def create_dfnd_mask(refimg, affine_func_2_anat, affine_anat_2_atlas,
warp_anat_2_atlas, reference):
import os
from p3.utility import get_basename
# get the current node directory
cwd = os.getcwd()
# get filename to output
out_file = os.path.join(cwd,
'{}_atlas.nii.gz'.format(get_basename(refimg)))
mask_file = os.path.join(
cwd, '{}_atlas_dfnd.nii.gz'.format(get_basename(refimg)))
# create dfnd mask
command = 'antsApplyTransforms -f 0.0 -d 3 -o {} -i {} -t {} -t {} -t {} -r {} -v'.format(
out_file, refimg, warp_anat_2_atlas, affine_anat_2_atlas,
affine_func_2_anat, reference)
print(command)
os.system(command)
# convert to binary
os.system('fslmaths {} -bin {}'.format(out_file, mask_file))
return mask_file
def format_reference(func, reference, bids_dir):
import os
import nibabel
from p3.utility import get_basename
from bids.grabbids import BIDSLayout
# save to node folder (go up 2 directories bc of iterfield)
cwd = os.path.dirname(os.path.dirname(os.getcwd()))
# get filename to output
formatted_reference = os.path.join(
cwd, '{}_format4D.nii.gz'.format(get_basename(func)))
# get dim 4 and TR of input image
dim4 = nibabel.load(func).header.get_data_shape()[3] # get the 4th dim
TR = BIDSLayout(bids_dir).get_metadata(func)[
'RepetitionTime'] # get the TR
# make the reference image the same dims as the input
print('Formatting reference image...')
command = 'ImageMath 3 {} ReplicateImage {} {} {} 0'.format(
formatted_reference, reference, dim4, TR)
print(command)
os.system(command)
return (formatted_reference, dim4, TR)
def applytransforms(in_file, reference4D, combined_transforms4D,
warp_func_2_refimg, dfnd_mask):
import os
from p3.utility import get_basename
# save to node folder (go up 2 directories bc of iterfield)
cwd = os.path.dirname(os.path.dirname(os.getcwd()))
# get filename to output
out_file = os.path.join(
cwd, '{}_moco_atlas.nii.gz'.format(get_basename(in_file)))
# set up command to run
command = 'antsApplyTransforms -f 0.0 -d 4 -i {} -r {} -o {} -t {} -t {} -v'.format(
in_file, reference4D, out_file, combined_transforms4D,
warp_func_2_refimg)
print(command)
# apply transforms
os.system(command)
# mask the aligned func with the dfnd mask
print('Applying dfnd mask...')
os.system('fslmaths {0} -mul {1} {0}'.format(out_file, dfnd_mask))
# return moco, atlas-aligned functional image
return out_file
def convertvsm2ANTSwarp(in_file, ped):
"""
Convert the voxel shift map to ants warp
Grabbed this from fmriprep
(https://github.com/poldracklab/fmriprep/blob/master/fmriprep/interfaces/itk.py#L178)
"""
import nibabel as nb
import numpy as np
import os
from p3.utility import get_basename
# save to node folder (go up 2 directories bc of iterfield)
cwd = os.path.dirname(os.path.dirname(os.getcwd()))
# load file
nii = nb.load(in_file)
phaseEncDim = {'x': 0, 'y': 1, 'z': 2}[ped[0]]
if len(ped) == 2:
phaseEncSign = 1.0
else:
phaseEncSign = -1.0
# Fix header
hdr = nii.header.copy()
hdr.set_data_dtype(np.dtype('<f4'))
hdr.set_intent('vector', (), '')
# Get data, convert to mm
data = nii.get_data()
aff = np.diag([1.0, 1.0, -1.0])
if np.linalg.det(aff) < 0 and phaseEncDim != 0:
# Reverse direction since ITK is LPS
aff *= -1.0
aff = aff.dot(nii.affine[:3, :3])
# scale the data correctly
data *= phaseEncSign * nii.header.get_zooms()[phaseEncDim]
