def model_fitting(source_img, prepped_img, subject_info, task):
taskdir = os.path.join(outputdir, task)
if not os.path.exists(taskdir):
os.mkdir(taskdir)
# skull strip the preprocessed BOLD
bet = fsl.BET()
bet.inputs.in_file = prepped_img
bet.inputs.frac = 0.7
bet.inputs.functional = True
bet.inputs.out_file = os.path.join(taskdir, task + "_input_functional_bet.nii.gz")
bet_res = bet.run()
bettedinput = bet_res.outputs.out_file
task_vs_baseline = [task + " vs baseline", 'T', [task], [1]] # set up contrasts
contrasts = [task_vs_baseline]
modelfit = pe.Workflow(name='modelfit', base_dir=taskdir) # generate the model fitting workflow
modelspec = pe.Node(interface=model.SpecifyModel(), name="modelspec") # generate design info
level1design = pe.Node(interface=fsl.Level1Design(), name="level1design") # generate fsf file
modelgen = pe.MapNode( # generate .mat file
interface=fsl.FEATModel(),
name='modelgen',
iterfield=['fsf_file', 'ev_files'])
feat = pe.Node( # feat statistics
interface=fsl.FEAT(),
name='feat',
iterfield=['fsf_file'])
# put it all together
modelfit.connect([
(modelspec, level1design, [('session_info', 'session_info')]),
(level1design, modelgen, [('fsf_files', 'fsf_file'), ('ev_files', 'ev_files')]),
(level1design, feat, [('fsf_files', 'fsf_file')])])
# define inputs to workflow
modelspec.inputs.input_units = 'secs'
modelspec.inputs.functional_runs = bettedinput
modelspec.inputs.time_repetition = source_img.entities['RepetitionTime']
modelspec.inputs.high_pass_filter_cutoff = 90
modelspec.inputs.subject_info = subject_info
level1design.inputs.interscan_interval = source_img.entities['RepetitionTime']
level1design.inputs.bases = {'gamma': {'gammasigma': 3, 'gammadelay': 6, 'derivs': True}}
level1design.inputs.contrasts = contrasts
level1design.inputs.model_serial_correlations = True
# Run the model-fitting pipeline. Main outputs are a feat directory (w/ functional img) and a design.mat file
res = modelfit.run()
# outputs
feat_dir = list(res.nodes)[3].result.outputs.feat_dir
thresh_img = feat_dir + "/thresh_zstat1.nii.gz"
return thresh_img
high_pass_filter_cutoff=100),
name="modelspec")
# first-level design
level1design = Node(fsl.Level1Design(bases={'dgamma':{'derivs': True}},
interscan_interval=TR,
model_serial_correlations=True,
contrasts=contrast_list),
name="level1design")
# creating all the other files necessary to run the model
modelgen = Node(fsl.FEATModel(),
name='modelgen')
# then running through FEAT
feat = Node(fsl.FEAT(),
name="feat")
# creating datasink to collect outputs
datasink = Node(DataSink(base_directory=outDir),
name='datasink')
## Use the following DataSink output substitutions
substitutions = [('_subject_id_', 'sub-'),
('_subsession_id_', '/ses-')
]
datasink.inputs.substitutions = substitutions
###########
#
high_pass_filter_cutoff=100),
name="modelspec")
# first-level design
level1design = Node(fsl.Level1Design(bases={'dgamma': {
'derivs': True
}},
interscan_interval=TR,
model_serial_correlations=True),
name="level1design")
# creating all the other files necessary to run the model
modelgen = Node(fsl.FEATModel(), name='modelgen')
# then running through FEAT
feat = Node(fsl.FEAT(), name="feat")
# creating datasink to collect outputs
datasink = Node(DataSink(base_directory=outDir), name='datasink')
## Use the following DataSink output substitutions
substitutions = [('_subject_id_', '/sub-'), ('_run_id_', '/run-')]
datasink.inputs.substitutions = substitutions
###########
#
# SETTING UP THE WORKFLOW NODES
#
###########
), name='l1_spec')
# l1_model creates a first-level model design
l1_model = pe.Node(fsl.Level1Design(
bases={'dgamma': {'derivs': True}},
model_serial_correlations=True,
