A growing number of datasets are available on OpenfMRI. This script demonstrates how to use nipype to analyze a data set.
python fmri_ants_openfmri.py –datasetdir ds107
from nipype import config
config.enable_provenance()
from nipype.external import six
from glob import glob
import os
import nipype.pipeline.engine as pe
import nipype.algorithms.modelgen as model
import nipype.algorithms.rapidart as ra
import nipype.interfaces.fsl as fsl
import nipype.interfaces.ants as ants
from nipype.interfaces.c3 import C3dAffineTool
import nipype.interfaces.io as nio
import nipype.interfaces.utility as niu
from nipype.workflows.fmri.fsl import (create_featreg_preproc,
create_modelfit_workflow,
create_fixed_effects_flow)
from nipype import LooseVersion
version = 0
if fsl.Info.version() and \
LooseVersion(fsl.Info.version()) > LooseVersion('5.0.6'):
version = 507
fsl.FSLCommand.set_default_output_type('NIFTI_GZ')
def create_reg_workflow(name='registration'):
"""Create a FEAT preprocessing workflow together with freesurfer
Parameters
----------
::
name : name of workflow (default: 'registration')
Inputs::
inputspec.source_files : files (filename or list of filenames to register)
inputspec.mean_image : reference image to use
inputspec.anatomical_image : anatomical image to coregister to
inputspec.target_image : registration target
Outputs::
outputspec.func2anat_transform : FLIRT transform
outputspec.anat2target_transform : FLIRT+FNIRT transform
outputspec.transformed_files : transformed files in target space
outputspec.transformed_mean : mean image in target space
Example
-------
register = pe.Workflow(name=name) inputnode = pe.Node(interface=niu.IdentityInterface(fields=[‘source_files’,
‘mean_image’, ‘anatomical_image’, ‘target_image’, ‘target_image_brain’, ‘config_file’]),name=’inputspec’)
‘anat2target_transform’, ‘transformed_files’, ‘transformed_mean’, ]),
name=’outputspec’)
Estimate the tissue classes from the anatomical image. But use spm’s segment as FSL appears to be breaking. “”“
stripper = pe.Node(fsl.BET(), name=’stripper’) register.connect(inputnode, ‘anatomical_image’, stripper, ‘in_file’) fast = pe.Node(fsl.FAST(), name=’fast’) register.connect(stripper, ‘out_file’, fast, ‘in_files’)
Binarize the segmentation
binarize = pe.Node(fsl.ImageMaths(op_string='-nan -thr 0.5 -bin'),
name='binarize')
pickindex = lambda x, i: x[i]
register.connect(fast, ('partial_volume_files', pickindex, 2),
binarize, 'in_file')
Calculate rigid transform from mean image to anatomical image
mean2anat = pe.Node(fsl.FLIRT(), name='mean2anat')
mean2anat.inputs.dof = 6
register.connect(inputnode, 'mean_image', mean2anat, 'in_file')
register.connect(stripper, 'out_file', mean2anat, 'reference')
Now use bbr cost function to improve the transform
mean2anatbbr = pe.Node(fsl.FLIRT(), name='mean2anatbbr')
mean2anatbbr.inputs.dof = 6
mean2anatbbr.inputs.cost = 'bbr'
mean2anatbbr.inputs.schedule = os.path.join(os.getenv('FSLDIR'),
'etc/flirtsch/bbr.sch')
register.connect(inputnode, 'mean_image', mean2anatbbr, 'in_file')
register.connect(binarize, 'out_file', mean2anatbbr, 'wm_seg')
register.connect(inputnode, 'anatomical_image', mean2anatbbr, 'reference')
register.connect(mean2anat, 'out_matrix_file',
mean2anatbbr, 'in_matrix_file')
"""
Convert the BBRegister transformation to ANTS ITK format