# Add missing dimensions
zeros = np.zeros_like(data)
field = [zeros, zeros]
field.insert(phaseEncDim, data)
field = np.stack(field, -1)
# Add empty axis
field = field[:, :, :, np.newaxis, :]
# Write out
out_file = os.path.join(cwd,
'{}_antswarp.nii.gz'.format(get_basename(in_file)))
nb.Nifti1Image(field.astype(np.dtype('<f4')), nii.affine,
hdr).to_filename(out_file)
return out_file
def combinetransforms(func,
reference,
dim4,
TR,
affine_func_2_anat,
affine_anat_2_atlas,
warp_anat_2_atlas,
warp_fmc=None):
import os
from p3.utility import get_basename
# save to node folder (go up 2 directories bc of iterfield)
cwd = os.path.dirname(os.path.dirname(os.getcwd()))
# get filename to output
name = get_basename(func)
combined_transforms = os.path.join(
cwd, '{}_combined_transforms.nii.gz'.format(name))
combined_transforms4D = os.path.join(
cwd, '{}_combined_transforms4D.nii.gz'.format(name))
# set up transforms (check in field map correction files exist)
# we exclude the func_2_refimg transform since it is already 4D
if warp_fmc:
transforms = '-t {} -t {} -t {} -t {}'.format(
warp_anat_2_atlas,
affine_anat_2_atlas,
affine_func_2_anat,
warp_fmc,
)
else:
transforms = '-t {} -t {} -t {}'.format(
warp_anat_2_atlas,
affine_anat_2_atlas,
affine_func_2_anat,
)
# convert the transforms to 4D
print('Combining transforms into one warp displacement field...')
command = 'antsApplyTransforms -f 0.0 -d 3 -o [{},1] {} -r {} -v'.format(
combined_transforms, transforms, reference)
print(command)
os.system(command)
# replicate the combined transform
print('Replicating the combined transform into 4D...')
command = 'ImageMath 3 {} ReplicateDisplacement {} {} {} 0'.format(
combined_transforms4D, combined_transforms, dim4, TR)
print(command)
os.system(command)
# return the 4D combined transform
return combined_transforms4D
def resample_2_epi(T1, epi, aparc_aseg=None):
"""
Resample images to epi resolution
"""
import os
import shutil
from p3.utility import get_basename
# get cwd
cwd = os.getcwd()
# get first epi from list
epi = epi[0]
# get filename of T1
out_file = os.path.join(cwd, '{}_funcres.nii.gz'.format(get_basename(T1)))
# resample the T1
os.system('3dresample -rmode Li -master {} -prefix {} -inset {}'.format(
epi, out_file, T1))
T1_epi = out_file
# ONLY if aparc_aseg defined
if aparc_aseg:
# get filename of aparc_aseg
out_file = os.path.join(
cwd, '{}_funcres.nii.gz'.format(get_basename(aparc_aseg)))
# resample the aparc_aseg
os.system(
'3dresample -rmode NN -master {} -prefix {} -inset {}'.format(
epi, out_file, aparc_aseg))
aparc_aseg_epi = out_file
else: # set to empty
aparc_aseg_epi = ''
# return resampled images
return (T1_epi, aparc_aseg_epi)
def antsMotionCorr(fixed_image, moving_image, transform, writewarp):
import os
from p3.utility import get_basename
# save to node folder (go up 2 directories bc of iterfield)
cwd = os.path.dirname(os.path.dirname(os.getcwd()))
# strip filename extension
name = get_basename(moving_image)
output_basename = os.path.join(cwd, name) # set the output basename
output_mocoparams = os.path.join(cwd, '{}MOCOparams.csv'.format(name))
output_warp = os.path.join(cwd, '{}Warp.nii.gz'.format(name))
output_warpedimg = os.path.join(cwd, '{}_Warped.nii.gz'.format(name))
# check write warp boolean
if writewarp:
writewarp = 1
else:
writewarp = 0
