import os
import numpy as np
import nibabel as nb
from CPAC.pipeline import nipype_pipeline_engine as pe
import nipype.interfaces.utility as util
from CPAC.utils.interfaces.function import Function
from nipype.interfaces.afni.base import (AFNICommand, AFNICommandInputSpec)
from nipype.interfaces.base import (TraitedSpec, traits, isdefined, File)
[docs]def motion_power_statistics(name='motion_stats',
motion_correct_tool='3dvolreg'):
"""
The main purpose of this workflow is to get various statistical measures
from the movement/motion parameters obtained in functional preprocessing.
Parameters
----------
:param str name: Name of the workflow, defaults to 'motion_stats'
:return: Nuisance workflow.
:rtype: nipype.pipeline.engine.Workflow
Notes
-----
Workflow Inputs::
inputspec.subject_id : string
Subject name or id
inputspec.scan_id : string
Functional Scan id or name
inputspec.motion_correct : string (func/rest file or a list of func/rest nifti file)
Path to motion corrected functional data
inputspec.max_displacement : string (Mat file)
maximum displacement (in mm) vector for brain voxels in each volume.
This file is obtained in functional preprocessing step
inputspec.movement_parameters : string (Mat file)
1D file containing six movement/motion parameters(3 Translation, 3 Rotations)
in different columns (roll pitch yaw dS dL dP), obtained in functional preprocessing step
Workflow Outputs::
outputspec.FDP_1D : 1D file
mean Framewise Displacement (FD)
outputspec.power_params : txt file
Text file containing various power parameters for scrubbing
outputspec.motion_params : txt file
Text file containing various movement parameters
Order of commands:
- Calculate Framewise Displacement FD as per power et al., 2012
Differentiating head realignment parameters across frames yields a six dimensional timeseries that represents instantaneous head motion.
Rotational displacements are converted from degrees to millimeters by calculating displacement on the surface of a sphere of radius 50 mm.[R5]
- Calculate Framewise Displacement FD as per jenkinson et al., 2002
- Calculate DVARS
DVARS (D temporal derivative of timecourses, VARS referring to RMS variance over voxels) indexes
the rate of change of BOLD signal across the entire brain at each frame of data.To calculate
DVARS, the volumetric timeseries is differentiated (by backwards differences) and RMS signal
change is calculated over the whole brain.DVARS is thus a measure of how much the intensity
of a brain image changes in comparison to the previous timepoint (as opposed to the global
signal, which is the average value of a brain image at a timepoint).[R5]
- Calculate Power parameters::
MeanFD : Mean (across time/frames) of the absolute values for Framewise Displacement (FD),
computed as described in Power et al., Neuroimage, 2012)
rootMeanSquareFD : Root mean square (RMS; across time/frames) of the absolute values for FD
rmsFD : Root mean square (RMS; across time/frames) of the absolute values for FD
FDquartile(top 1/4th FD) : Mean of the top 25% highest FD values
MeanDVARS : Mean of voxel DVARS
- Calculate Motion Parameters
Following motion parameters are calculated::
Subject
Scan
Mean Relative RMS Displacement
Max Relative RMS Displacement
Movements > threshold
Mean Relative Mean Rotation
Mean Relative Maxdisp
Max Relative Maxdisp
Max Abs Maxdisp
Max Relative Roll
Max Relative Pitch
Max Relative Yaw
Max Relative dS-I
Max Relative dL-R
Max Relative dP-A
Mean Relative Roll
Mean Relative Pitch
Mean Relative Yaw
Mean Relative dS-I
Mean Relative dL-R
Mean Relative dP-A
Max Abs Roll
Max Abs Pitch
Max Abs Yaw
Max Abs dS-I
Max Abs dL-R
Max Abs dP-A
Mean Abs Roll
Mean Abs Pitch
Mean Abs Yaw
Mean Abs dS-I
Mean Abs dL-R
Mean Abs dP-A
.. exec::
from CPAC.generate_motion_statistics import motion_power_statistics
wf = motion_power_statistics()
wf.write_graph(
graph2use='orig',
dotfilename='./images/generated/motion_statistics.dot'
)
High Level Workflow Graph:
.. image:: ../../images/generated/motion_statistics.png
:width: 1000
Detailed Workflow Graph:
.. image:: ../../images/generated/motion_statistics_detailed.png
:width: 1000
Examples
--------
>>> from CPAC import generate_motion_statistics
>>> wf = generate_motion_statistics.motion_power_statistics("generate_statistics")
>>> wf.inputs.inputspec.movement_parameters = 'CPAC_outupts/sub01/func/movement_parameteres/rest_mc.1D' # doctest: +SKIP
>>> wf.inputs.inputspec.max_displacement = 'CPAC_outputs/sub01/func/max_dispalcement/max_disp.1D' # doctest: +SKIP
>>> wf.inputs.inputspec.motion_correct = 'CPAC_outputs/sub01/func/motion_correct/rest_mc.nii.gz' # doctest: +SKIP
>>> wf.inputs.inputspec.mask = 'CPAC_outputs/sub01/func/func_mask/rest_mask.nii.gz' # doctest: +SKIP
>>> wf.inputs.inputspec.transformations = 'CPAC_outputs/sub01/func/coordinate_transformation/rest_mc.aff12.1D' # doctest: +SKIP
>>> wf.inputs.inputspec.subject_id = 'sub01'
>>> wf.inputs.inputspec.scan_id = 'rest_1'
>>> wf.base_dir = './working_dir' # doctest: +SKIP
>>> wf.run() # doctest: +SKIP
References
----------
.. [1] Power, J. D., Barnes, K. A., Snyder, A. Z., Schlaggar, B. L., & Petersen, S. E. (2012). Spurious
but systematic correlations in functional connectivity MRI networks arise from subject motion. NeuroImage, 59(3),
2142-2154. doi:10.1016/j.neuroimage.2011.10.018
.. [2] Power, J. D., Barnes, K. A., Snyder, A. Z., Schlaggar, B. L., & Petersen, S. E. (2012). Steps
toward optimizing motion artifact removal in functional connectivity MRI; a reply to Carp.