interscan_interval = tr,
contrasts=contrasts
# orthogonalization=orthogonality,
), name='l1_model')
# feat_spec generates an fsf model specification file
feat_spec = pe.Node(fsl.FEATModel(), name='feat_spec')
# feat_fit actually runs FEAT
feat_fit = pe.Node(fsl.FEAT(), name='feat_fit', mem_gb=5)
## instead of FEAT
#modelestimate = pe.MapNode(interface=fsl.FILMGLS(smooth_autocorr=True,
# mask_size=5,
# threshold=1000),
# name='modelestimate',
# iterfield = ['design_file',
# 'in_file',
# 'tcon_file'])
feat_select = pe.Node(nio.SelectFiles({
'cope': 'stats/cope*.nii.gz',
'pe': 'stats/pe[0-9][0-9].nii.gz',
'tstat': 'stats/tstat*.nii.gz',
'varcope': 'stats/varcope*.nii.gz',
'zstat': 'stats/zstat*.nii.gz',
preproc.connect(inputnode, 'func', get_totalVoxels, 'in_file') # get the number of volumes in a scan get_nVols = pe.MapNode(interface=Function(input_names=['in_file'], output_names=['output'], function=Get_nVols), name='get_nVols', iterfield=['in_file']) preproc.connect(inputnode, 'func', get_nVols, 'in_file') # customize template melodic.fsf file for data of one participant prepare_design_fsf = pe.MapNode(interface=Function(input_names=['feat_files','initial_highres_files','highres_files','npts','total_voxels'], output_names=['out_file'], function=Prepare_Design_FSF), name='prepare_design_fsf', iterfield=['feat_files','initial_highres_files', 'npts','total_voxels']) preproc.connect(inputnode, 'func', prepare_design_fsf, 'feat_files') preproc.connect(n4biasCorrMbRef, 'output_image', prepare_design_fsf, 'initial_highres_files') preproc.connect(maskT2, 'out_file', prepare_design_fsf, 'highres_files') preproc.connect(get_totalVoxels, 'output', prepare_design_fsf, 'total_voxels') preproc.connect(get_nVols, 'output', prepare_design_fsf, 'npts') # define the feat process for preprocessing the data feat = pe.MapNode(interface=fsl.FEAT(), name='feat', iterfield=['fsf_file']) preproc.connect(prepare_design_fsf, 'out_file', feat, 'fsf_file') # # ----------------- denoise --------- # call the ica-fix feature extraction, mel_ica is the output directory of a melodic run extract_features = pe.MapNode(interface=fix.FeatureExtractor(), name='extract_features', iterfield=['mel_ica']) preproc.connect(feat, 'feat_dir', extract_features, 'mel_ica') # assemble inputs for training the ica-fix classifier
def first_level_wf(in_files, output_dir, fwhm=6.0, name='wf_1st_level'):
workflow = pe.Workflow(name=name)
datasource = pe.Node(niu.Function(function=_dict_ds,
output_names=DATA_ITEMS),
name='datasource')
datasource.inputs.in_dict = in_files
datasource.iterables = ('sub', sorted(in_files.keys()))
# Extract motion parameters from regressors file
runinfo = pe.Node(niu.Function(input_names=[
'in_file', 'events_file', 'regressors_file', 'regressors_names'
],
function=_bids2nipypeinfo,
output_names=['info', 'realign_file']),
name='runinfo')
# Set the column names to be used from the confounds file
runinfo.inputs.regressors_names = ['dvars', 'framewise_displacement'] + \
['a_comp_cor_%02d' % i for i in range(6)] + ['cosine%02d' % i for i in range(4)]
# SUSAN smoothing
susan = create_susan_smooth()
susan.inputs.inputnode.fwhm = fwhm
l1_spec = pe.Node(SpecifyModel(parameter_source='FSL',
input_units='secs',
high_pass_filter_cutoff=100),
name='l1_spec')
# l1_model creates a first-level model design
l1_model = pe.Node(
fsl.Level1Design(
bases={'dgamma': {
'derivs': True
}},
model_serial_correlations=True,
#ENTER YOUR OWN CONTRAST HERE
contrasts=[],
# orthogonalization=orthogonality,
),
name='l1_model')
# feat_spec generates an fsf model specification file
feat_spec = pe.Node(fsl.FEATModel(), name='feat_spec')