# setup commandline execution
command = 'antsMotionCorr -d 3 -o [{},{}] -m MI[{},{},{},{},{},{}] -t {}[{}] -u 1 -e 1 ' \
'-s {} -f {} -i {} -n 1 -w {} -v'.format(
output_basename,
output_warpedimg,
fixed_image,
moving_image,
1, # metric weight
32, # number of bins
'Regular', # sampling Strategy
0.2, # sampling percentage
transform,
0.1, # gradient step
'1x0', # smoothing sigmas
'2x1', # shrink factors
'20x5', # iterations,
writewarp # set flag for writing warps
)
print(command) # print command before running
# run antsMotionCorr
os.system(command)
# remove the InverseWarp image to save space
os.system('rm {}'.format(os.path.join(cwd, '*InverseWarp.nii.gz')))
# return the outputs
return (output_warp, output_mocoparams, output_warpedimg)
def avganats(anat_list):
import os
from p3.utility import get_basename
# get current path
path = os.getcwd()
# join list into string
filelist = ' '.join(anat_list)
# get filename of first file
outfile = '{}_avg.nii.gz'.format(get_basename(anat_list[0]))
os.system('3dMean -prefix {} {}'.format(outfile, filelist))
# return avg anat
return os.path.join(path, outfile)
def fsl_prepare_fieldmap(phasediff, magnitude, TE):
import os
from p3.utility import get_basename
# save to node folder (go up 2 directories bc of iterfield)
cwd = os.path.dirname(os.path.dirname(os.getcwd()))
# get filename to output
out_file = os.path.join(
cwd, '{}_fieldmap.nii.gz'.format(get_basename(phasediff)))
# run prepare field map
os.system('fsl_prepare_fieldmap SIEMENS {} {} {} {}'.format(
phasediff, magnitude, out_file, TE))
# return the fieldmap file
return out_file
def apply_warp(in_file, reference, transform):
import os
from p3.utility import get_basename
# get cwd
cwd = os.getcwd()
# get filename to output
out_file = os.path.join(cwd,
'{}_atlas.nii.gz'.format(get_basename(in_file)))
# set up command to run
command = 'antsApplyTransforms -f 0.0 -d 3 -i {} -r {} -o {} -t {} -v'.format(
in_file, reference, out_file, transform)
print(command)
# run transform
os.system(command)
# return output
return out_file
def combinetransforms(avgepi, reference, unwarp, realign):
import os
from p3.utility import get_basename
# save to node folder (go up 2 directories bc of iterfield)
cwd = os.path.dirname(os.path.dirname(os.getcwd()))
# get filename to output
combined_transforms = os.path.join(
cwd, '{}_unwarpedtransform.nii.gz'.format(get_basename(avgepi)))
# set up transforms
transforms = '-t {} -t {}'.format(realign, unwarp)
# convert the transforms to 4D
print('Combining transforms into one warp displacement field...')
command = 'antsApplyTransforms -f 0.0 -d 3 -o [{},1] {} -r {} -v'.format(
combined_transforms, transforms, reference)
print(command)
os.system(command)
# return the 4D combined transform
return combined_transforms
def get_prefix(filename):
from p3.utility import get_basename
return '{}_'.format(get_basename(filename))
def calcFD(moco_params, brain_radius, threshold, filtered_threshold, TR,
min_bpm, max_bpm):
import os
import math
from p3.utility import get_basename
import numpy as np
from scipy import signal
def respiration_iirnotch(TR_in_sec, bpm_min=18.582, bpm_max=25.7263):
""" Function for calculating filter parameters for respiration filter
Takes in the TR (optional min/max breaths-per-min, bpm_min, bpm_max).
Returns the parameters for IIR Notch filter.
