sMRI: Using new ANTS for creating a T1 template (ITK4)

In this tutorial we will use ANTS (new ITK4 version aka “antsRegistration”) based workflow to create a template out of multiple T1 volumes. We will also showcase how to fine tune SGE jobs requirements.

  1. Tell python where to find the appropriate functions.

from __future__ import print_function
from future import standard_library
standard_library.install_aliases()

import os
import nipype.interfaces.utility as util
import nipype.interfaces.ants as ants
import nipype.interfaces.io as io
import nipype.pipeline.engine as pe  # pypeline engine

from niflow.nipype1.workflows.smri.ants import antsRegistrationTemplateBuildSingleIterationWF

  1. Download T1 volumes into home directory

import urllib.request
import urllib.error
import urllib.parse
homeDir = os.getenv("HOME")
requestedPath = os.path.join(homeDir, 'nipypeTestPath')
mydatadir = os.path.realpath(requestedPath)
if not os.path.exists(mydatadir):
    os.makedirs(mydatadir)
print(mydatadir)

MyFileURLs = [
    ('http://slicer.kitware.com/midas3/download?bitstream=13121',
     '01_T1_half.nii.gz'),
    ('http://slicer.kitware.com/midas3/download?bitstream=13122',
     '02_T1_half.nii.gz'),
    ('http://slicer.kitware.com/midas3/download?bitstream=13124',
     '03_T1_half.nii.gz'),
    ('http://slicer.kitware.com/midas3/download?bitstream=13128',
     '01_T1_inv_half.nii.gz'),
    ('http://slicer.kitware.com/midas3/download?bitstream=13123',
     '02_T1_inv_half.nii.gz'),
    ('http://slicer.kitware.com/midas3/download?bitstream=13125',
     '03_T1_inv_half.nii.gz'),
]
for tt in MyFileURLs:
    myURL = tt[0]
    localFilename = os.path.join(mydatadir, tt[1])
    if not os.path.exists(localFilename):
        remotefile = urllib.request.urlopen(myURL)

        localFile = open(localFilename, 'wb')
        localFile.write(remotefile.read())
        localFile.close()
        print("Downloaded file: {0}".format(localFilename))
    else:
        print("File previously downloaded {0}".format(localFilename))

ListOfImagesDictionaries - a list of dictionaries where each dictionary is for one scan session, and the mappings in the dictionary are for all the co-aligned images for that one scan session

ListOfImagesDictionaries = [{
    'T1':
    os.path.join(mydatadir, '01_T1_half.nii.gz'),
    'INV_T1':
    os.path.join(mydatadir, '01_T1_inv_half.nii.gz'),
    'LABEL_MAP':
    os.path.join(mydatadir, '01_T1_inv_half.nii.gz')
}, {
    'T1':
    os.path.join(mydatadir, '02_T1_half.nii.gz'),
    'INV_T1':
    os.path.join(mydatadir, '02_T1_inv_half.nii.gz'),
    'LABEL_MAP':
    os.path.join(mydatadir, '02_T1_inv_half.nii.gz')
}, {
    'T1':
    os.path.join(mydatadir, '03_T1_half.nii.gz'),
    'INV_T1':
    os.path.join(mydatadir, '03_T1_inv_half.nii.gz'),
    'LABEL_MAP':
    os.path.join(mydatadir, '03_T1_inv_half.nii.gz')
}]
input_passive_images = [{
    'INV_T1':
    os.path.join(mydatadir, '01_T1_inv_half.nii.gz')
}, {
    'INV_T1':
    os.path.join(mydatadir, '02_T1_inv_half.nii.gz')
}, {
    'INV_T1':
    os.path.join(mydatadir, '03_T1_inv_half.nii.gz')
}]

registrationImageTypes - A list of the image types to be used actively during the estimation process of registration, any image type not in this list will be passively resampled with the estimated transforms. [‘T1’,’T2’]

registrationImageTypes = ['T1']

interpolationMap - A map of image types to interpolation modes. If an image type is not listed, it will be linearly interpolated. { ‘labelmap’:’NearestNeighbor’, ‘FLAIR’:’WindowedSinc’ }

interpolationMapping = {
    'INV_T1': 'LanczosWindowedSinc',
    'LABEL_MAP': 'NearestNeighbor',
    'T1': 'Linear'
}