"""
convert2itk = pe.Node(C3dAffineTool(),
name='convert2itk')
convert2itk.inputs.fsl2ras = True
convert2itk.inputs.itk_transform = True
register.connect(mean2anatbbr, 'out_matrix_file', convert2itk, 'transform_file')
register.connect(inputnode, 'mean_image',convert2itk, 'source_file')
register.connect(stripper, 'out_file', convert2itk, 'reference_file')
Compute registration between the subject’s structural and MNI template This is currently set to perform a very quick registration. However, the registration can be made significantly more accurate for cortical structures by increasing the number of iterations All parameters are set using the example from: #https://github.com/stnava/ANTs/blob/master/Scripts/newAntsExample.sh
reg = pe.Node(ants.Registration(), name='antsRegister')
reg.inputs.output_transform_prefix = "output_"
reg.inputs.transforms = ['Rigid', 'Affine', 'SyN']
reg.inputs.transform_parameters = [(0.1,), (0.1,), (0.2, 3.0, 0.0)]
#reg.inputs.number_of_iterations = ([[10000, 111110, 11110]] * 2 + [[100, 50, 30]])
reg.inputs.number_of_iterations = [[10000, 11110, 11110]] * 2 + [[100, 30, 20]]
reg.inputs.dimension = 3
reg.inputs.write_composite_transform = True
reg.inputs.collapse_output_transforms = True
reg.inputs.initial_moving_transform_com = True
reg.inputs.metric = ['Mattes'] * 2 + [['Mattes', 'CC']]
reg.inputs.metric_weight = [1] * 2 + [[0.5, 0.5]]
reg.inputs.radius_or_number_of_bins = [32] * 2 + [[32, 4]]
reg.inputs.sampling_strategy = ['Regular'] * 2 + [[None, None]]
reg.inputs.sampling_percentage = [0.3] * 2 + [[None, None]]
reg.inputs.convergence_threshold = [1.e-8] * 2 + [-0.01]
reg.inputs.convergence_window_size = [20] * 2 + [5]
reg.inputs.smoothing_sigmas = [[4, 2, 1]] * 2 + [[1, 0.5, 0]]
reg.inputs.sigma_units = ['vox'] * 3
reg.inputs.shrink_factors = [[3, 2, 1]]*2 + [[4, 2, 1]]
reg.inputs.use_estimate_learning_rate_once = [True] * 3
reg.inputs.use_histogram_matching = [False] * 2 + [True]
reg.inputs.winsorize_lower_quantile = 0.005
reg.inputs.winsorize_upper_quantile = 0.995
reg.inputs.args = '--float'
reg.inputs.output_warped_image = 'output_warped_image.nii.gz'
reg.inputs.num_threads = 4
reg.plugin_args = {'qsub_args': '-l nodes=1:ppn=4'}
register.connect(stripper, 'out_file', reg, 'moving_image')
register.connect(inputnode,'target_image_brain', reg,'fixed_image')
Concatenate the affine and ants transforms into a list
pickfirst = lambda x: x[0]
merge = pe.Node(niu.Merge(2), iterfield=['in2'], name='mergexfm')
register.connect(convert2itk, 'itk_transform', merge, 'in2')
register.connect(reg, ('composite_transform', pickfirst), merge, 'in1')
Transform the mean image. First to anatomical and then to target
warpmean.inputs.input_image_type = 3 warpmean.inputs.interpolation = ‘BSpline’ warpmean.inputs.invert_transform_flags = [False, False] warpmean.inputs.terminal_output = ‘file’
register.connect(inputnode,’target_image_brain’, warpmean,’reference_image’) register.connect(inputnode, ‘mean_image’, warpmean, ‘input_image’) register.connect(merge, ‘out’, warpmean, ‘transforms’)
Transform the remaining images. First to anatomical and then to target
warpall = pe.MapNode(ants.ApplyTransforms(),
iterfield=['input_image'],
name='warpall')
warpall.inputs.input_image_type = 3