"""
fs = 1.0 / TR_in_sec # Sampling frequency (Hz)
fn = fs / 2.0 # Nyquist frequency (Hz)
# RR MIN
rr_min = bpm_min / 60.0 # respiration rate minimum in Hz
fa_min = abs(rr_min - math.floor(
(rr_min + fn) / fs) * fs) # Aliased minimum frequency (Hz)
w0_min = fa_min / fn # Normalized minimum frequency
# RR MAX
rr_max = bpm_max / 60.0 # respiration rate maximum in Hz
fa_max = abs(rr_max - math.floor(
(rr_max + fn) / fs) * fs) # Aliased maximum frequency (Hz)
w0_max = fa_max / fn # Normalized maximum frequency
# RR iirnotch filter
w0 = np.mean([w0_min, w0_max]) # Mean normalized frequency
bw = abs(w0_max - w0_min) # Normalized bandwidth
Q = w0 / bw # Quality factor
b, a = signal.iirnotch(w0, Q) # Filter design
return b, a
# save to node folder (go up 2 directories bc of iterfield)
cwd = os.path.dirname(os.path.dirname(os.getcwd()))
# set output filename
out_file = os.path.join(cwd, '{}.FD'.format(get_basename(moco_params)))
tmask_out_file = os.path.join(cwd,
'{}.tmask'.format(get_basename(moco_params)))
filt_moco_file = os.path.join(
cwd, '{}_filtered.1D'.format(get_basename(moco_params)))
filt_out_file = os.path.join(
cwd, '{}_filtered.FD'.format(get_basename(moco_params)))
filt_tmask_out_file = os.path.join(
cwd, '{}_filtered.tmask'.format(get_basename(moco_params)))
# open file
with open(moco_params, 'r') as moco_file:
moco_nums = moco_file.readlines()
# format the moco numbers from strings to float
f_moco_nums = [
tuple(map(float, list(filter(bool,
moco_num.rstrip().split(" ")))))
for moco_num in moco_nums
]
fr_moco_nums = [(f_moco_num[0] * brain_radius * math.pi / 180,
f_moco_num[1] * brain_radius * math.pi / 180,
f_moco_num[2] * brain_radius * math.pi / 180,
f_moco_num[3], f_moco_num[4], f_moco_num[5])
for f_moco_num in f_moco_nums]
# convert to numpy array and filter the motion numbers
np_moco_nums = np.array(f_moco_nums)
TR = float(TR) # convert TR to float
b, a = respiration_iirnotch(TR, min_bpm, max_bpm) # create filter
# filter data (run twice for 4th order)
filt_moco_stage1 = signal.filtfilt(b,
a,
np_moco_nums,
axis=0,
padtype=None)
filt_moco_stage2 = signal.filtfilt(b,
a,
filt_moco_stage1,
axis=0,
padtype=None)
filt_moco = filt_moco_stage2
# Convert rotations to mm
filt_moco[:, 4:7] = filt_moco[:, 4:7] * brain_radius * math.pi / 180
# Calculate FD values for the series
filtered_FD = np.array([
np.concatenate(
(np.array([0]),
np.sum(np.absolute(filt_moco[1:, :] - filt_moco[0:-1, :]),
axis=1)),
axis=0)
]).transpose()
# convert to lists
filt_moco = np.ndarray.tolist(filt_moco)
filtered_FD = [FD_val[0] for FD_val in np.ndarray.tolist(filtered_FD)]
# calculate FD
FD = [0]
for f1, f2 in zip(fr_moco_nums[0:-1], fr_moco_nums[1:]):
FD.append(sum(map(abs, [v1 - v2 for v1, v2 in zip(f1, f2)])))
# write FD to file
with open(out_file, 'w') as FD_file:
for val in FD:
FD_file.write(str(val))
FD_file.write('\n')
# write tmask to file
with open(tmask_out_file, 'w') as tmask_file:
for val in FD:
tmask_file.write(str(int(val < threshold)))
tmask_file.write('\n')
# write filtered motion numbers to file
with open(filt_moco_file, 'w') as fm_file:
for val in filt_moco:
fm_file.write(' '.join([str(v) for v in val]))
fm_file.write('\n')
# write filtered FD to file
with open(filt_out_file, 'w') as filt_FD_file:
