Skip to content

Commit

Permalink
fix: update dwi workflow to new metadata crawling
Browse files Browse the repository at this point in the history
  • Loading branch information
oesteban committed Aug 15, 2024
1 parent 538a282 commit 7025753
Showing 1 changed file with 35 additions and 59 deletions.
94 changes: 35 additions & 59 deletions mriqc/workflows/diffusion/base.py
Original file line number Diff line number Diff line change
Expand Up @@ -43,9 +43,6 @@
This workflow is orchestrated by :py:func:`dmri_qc_workflow`.
"""

from pathlib import Path

import numpy as np
from nipype.interfaces import utility as niu
from nipype.pipeline import engine as pe

Expand Down Expand Up @@ -87,45 +84,29 @@ def dmri_qc_workflow(name='dwiMRIQC'):
from mriqc.messages import BUILDING_WORKFLOW
from mriqc.workflows.shared import synthstrip_wf as dmri_bmsk_workflow

workflow = pe.Workflow(name=name)

dataset = config.workflow.inputs.get('dwi', [])

full_data = []

for dwi_path in dataset:
bval = config.execution.layout.get_bval(dwi_path)
if bval and Path(bval).exists() and len(np.loadtxt(bval)) > config.workflow.min_len_dwi:
full_data.append(dwi_path)
else:
config.loggers.workflow.warn(
f'Dismissing {dwi_path} for processing. b-values are missing or '
'insufficient in number to execute the workflow.'
)

if set(dataset) - set(full_data):
config.workflow.inputs['dwi'] = full_data
config.to_filename()

# Enable if necessary
# mem_gb = config.workflow.biggest_file_gb['dwi']
dataset = config.workflow.inputs['dwi']
metadata = config.workflow.inputs_metadata['dwi']
entities = config.workflow.inputs_entities['dwi']
message = BUILDING_WORKFLOW.format(
modality='diffusion',
detail=(
f'for {len(full_data)} NIfTI files.'
if len(full_data) > 2
else f"({' and '.join('<%s>' % v for v in full_data)})."
),
detail=f'for {len(dataset)} NIfTI files.',
)
config.loggers.workflow.info(message)

if config.execution.datalad_get:
from mriqc.utils.misc import _datalad_get

_datalad_get(full_data)

# Define workflow, inputs and outputs
# 0. Get data, put it in RAS orientation
inputnode = pe.Node(niu.IdentityInterface(fields=['in_file']), name='inputnode')
inputnode.iterables = [('in_file', full_data)]
workflow = pe.Workflow(name=name)
inputnode = pe.Node(niu.IdentityInterface(
fields=['in_file', 'metadata', 'entities'],
), name='inputnode')
inputnode.synchronize = True # Do not test combinations of iterables
inputnode.iterables = [
('in_file', dataset),
('metadata', metadata),
('entities', entities),
]

sanitize = pe.Node(
SanitizeImage(
Expand Down Expand Up @@ -245,7 +226,11 @@ def dmri_qc_workflow(name='dwiMRIQC'):
(inputnode, dwi_report_wf, [
('in_file', 'inputnode.name_source'),
]),
(inputnode, iqms_wf, [('in_file', 'inputnode.in_file')]),
(inputnode, iqms_wf, [
('in_file', 'inputnode.in_file'),
('metadata', 'inputnode.metadata'),
('entities', 'inputnode.entities'),
]),
(inputnode, sanitize, [('in_file', 'in_file')]),
(sanitize, dwi_ref, [('out_file', 'in_file')]),
(sanitize, sp_mask, [('out_file', 'in_file')]),
Expand Down Expand Up @@ -335,7 +320,6 @@ def compute_iqms(name='ComputeIQMs'):
wf = compute_iqms()
"""
from niworkflows.interfaces.bids import ReadSidecarJSON