warpall.inputs.interpolation = 'BSpline'
warpall.inputs.invert_transform_flags = [False, False]
warpall.inputs.terminal_output = 'file'
register.connect(inputnode,'target_image_brain',warpall,'reference_image')
register.connect(inputnode,'source_files', warpall, 'input_image')
register.connect(merge, 'out', warpall, 'transforms')
Assign all the output files
register.connect(warpmean, 'output_image', outputnode, 'transformed_mean')
register.connect(warpall, 'output_image', outputnode, 'transformed_files')
register.connect(mean2anatbbr, 'out_matrix_file',
outputnode, 'func2anat_transform')
register.connect(reg, 'composite_transform',
outputnode, 'anat2target_transform')
return register
def get_subjectinfo(subject_id, base_dir, task_id, model_id):
"""Get info for a given subject
Parameters
----------
subject_id : string
Subject identifier (e.g., sub001)
base_dir : string
Path to base directory of the dataset
task_id : int
Which task to process
model_id : int
Which model to process
Returns
-------
run_ids : list of ints
Run numbers
conds : list of str
Condition names
TR : float
Repetition time
"""
from glob import glob
import os
import numpy as np
condition_info = []
cond_file = os.path.join(base_dir, 'models', 'model%03d' % model_id,
'condition_key.txt')
with open(cond_file, 'rt') as fp:
for line in fp:
info = line.strip().split()
condition_info.append([info[0], info[1], ' '.join(info[2:])])
if len(condition_info) == 0:
raise ValueError('No condition info found in %s' % cond_file)
taskinfo = np.array(condition_info)
n_tasks = len(np.unique(taskinfo[:, 0]))
conds = []
run_ids = []
if task_id > n_tasks:
raise ValueError('Task id %d does not exist' % task_id)
for idx in range(n_tasks):
taskidx = np.where(taskinfo[:, 0] == 'task%03d' % (idx + 1))
conds.append([condition.replace(' ', '_') for condition
in taskinfo[taskidx[0], 2]])
files = glob(os.path.join(base_dir,
subject_id,
'BOLD',
'task%03d_run*' % (idx + 1)))
run_ids.insert(idx, range(1, len(files) + 1))
TR = np.genfromtxt(os.path.join(base_dir, 'scan_key.txt'))[1]
return run_ids[task_id - 1], conds[task_id - 1], TR
def analyze_openfmri_dataset(data_dir, subject=None, model_id=None,
task_id=None, output_dir=None, subj_prefix='*',
hpcutoff=120., use_derivatives=True,
fwhm=6.0):
"""Analyzes an open fmri dataset
Parameters
----------
data_dir : str
Path to the base data directory
work_dir : str
Nipype working directory (defaults to cwd)
"""
Load nipype workflows
preproc = create_featreg_preproc(whichvol='first')
modelfit = create_modelfit_workflow()
fixed_fx = create_fixed_effects_flow()
registration = create_reg_workflow()
Remove the plotting connection so that plot iterables don’t propagate to the model stage
preproc.disconnect(preproc.get_node('plot_motion'), 'out_file',
preproc.get_node('outputspec'), 'motion_plots')
Set up openfmri data specific components
subjects = sorted([path.split(os.path.sep)[-1] for path in
glob(os.path.join(data_dir, subj_prefix))])
infosource = pe.Node(niu.IdentityInterface(fields=['subject_id',
'model_id',
'task_id']),
name='infosource')
if len(subject) == 0:
infosource.iterables = [('subject_id', subjects),
('model_id', [model_id]),
('task_id', task_id)]
else:
infosource.iterables = [('subject_id',
[subjects[subjects.index(subj)] for subj in subject]),