for val in filtered_FD:
filt_FD_file.write(str(val))
filt_FD_file.write('\n')
# write filtered tmask to file
with open(filt_tmask_out_file, 'w') as filt_tmask_file:
for val in filtered_FD:
filt_tmask_file.write(str(int(val < filtered_threshold)))
filt_tmask_file.write('\n')
# return the FD file
return out_file, tmask_out_file, filt_moco_file, filt_out_file, filt_tmask_out_file
def gett1name(T1):
from p3.utility import get_basename
return get_basename(T1)
def __init__(self, settings):
# call base constructor
super().__init__(settings)
# define input/output node
self.set_input([
'func', # I pass this in so I can get the TR info from BIDS
'func_stc_despike',
'warp_func_2_refimg',
'warp_fmc',
'refimg',
'affine_func_2_anat',
'warp_func_2_anat',
'affine_anat_2_atlas',
'warp_anat_2_atlas'
])
self.set_output(['func_atlas'])
# define datasink substitutions
atlas_base = get_basename(settings['atlas']) # get atlas basename
self.set_subs([
('_flirt', '_funcres'),
(atlas_base, 'atlas/{}'.format(atlas_base)
) # save resampled atlas to atlas folder
])
self.set_resubs([
('sub-(?P<subject>\w+_)',
'func/sub-\g<subject>'), # place file under anat folder
('func/sub-(?P<subject>\w+)_ses-(?P<session>\w+)_',
'func/ses-\g<session>/sub-\g<subject>_ses-\g<session>_'
) # add session folders if they exist
])
# grab the resolution of the refimg
self.get_resolution = Node(Function(input_names=['reference'],
output_names=['resolution'],
function=get_resolution),
name='get_resolution')
# resample atlas to epi space so we can use it as a reference
self.resample = Node(fsl.FLIRT(no_search=True), name='resample')
self.resample.inputs.in_file = set_atlas_path(settings['atlas'])
self.resample.inputs.reference = set_atlas_path(settings['atlas'])
# format the reference image (which should be the resampled atlas)
self.format_reference = MapNode(Function(
input_names=['func', 'reference', 'bids_dir'],
output_names=['formatted_reference', 'dim4', 'TR'],
function=format_reference),
iterfield=['func'],
name='format_reference')
self.format_reference.inputs.bids_dir = settings['bids_dir']
# combine 3D transforms and replicate to 4D
self.combinetransforms = MapNode(
Function(input_names=[
'func', 'reference', 'dim4', 'TR', 'affine_func_2_anat',
'affine_anat_2_atlas', 'warp_anat_2_atlas', 'warp_fmc'
],
output_names=['combined_transforms4D'],
function=combinetransforms),
iterfield=['func', 'dim4', 'TR', 'warp_fmc'],
name='combinetransforms')
self.combinetransforms.n_procs = settings['num_threads']
# align the reference image to atlas and create a dfnd mask from it
self.create_dfnd_mask = Node(Function(input_names=[
'refimg', 'affine_func_2_anat', 'affine_anat_2_atlas',
'warp_anat_2_atlas', 'reference'
],
output_names=['mask_file'],
function=create_dfnd_mask),
name='create_dfnd_mask')
self.create_dfnd_mask.n_procs = settings['num_threads']
# apply nonlinear transform
self.applytransforms = MapNode(Function(input_names=[
'in_file', 'reference4D', 'combined_transforms4D',
'warp_func_2_refimg', 'dfnd_mask'
],
output_names=['out_file'],
function=applytransforms),
iterfield=[
'in_file', 'reference4D',
'combined_transforms4D',
'warp_func_2_refimg'
],
name='applytransforms')
self.applytransforms.n_procs = settings['num_threads']