# feat_fit actually runs FEAT
feat_fit = pe.Node(fsl.FEAT(), name='feat_fit', mem_gb=12)
feat_select = pe.Node(nio.SelectFiles({
'cope': 'stats/cope1.nii.gz',
'pe': 'stats/pe[0-9][0-9].nii.gz',
'tstat': 'stats/tstat1.nii.gz',
'varcope': 'stats/varcope1.nii.gz',
'zstat': 'stats/zstat1.nii.gz',
}),
name='feat_select')
ds_cope = pe.Node(DerivativesDataSink(base_directory=str(output_dir),
keep_dtype=False,
suffix='cope',
desc='intask'),
name='ds_cope',
run_without_submitting=True)
ds_varcope = pe.Node(DerivativesDataSink(base_directory=str(output_dir),
keep_dtype=False,
suffix='varcope',
desc='intask'),
name='ds_varcope',
run_without_submitting=True)
ds_zstat = pe.Node(DerivativesDataSink(base_directory=str(output_dir),
keep_dtype=False,
suffix='zstat',
desc='intask'),
name='ds_zstat',
run_without_submitting=True)
ds_tstat = pe.Node(DerivativesDataSink(base_directory=str(output_dir),
keep_dtype=False,
suffix='tstat',
desc='intask'),
name='ds_tstat',
run_without_submitting=True)
workflow.connect([
(datasource, susan, [('bold', 'inputnode.in_files'),
('mask', 'inputnode.mask_file')]),
(datasource, runinfo, [('events', 'events_file'),
('regressors', 'regressors_file')]),
(susan, l1_spec, [('outputnode.smoothed_files', 'functional_runs')]),
(datasource, l1_spec, [('tr', 'time_repetition')]),
(datasource, l1_model, [('tr', 'interscan_interval')]),
(datasource, ds_cope, [('bold', 'source_file')]),
(datasource, ds_varcope, [('bold', 'source_file')]),
(datasource, ds_zstat, [('bold', 'source_file')]),
(datasource, ds_tstat, [('bold', 'source_file')]),
(susan, runinfo, [('outputnode.smoothed_files', 'in_file')]),
(runinfo, l1_spec, [('info', 'subject_info'),
('realign_file', 'realignment_parameters')]),
(l1_spec, l1_model, [('session_info', 'session_info')]),
(l1_model, feat_spec, [('fsf_files', 'fsf_file'),
('ev_files', 'ev_files')]),
(l1_model, feat_fit, [('fsf_files', 'fsf_file')]),
(feat_fit, feat_select, [('feat_dir', 'base_directory')]),
(feat_select, ds_cope, [('cope', 'in_file')]),
(feat_select, ds_varcope, [('varcope', 'in_file')]),
(feat_select, ds_zstat, [('zstat', 'in_file')]),
(feat_select, ds_tstat, [('tstat', 'in_file')]),
])
return workflow
def first_level_wf(pipeline, subject_id, task_id, output_dir):
"""
First level workflow
"""
workflow = pe.Workflow(name='_'.join((pipeline, subject_id, task_id)))
inputnode = pe.Node(niu.IdentityInterface(fields=[
'bold_preproc', 'contrasts', 'confounds', 'brainmask', 'events_file'
]),
name='inputnode')
outputnode = pe.Node(
niu.IdentityInterface(fields=['sigma_pre', 'sigma_post', 'out_stats']),
name='outputnode')
conf2movpar = pe.Node(niu.Function(function=_confounds2movpar),
name='conf2movpar')
masker = pe.Node(fsl.ApplyMask(), name='masker')
bim = pe.Node(afni.BlurInMask(fwhm=5.0, outputtype='NIFTI_GZ'),
name='bim',
mem_gb=20)
ev = pe.Node(EventsFilesForTask(task=task_id), name='events')
l1 = pe.Node(SpecifyModel(
input_units='secs',
time_repetition=2,
high_pass_filter_cutoff=100,
parameter_source='FSL',
),
name='l1')
l1model = pe.Node(fsl.Level1Design(interscan_interval=2,
bases={'dgamma': {
'derivs': True
}},
model_serial_correlations=True),
name='l1design')
l1featmodel = pe.Node(fsl.FEATModel(), name='l1model')
l1estimate = pe.Node(fsl.FEAT(), name='l1estimate', mem_gb=40)
pre_smooth_afni = pe.Node(afni.FWHMx(combine=True,
detrend=True,
args='-ShowMeClassicFWHM'),
name='smooth_pre_afni',
mem_gb=20)
post_smooth_afni = pe.Node(afni.FWHMx(combine=True,
detrend=True,
args='-ShowMeClassicFWHM'),
name='smooth_post_afni',
mem_gb=20)
pre_smooth = pe.Node(fsl.SmoothEstimate(), name='smooth_pre', mem_gb=20)
post_smooth = pe.Node(fsl.SmoothEstimate(), name='smooth_post', mem_gb=20)
def _resels(val):
return val**(1 / 3.)
