dMRI: Group connectivity - MRtrix, FSL, FreeSurfer


This script,, runs group-based connectivity analysis using the dmri.mrtrix.connectivity_mapping Nipype workflow. Further detail on the processing can be found in dMRI: Connectivity - MRtrix, CMTK, FreeSurfer. This tutorial can be run using:


We perform this analysis using one healthy subject and two subjects who suffer from Parkinson’s disease.

The whole package (960 mb as .tar.gz / 1.3 gb uncompressed) including the Freesurfer directories for these subjects, can be acquired from here:

A data package containing the outputs of this pipeline can be obtained from here:

Along with MRtrix, FSL, and Freesurfer, you must also have the Connectome File Format library installed as well as the Connectome Mapper (cmp).

Or on github at:

Output data can be visualized in ConnectomeViewer, TrackVis, Gephi, the MRtrix Viewer (mrview), and anything that can view Nifti files.

The fiber data is available in Numpy arrays, and the connectivity matrix is also produced as a MATLAB matrix.

Import the workflows

First, we import the necessary modules from nipype.

import nipype.interfaces.fsl as fsl
import nipype.interfaces.freesurfer as fs  # freesurfer
import os.path as op  # system functions
import cmp
from nipype.workflows.dmri.mrtrix.group_connectivity import create_group_connectivity_pipeline
from nipype.workflows.dmri.connectivity.group_connectivity import (

Set the proper directories

First, we import the necessary modules from nipype.

subjects_dir = op.abspath('groupcondatapackage/subjects/')
data_dir = op.abspath('groupcondatapackage/data/')

Define the groups

Here we define the groups for this study. We would like to search for differences between the healthy subject and the two vegetative patients. The group list is defined as a Python dictionary (see, with group IDs (‘controls’, ‘parkinsons’) as keys, and subject/patient names as values. We set the main output directory as ‘groupcon’.

group_list = {}
group_list['controls'] = ['cont17']
group_list['parkinsons'] = ['pat10', 'pat20']

The output directory must be named as well.

global output_dir
output_dir = op.abspath('dmri_group_connectivity_mrtrix')

Main processing loop

The title for the final grouped-network connectome file is dependent on the group names. The resulting file for this example is ‘parkinsons-controls.cff’. The following code implements the format a-b-c-…x.cff for an arbitary number of groups.


The ‘info’ dictionary below is used to define the input files. In this case, the diffusion weighted image contains the string ‘dti’. The same applies to the b-values and b-vector files, and this must be changed to fit your naming scheme.

The workflow is created given the information input about the groups and subjects.

See also

We set values for absolute threshold used on the fractional anisotropy map. This is done in order to identify single-fiber voxels. In brains with more damage, however, it may be necessary to reduce the threshold, since their brains are have lower average fractional anisotropy values.

We invert the b-vectors in the encoding file, and set the maximum harmonic order of the pre-tractography spherical deconvolution step. This is done to show how to set inputs that will affect both groups.

Next we create and run the second-level pipeline. The purpose of this workflow is simple: It is used to merge each subject’s CFF file into one, so that there is a single file containing all of the networks for each group. This can be useful for performing Network Brain Statistics using the NBS plugin in ConnectomeViewer.

title = ''
for idx, group_id in enumerate(group_list.keys()):
    title += group_id
    if not idx == len(list(group_list.keys())) - 1:
        title += '-'

    info = dict(
        dwi=[['subject_id', 'dti']],
        bvecs=[['subject_id', 'bvecs']],
        bvals=[['subject_id', 'bvals']])

    l1pipeline = create_group_connectivity_pipeline(
        group_list, group_id, data_dir, subjects_dir, output_dir, info)

    # Here with invert the b-vectors in the Y direction and set the maximum harmonic order of the
    # spherical deconvolution step
    l1pipeline.inputs.connectivity.mapping.fsl2mrtrix.invert_y = True
    l1pipeline.inputs.connectivity.mapping.csdeconv.maximum_harmonic_order = 6

    # Here we define the parcellation scheme and the number of tracks to produce
    parcellation_name = 'scale500'
    l1pipeline.inputs.connectivity.mapping.Parcellate.parcellation_name = parcellation_name
    cmp_config = cmp.configuration.PipelineConfiguration()
    cmp_config.parcellation_scheme = "Lausanne2008"
    l1pipeline.inputs.connectivity.mapping.inputnode_within.resolution_network_file = cmp_config._get_lausanne_parcellation(
    l1pipeline.inputs.connectivity.mapping.probCSDstreamtrack.desired_number_of_tracks = 100000
    l1pipeline.write_graph(format='eps', graph2use='flat')

    # The second-level pipeline is created here
    l2pipeline = create_merge_network_results_by_group_workflow(
        group_list, group_id, data_dir, subjects_dir, output_dir)
    l2pipeline.inputs.l2inputnode.network_file = cmp_config._get_lausanne_parcellation(
    l2pipeline.write_graph(format='eps', graph2use='flat')

Now that the for loop is complete there are two grouped CFF files each containing the appropriate subjects. It is also convenient to have every subject in a single CFF file, so that is what the third-level pipeline does.

l3pipeline = create_merge_group_network_results_workflow(
    group_list, data_dir, subjects_dir, output_dir, title)
l3pipeline.write_graph(format='eps', graph2use='flat')

The fourth and final workflow averages the networks and saves them in another CFF file

l4pipeline = create_average_networks_by_group_workflow(
    group_list, data_dir, subjects_dir, output_dir, title)
l4pipeline.write_graph(format='eps', graph2use='flat')

Example source code

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