from mriqc.interfaces import IQMFileSink
from mriqc.interfaces.diffusion import DiffusionQC
Expand All @@ -348,6 +332,8 @@ def compute_iqms(name='ComputeIQMs'):
niu.IdentityInterface(
fields=[
'in_file',
'metadata',
'entities',
'in_shells',
'n_shells',
'b_values_file',
Expand Down Expand Up @@ -377,7 +363,6 @@ def compute_iqms(name='ComputeIQMs'):
niu.IdentityInterface(
fields=[
'out_file',
'meta_sidecar',
'noise_floor',
]
),
Expand All @@ -389,8 +374,6 @@ def compute_iqms(name='ComputeIQMs'):
name='estimate_sigma',
)

meta = pe.Node(ReadSidecarJSON(index_db=config.execution.bids_database_dir), name='metadata')

measures = pe.Node(DiffusionQC(), name='measures')

addprov = pe.Node(
Expand All @@ -413,10 +396,11 @@ def compute_iqms(name='ComputeIQMs'):
# fmt: off
workflow.connect([
(inputnode, datasink, [('in_file', 'in_file'),
('entities', 'entities'),
(('metadata', _filter_metadata), 'metadata'),
('n_shells', 'NumberOfShells'),
('b_values_shells', 'bValuesEstimation'),
(('b_values_file', _bvals_report), 'bValues')]),
(inputnode, meta, [('in_file', 'in_file')]),
(inputnode, measures, [('in_file', 'in_file'),
('b_values_file', 'in_bval_file'),
('b_values_shells', 'in_shells_bval'),
Expand All @@ -439,15 +423,7 @@ def compute_iqms(name='ComputeIQMs'):
('piesno_sigma', 'piesno_sigma')]),
(inputnode, addprov, [('in_file', 'in_file')]),
(addprov, datasink, [('out_prov', 'provenance')]),
(meta, datasink, [('subject', 'subject_id'),
('session', 'session_id'),
('task', 'task_id'),
('acquisition', 'acq_id'),
('reconstruction', 'rec_id'),
('run', 'run_id'),
(('out_dict', _filter_metadata), 'metadata')]),
(datasink, outputnode, [('out_file', 'out_file')]),
(meta, outputnode, [('out_dict', 'meta_sidecar')]),
(measures, datasink, [('out_qc', 'root')]),
(inputnode, estimate_sigma, [('in_noise', 'in_file'),
('brain_mask', 'mask')]),
Expand Down Expand Up @@ -676,23 +652,23 @@ def epi_mni_align(name='SpatialNormalization'):


def _mean(inlist):
import numpy as np
from numpy import mean

return np.mean(inlist)
return mean(inlist)


def _parse_tqual(in_file):
import numpy as np
from numpy import mean

with open(in_file) as fin:
lines = fin.readlines()
return np.mean([float(line.strip()) for line in lines if not line.startswith('++')])
return mean([float(line.strip()) for line in lines if not line.startswith('++')])


def _parse_tout(in_file):
import numpy as np
from numpy import loadtxt

data = np.loadtxt(in_file) # pylint: disable=no-member
data = loadtxt(in_file) # pylint: disable=no-member
return data.mean()


Expand All @@ -701,9 +677,9 @@ def _tolist(value):


def _get_bvals(bmatrix):
import numpy as np
from numpy import squeeze

return np.squeeze(bmatrix[:, -1]).tolist()
return squeeze(bmatrix[:, -1]).tolist()


def _first(inlist):
Expand All @@ -722,11 +698,11 @@ def _all_but_first(inlist):

def _estimate_sigma(in_file, mask):
import nibabel as nb
import numpy as np
from numpy import median

msk = nb.load(mask).get_fdata() > 0.5
return round(
float(np.median(nb.load(in_file).get_fdata()[msk])),
float(median(nb.load(in_file).get_fdata()[msk])),
6,
)

Expand Down

0 comments on commit 7025753

Please sign in to comment.