def _fwhm(fwhm):
from numpy import mean
return float(mean(fwhm, dtype=float))
workflow.connect([
(inputnode, masker, [('bold_preproc', 'in_file'),
('brainmask', 'mask_file')]),
(inputnode, ev, [('events_file', 'in_file')]),
(inputnode, l1model, [('contrasts', 'contrasts')]),
(inputnode, conf2movpar, [('confounds', 'in_confounds')]),
(inputnode, bim, [('brainmask', 'mask')]),
(masker, bim, [('out_file', 'in_file')]),
(bim, l1, [('out_file', 'functional_runs')]),
(ev, l1, [('event_files', 'event_files')]),
(conf2movpar, l1, [('out', 'realignment_parameters')]),
(l1, l1model, [('session_info', 'session_info')]),
(ev, l1model, [('orthogonalization', 'orthogonalization')]),
(l1model, l1featmodel, [('fsf_files', 'fsf_file'),
('ev_files', 'ev_files')]),
(l1model, l1estimate, [('fsf_files', 'fsf_file')]),
# Smooth
(inputnode, pre_smooth, [('bold_preproc', 'zstat_file'),
('brainmask', 'mask_file')]),
(bim, post_smooth, [('out_file', 'zstat_file')]),
(inputnode, post_smooth, [('brainmask', 'mask_file')]),
(pre_smooth, outputnode, [(('resels', _resels), 'sigma_pre')]),
(post_smooth, outputnode, [(('resels', _resels), 'sigma_post')]),
# Smooth with AFNI
(inputnode, pre_smooth_afni, [('bold_preproc', 'in_file'),
('brainmask', 'mask')]),
(bim, post_smooth_afni, [('out_file', 'in_file')]),
(inputnode, post_smooth_afni, [('brainmask', 'mask')]),
])
# Writing outputs
csv = pe.Node(AddCSVRow(in_file=str(output_dir / 'smoothness.csv')),
name='addcsv_%s_%s' % (subject_id, pipeline))
csv.inputs.sub_id = subject_id
csv.inputs.pipeline = pipeline
# Datasinks
ds_stats = pe.Node(niu.Function(function=_feat_stats), name='ds_stats')
ds_stats.inputs.subject_id = subject_id
ds_stats.inputs.task_id = task_id
ds_stats.inputs.variant = pipeline
ds_stats.inputs.out_path = output_dir
setattr(ds_stats.interface, '_always_run', True)
workflow.connect([
(outputnode, csv, [('sigma_pre', 'smooth_pre'),
('sigma_post', 'smooth_post')]),
(pre_smooth_afni, csv, [(('fwhm', _fwhm), 'fwhm_pre')]),
(post_smooth_afni, csv, [(('fwhm', _fwhm), 'fwhm_post')]),
(l1estimate, ds_stats, [('feat_dir', 'feat_dir')]),
(ds_stats, outputnode, [('out', 'out_stats')]),
])
return workflow
def feat_o(project_name, file_name):
feat = fsl.FEAT()
feat.inputs.fsf_file = f'/home/fsluser/Documents/{project_name}' \
f'/{file_name}/design.fsf'
feat.run()