWorkflow is a series of DIPY workflows organized together in a way that the output of a workflow serves as input for the next one.

First create your CombinedWorkflow class. Your CombinedWorkflow class file is usually located in the dipy/workflows directory.

from dipy.workflows.combined_workflow import CombinedWorkflow


CombinedWorkflow is the base class that will be extended to create our combined workflow.

from dipy.workflows.denoise import NLMeansFlow
from dipy.workflows.segment import MedianOtsuFlow


MedianOtsuFlow and NLMeansFlow will be combined to create our processing section.

class DenoiseAndSegment(CombinedWorkflow):


DenoiseAndSegment is the name of our combined workflow. Note that it needs to extend CombinedWorkflow for everything to work properly.

def _get_sub_flows(self):
return [
NLMeansFlow,
MedianOtsuFlow
]


It is mandatory to implement this method if you want to make all the sub workflows parameters available in commandline.

def run(self, input_files, out_dir='', out_file='processed.nii.gz'):
"""
Parameters
----------
input_files : string
Path to the input files. This path may contain wildcards to
process multiple inputs at once.

out_dir : string, optional
Where the resulting file will be saved. (default '')

out_file : string, optional
Name of the result file to be saved. (default 'processed.nii.gz')
"""


Just like a normal workflow, it is mandatory to have out_dir as a parameter. It is also mandatory to put ‘out_’ in front of every parameter that is going to be an output. Lastly, all out_ params needs to be at the end of the params list. The class docstring part is very important, you need to document every parameter as they will be used with inspection to build the command line argument parser.

io_it = self.get_io_iterator()

for in_file, out_file in io_it:
nl_flow = NLMeansFlow()
self.run_sub_flow(nl_flow, in_file, out_dir=out_dir)
denoised = nl_flow.last_generated_outputs['out_denoised']

me_flow = MedianOtsuFlow()
self.run_sub_flow(me_flow, denoised, out_dir=out_dir)


Use self.get_io_iterator() in every workflow you create. This creates an IOIterator object that create output file names and directory structure based on the inputs and some other advanced output strategy parameters.

Iterating on the IOIterator object you created previously you conveniently get all input and output paths for every input file found when globbin the input parameters.

In the IOIterator loop you can see how we create a new NLMeans workflow then run it using self.run_sub_flow. Running it this way will pass any workflow specific parameter that was retreived from the command line and will append the ones you specify as optional parameters (out_dir in this case).

Lastly, the outputs paths are retrived using workflow.last_generated_outputs. This allows to use denoise as the input for the MedianOtsuFlow.

This is it for the combined workflow class! Now to be able to call it easily via command line, you need this last bit of code. It is usually in an executable file located in bin.

from dipy.workflows.flow_runner import run_flow


This is the method that will wrap everything that is needed to make a workflow ready then run it.

if __name__ == "__main__":
run_flow(DenoiseAndSegment())


This is the only thing needed to make your workflow available through command line.

Now just call the script you just made with -h to see the argparser help text:

python combined_workflow_creation.py --help


You should see all your parameters available along with some extra common ones like logging file and force overwrite. Also all the documentation you wrote about each parameter is there. Also note that every sub workflow optional parameter is available.

Now call it for real with a nifti file to see the results. Experiment with the parameters and see the results:

python combined_workflow_creation.py volume.nii.gz


Example source code

You can download the full source code of this example. This same script is also included in the dipy source distribution under the doc/examples/ directory.