('model_id', [model_id]),
('task_id', task_id)]
subjinfo = pe.Node(niu.Function(input_names=['subject_id', 'base_dir',
'task_id', 'model_id'],
output_names=['run_id', 'conds', 'TR'],
function=get_subjectinfo),
name='subjectinfo')
subjinfo.inputs.base_dir = data_dir
Return data components as anat, bold and behav
datasource = pe.Node(nio.DataGrabber(infields=['subject_id', 'run_id',
'task_id', 'model_id'],
outfields=['anat', 'bold', 'behav',
'contrasts']),
name='datasource')
datasource.inputs.base_directory = data_dir
datasource.inputs.template = '*'
datasource.inputs.field_template = {'anat': '%s/anatomy/highres001.nii.gz',
'bold': '%s/BOLD/task%03d_r*/bold.nii.gz',
'behav': ('%s/model/model%03d/onsets/task%03d_'
'run%03d/cond*.txt'),
'contrasts': ('models/model%03d/'
'task_contrasts.txt')}
datasource.inputs.template_args = {'anat': [['subject_id']],
'bold': [['subject_id', 'task_id']],
'behav': [['subject_id', 'model_id',
'task_id', 'run_id']],
'contrasts': [['model_id']]}
datasource.inputs.sort_filelist = True
Create meta workflow
wf = pe.Workflow(name='openfmri')
wf.connect(infosource, 'subject_id', subjinfo, 'subject_id')
wf.connect(infosource, 'model_id', subjinfo, 'model_id')
wf.connect(infosource, 'task_id', subjinfo, 'task_id')
wf.connect(infosource, 'subject_id', datasource, 'subject_id')
wf.connect(infosource, 'model_id', datasource, 'model_id')
wf.connect(infosource, 'task_id', datasource, 'task_id')
wf.connect(subjinfo, 'run_id', datasource, 'run_id')
wf.connect([(datasource, preproc, [('bold', 'inputspec.func')]),
])
def get_highpass(TR, hpcutoff):
return hpcutoff / (2 * TR)
gethighpass = pe.Node(niu.Function(input_names=['TR', 'hpcutoff'],
output_names=['highpass'],
function=get_highpass),
name='gethighpass')
wf.connect(subjinfo, 'TR', gethighpass, 'TR')
wf.connect(gethighpass, 'highpass', preproc, 'inputspec.highpass')
Setup a basic set of contrasts, a t-test per condition
def get_contrasts(contrast_file, task_id, conds):
import numpy as np
contrast_def = np.genfromtxt(contrast_file, dtype=object)
if len(contrast_def.shape) == 1:
contrast_def = contrast_def[None, :]
contrasts = []
for row in contrast_def:
if row[0] != 'task%03d' % task_id:
continue
con = [row[1], 'T', ['cond%03d' % (i + 1) for i in range(len(conds))],
row[2:].astype(float).tolist()]
contrasts.append(con)
# add auto contrasts for each column
for i, cond in enumerate(conds):
con = [cond, 'T', ['cond%03d' % (i + 1)], [1]]
contrasts.append(con)
return contrasts
contrastgen = pe.Node(niu.Function(input_names=['contrast_file',
'task_id', 'conds'],
output_names=['contrasts'],
function=get_contrasts),
name='contrastgen')
art = pe.MapNode(interface=ra.ArtifactDetect(use_differences=[True, False],
use_norm=True,
norm_threshold=1,
zintensity_threshold=3,
parameter_source='FSL',
mask_type='file'),
iterfield=['realigned_files', 'realignment_parameters',
'mask_file'],
name="art")
modelspec = pe.Node(interface=model.SpecifyModel(),
name="modelspec")
modelspec.inputs.input_units = 'secs'
def check_behav_list(behav):
out_behav = []
if isinstance(behav, six.string_types):
behav = [behav]
for val in behav:
if not isinstance(val, list):
out_behav.append([val])
else:
out_behav.append(val)
return out_behav
wf.connect(subjinfo, 'TR', modelspec, 'time_repetition')