  1. Define the workflow and its working directory

tbuilder = pe.Workflow(name="antsRegistrationTemplateBuilder")
tbuilder.base_dir = requestedPath

  1. Define data sources. In real life these would be replace by DataGrabbers

InitialTemplateInputs = [mdict['T1'] for mdict in ListOfImagesDictionaries]

datasource = pe.Node(
    interface=util.IdentityInterface(fields=[
        'InitialTemplateInputs', 'ListOfImagesDictionaries',
        'registrationImageTypes', 'interpolationMapping'
    ]),
    run_without_submitting=True,
    name='InputImages')
datasource.inputs.InitialTemplateInputs = InitialTemplateInputs
datasource.inputs.ListOfImagesDictionaries = ListOfImagesDictionaries
datasource.inputs.registrationImageTypes = registrationImageTypes
datasource.inputs.interpolationMapping = interpolationMapping
datasource.inputs.sort_filelist = True

  1. Template is initialized by a simple average in this simple example, any reference image could be used (i.e. a previously created template)

initAvg = pe.Node(interface=ants.AverageImages(), name='initAvg')
initAvg.inputs.dimension = 3
initAvg.inputs.normalize = True

tbuilder.connect(datasource, "InitialTemplateInputs", initAvg, "images")

  1. Define the first iteration of template building

buildTemplateIteration1 = antsRegistrationTemplateBuildSingleIterationWF(
    'iteration01')

Here we are fine tuning parameters of the SGE job (memory limit, numebr of cores etc.)

BeginANTS = buildTemplateIteration1.get_node("BeginANTS")
BeginANTS.plugin_args = {
    'qsub_args':
    '-S /bin/bash -pe smp1 8-12 -l mem_free=6000M -o /dev/null -e /dev/null queue_name',
    'overwrite':
    True
}

tbuilder.connect(initAvg, 'output_average_image', buildTemplateIteration1,
                 'inputspec.fixed_image')
tbuilder.connect(datasource, 'ListOfImagesDictionaries',
                 buildTemplateIteration1, 'inputspec.ListOfImagesDictionaries')
tbuilder.connect(datasource, 'registrationImageTypes', buildTemplateIteration1,
                 'inputspec.registrationImageTypes')
tbuilder.connect(datasource, 'interpolationMapping', buildTemplateIteration1,
                 'inputspec.interpolationMapping')

  1. Define the second iteration of template building

buildTemplateIteration2 = antsRegistrationTemplateBuildSingleIterationWF(
    'iteration02')
BeginANTS = buildTemplateIteration2.get_node("BeginANTS")
BeginANTS.plugin_args = {
    'qsub_args':
    '-S /bin/bash -pe smp1 8-12 -l mem_free=6000M -o /dev/null -e /dev/null queue_name',
    'overwrite':
    True
}
tbuilder.connect(buildTemplateIteration1, 'outputspec.template',
                 buildTemplateIteration2, 'inputspec.fixed_image')
tbuilder.connect(datasource, 'ListOfImagesDictionaries',
                 buildTemplateIteration2, 'inputspec.ListOfImagesDictionaries')
tbuilder.connect(datasource, 'registrationImageTypes', buildTemplateIteration2,
                 'inputspec.registrationImageTypes')
tbuilder.connect(datasource, 'interpolationMapping', buildTemplateIteration2,
                 'inputspec.interpolationMapping')

  1. Move selected files to a designated results folder

datasink = pe.Node(io.DataSink(), name="datasink")
datasink.inputs.base_directory = os.path.join(requestedPath, "results")

tbuilder.connect(buildTemplateIteration2, 'outputspec.template', datasink,
                 'PrimaryTemplate')
tbuilder.connect(buildTemplateIteration2,
                 'outputspec.passive_deformed_templates', datasink,
                 'PassiveTemplate')
tbuilder.connect(initAvg, 'output_average_image', datasink,
                 'PreRegisterAverage')

  1. Run the workflow

tbuilder.run(plugin="SGE")

Example source code

You can download the full source code of this example. This same script is also included in Nipype1 Examples Niflow under the package/niflow/nipype1/examples directory.