wf.connect(datasource, ('behav', check_behav_list), modelspec, 'event_files')
wf.connect(subjinfo, 'TR', modelfit, 'inputspec.interscan_interval')
wf.connect(subjinfo, 'conds', contrastgen, 'conds')
wf.connect(datasource, 'contrasts', contrastgen, 'contrast_file')
wf.connect(infosource, 'task_id', contrastgen, 'task_id')
wf.connect(contrastgen, 'contrasts', modelfit, 'inputspec.contrasts')
wf.connect([(preproc, art, [('outputspec.motion_parameters',
'realignment_parameters'),
('outputspec.realigned_files',
'realigned_files'),
('outputspec.mask', 'mask_file')]),
(preproc, modelspec, [('outputspec.highpassed_files',
'functional_runs'),
('outputspec.motion_parameters',
'realignment_parameters')]),
(art, modelspec, [('outlier_files', 'outlier_files')]),
(modelspec, modelfit, [('session_info',
'inputspec.session_info')]),
(preproc, modelfit, [('outputspec.highpassed_files',
'inputspec.functional_data')])
])
Reorder the copes so that now it combines across runs
def sort_copes(files):
numelements = len(files[0])
outfiles = []
for i in range(numelements):
outfiles.insert(i, [])
for j, elements in enumerate(files):
outfiles[i].append(elements[i])
return outfiles
def num_copes(files):
return len(files)
pickfirst = lambda x: x[0]
wf.connect([(preproc, fixed_fx, [(('outputspec.mask', pickfirst),
'flameo.mask_file')]),
(modelfit, fixed_fx, [(('outputspec.copes', sort_copes),
'inputspec.copes'),
('outputspec.dof_file',
'inputspec.dof_files'),
(('outputspec.varcopes',
sort_copes),
'inputspec.varcopes'),
(('outputspec.copes', num_copes),
'l2model.num_copes'),
])
])
wf.connect(preproc, 'outputspec.mean', registration, 'inputspec.mean_image')
wf.connect(datasource, 'anat', registration, 'inputspec.anatomical_image')
registration.inputs.inputspec.target_image = fsl.Info.standard_image('MNI152_T1_2mm.nii.gz')
registration.inputs.inputspec.target_image_brain = fsl.Info.standard_image('MNI152_T1_2mm_brain.nii.gz')
registration.inputs.inputspec.config_file = 'T1_2_MNI152_2mm'
def merge_files(copes, varcopes, zstats):
out_files = []
splits = []
out_files.extend(copes)
splits.append(len(copes))
out_files.extend(varcopes)
splits.append(len(varcopes))
out_files.extend(zstats)
splits.append(len(zstats))
return out_files, splits
mergefunc = pe.Node(niu.Function(input_names=['copes', 'varcopes',
'zstats'],
output_names=['out_files', 'splits'],
function=merge_files),
name='merge_files')
wf.connect([(fixed_fx.get_node('outputspec'), mergefunc,
[('copes', 'copes'),
('varcopes', 'varcopes'),
('zstats', 'zstats'),
])])
wf.connect(mergefunc, 'out_files', registration, 'inputspec.source_files')
def split_files(in_files, splits):
copes = in_files[:splits[0]]
varcopes = in_files[splits[0]:(splits[0] + splits[1])]
zstats = in_files[(splits[0] + splits[1]):]
return copes, varcopes, zstats
splitfunc = pe.Node(niu.Function(input_names=['in_files', 'splits'],
output_names=['copes', 'varcopes',
'zstats'],
function=split_files),
name='split_files')
wf.connect(mergefunc, 'splits', splitfunc, 'splits')
wf.connect(registration, 'outputspec.transformed_files',
splitfunc, 'in_files')
Connect to a datasink
def get_subs(subject_id, conds, model_id, task_id):
subs = [('_subject_id_%s_' % subject_id, '')]
subs.append(('_model_id_%d' % model_id, 'model%03d' %model_id))
subs.append(('task_id_%d/' % task_id, '/task%03d_' % task_id))
subs.append(('bold_dtype_mcf_mask_smooth_mask_gms_tempfilt_mean_warp',
'mean'))
subs.append(('bold_dtype_mcf_mask_smooth_mask_gms_tempfilt_mean_flirt',
'affine'))
for i in range(len(conds)):
subs.append(('_flameo%d/cope1.' % i, 'cope%02d.' % (i + 1)))
subs.append(('_flameo%d/varcope1.' % i, 'varcope%02d.' % (i + 1)))
subs.append(('_flameo%d/zstat1.' % i, 'zstat%02d.' % (i + 1)))
subs.append(('_flameo%d/tstat1.' % i, 'tstat%02d.' % (i + 1)))
subs.append(('_flameo%d/res4d.' % i, 'res4d%02d.' % (i + 1)))
subs.append(('_warpall%d/cope1_warp.' % i,
'cope%02d.' % (i + 1)))
subs.append(('_warpall%d/varcope1_warp.' % (len(conds) + i),
'varcope%02d.' % (i + 1)))
subs.append(('_warpall%d/zstat1_warp.' % (2 * len(conds) + i),
'zstat%02d.' % (i + 1)))
subs.append(('_warpall%d/cope1_trans.' % i,
'cope%02d.' % (i + 1)))
subs.append(('_warpall%d/varcope1_trans.' % (len(conds) + i),
'varcope%02d.' % (i + 1)))
subs.append(('_warpall%d/zstat1_trans.' % (2 * len(conds) + i),
'zstat%02d.' % (i + 1)))
return subs
subsgen = pe.Node(niu.Function(input_names=['subject_id', 'conds',
'model_id', 'task_id'],
output_names=['substitutions'],
function=get_subs),
name='subsgen')
datasink = pe.Node(interface=nio.DataSink(),
name="datasink")
wf.connect(infosource, 'subject_id', datasink, 'container')
wf.connect(infosource, 'subject_id', subsgen, 'subject_id')
wf.connect(infosource, 'model_id', subsgen, 'model_id')
wf.connect(infosource, 'task_id', subsgen, 'task_id')
wf.connect(contrastgen, 'contrasts', subsgen, 'conds')
wf.connect(subsgen, 'substitutions', datasink, 'substitutions')
wf.connect([(fixed_fx.get_node('outputspec'), datasink,
[('res4d', 'res4d'),
('copes', 'copes'),
('varcopes', 'varcopes'),
('zstats', 'zstats'),
('tstats', 'tstats')])
])
wf.connect([(splitfunc, datasink,
[('copes', 'copes.mni'),
('varcopes', 'varcopes.mni'),
('zstats', 'zstats.mni'),
])])
wf.connect(registration, 'outputspec.transformed_mean', datasink, 'mean.mni')
wf.connect(registration, 'outputspec.func2anat_transform', datasink, 'xfm.mean2anat')
wf.connect(registration, 'outputspec.anat2target_transform', datasink, 'xfm.anat2target')
Set processing parameters
preproc.inputs.inputspec.fwhm = fwhm gethighpass.inputs.hpcutoff = hpcutoff modelspec.inputs.high_pass_filter_cutoff = hpcutoff modelfit.inputs.inputspec.bases = {‘dgamma’: {‘derivs’: use_derivatives}} modelfit.inputs.inputspec.model_serial_correlations = True if version < 507:
modelfit.inputs.inputspec.film_threshold = 1000
- else:
- modelfit.inputs.inputspec.film_threshold = -1000
datasink.inputs.base_directory = output_dir return wf
import argparse defstr = ‘ (default %(default)s)’ parser = argparse.ArgumentParser(prog=’fmri_openfmri.py’,
description=__doc__)
parser.add_argument(‘-d’, ‘–datasetdir’, required=True) parser.add_argument(‘-s’, ‘–subject’, default=[],
nargs=’+’, type=str, help=”Subject name (e.g. ‘sub001’)”)
args = parser.parse_args() outdir = args.outdir work_dir = os.getcwd() if args.work_dir:
work_dir = os.path.abspath(args.work_dir)
derivatives = args.derivatives if derivatives is None:
derivatives = False
wf.base_dir = work_dir if args.plugin_args:
wf.run(args.plugin, plugin_args=eval(args.plugin_args))
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.