NeuroImage. doi:10.1016/j.neuroimage.2012.03.017
.. [3] Jenkinson, M., Bannister, P., Brady, M., Smith, S., 2002. Improved optimization for the robust
and accurate linear registration and motion correction of brain images. Neuroimage 17, 825-841.
"""
wf = pe.Workflow(name=name)
input_node = pe.Node(util.IdentityInterface(fields=['subject_id',
'scan_id',
'movement_parameters',
'max_displacement',
'rels_displacement',
'motion_correct',
'mask',
'transformations']),
name='inputspec')
output_node = pe.Node(util.IdentityInterface(fields=['FDP_1D',
'FDJ_1D',
'DVARS_1D',
'power_params',
'motion_params']),
name='outputspec')
cal_DVARS = pe.Node(ImageTo1D(method='dvars'),
name='cal_DVARS',
mem_gb=0.4,
mem_x=(739971956005215 / 151115727451828646838272,
'in_file'))
cal_DVARS_strip = pe.Node(Function(input_names=['file_1D'],
output_names=['out_file'],
function=DVARS_strip_t0,
as_module=True),
name='cal_DVARS_strip')
# calculate mean DVARS
wf.connect(input_node, 'motion_correct', cal_DVARS, 'in_file')
wf.connect(input_node, 'mask', cal_DVARS, 'mask')
wf.connect(cal_DVARS, 'out_file', cal_DVARS_strip, 'file_1D')
wf.connect(cal_DVARS_strip, 'out_file', output_node, 'DVARS_1D')
# Calculating mean Framewise Displacement as per power et al., 2012
calculate_FDP = pe.Node(Function(input_names=['in_file'],
output_names=['out_file'],
function=calculate_FD_P,
as_module=True),
name='calculate_FD')
wf.connect(input_node, 'movement_parameters', calculate_FDP, 'in_file')
wf.connect(calculate_FDP, 'out_file', output_node, 'FDP_1D')
# Calculating mean Framewise Displacement as per jenkinson et al., 2002
calculate_FDJ = pe.Node(Function(input_names=['in_file',
'motion_correct_tool'],
output_names=['out_file'],
function=calculate_FD_J,
as_module=True),
name='calculate_FDJ')
calculate_FDJ.inputs.motion_correct_tool = motion_correct_tool
if motion_correct_tool == '3dvolreg':
wf.connect(input_node, 'transformations', calculate_FDJ, 'in_file')
elif motion_correct_tool == 'mcflirt':
wf.connect(input_node, 'rels_displacement', calculate_FDJ, 'in_file')
wf.connect(calculate_FDJ, 'out_file', output_node, 'FDJ_1D')
calc_motion_parameters = pe.Node(Function(input_names=['subject_id',
'scan_id',
'movement_'
'parameters',
'max_displacement',
'motion_correct_'
'tool'],
output_names=['out_file'],
function=gen_motion_parameters,
as_module=True),
name='calc_motion_parameters')
calc_motion_parameters.inputs.motion_correct_tool = motion_correct_tool
wf.connect(input_node, 'subject_id',
calc_motion_parameters, 'subject_id')
wf.connect(input_node, 'scan_id',
calc_motion_parameters, 'scan_id')
wf.connect(input_node, 'movement_parameters',
calc_motion_parameters, 'movement_parameters')
wf.connect(input_node, 'max_displacement',
calc_motion_parameters, 'max_displacement')
wf.connect(calc_motion_parameters, 'out_file',
output_node, 'motion_params')
calc_power_parameters = pe.Node(Function(input_names=['subject_id',
'scan_id',
'fdp',
'fdj',
'dvars',
'motion_correct_tool'],
output_names=['out_file'],
function=gen_power_parameters,
as_module=True),
name='calc_power_parameters')
calc_power_parameters.inputs.motion_correct_tool = motion_correct_tool
wf.connect(input_node, 'subject_id',
calc_power_parameters, 'subject_id')
wf.connect(input_node, 'scan_id',
calc_power_parameters, 'scan_id')
wf.connect(cal_DVARS, 'out_file',
calc_power_parameters, 'dvars')
wf.connect(calculate_FDP, 'out_file',
calc_power_parameters, 'fdp')
if motion_correct_tool == '3dvolreg':
wf.connect(calculate_FDJ, 'out_file', calc_power_parameters, 'fdj')
wf.connect(calc_power_parameters, 'out_file',
output_node, 'power_params')
return wf
[docs]def calculate_FD_P(in_file):
"""
Method to calculate Framewise Displacement (FD) as per Power et al., 2012
Parameters
----------
in_file : string
movement parameters vector file path
Returns
-------
out_file : string
Frame-wise displacement mat
file path
"""
motion_params = np.genfromtxt(in_file).T
rotations = np.transpose(np.abs(np.diff(motion_params[0:3, :])))
translations = np.transpose(np.abs(np.diff(motion_params[3:6, :])))
fd = np.sum(translations, axis=1) + \
(50 * np.pi / 180) * np.sum(rotations, axis=1)
fd = np.insert(fd, 0, 0)
out_file = os.path.join(os.getcwd(), 'FD.1D')
np.savetxt(out_file, fd)
return out_file
[docs]def calculate_FD_J(in_file, motion_correct_tool='3dvolreg', center=None):
"""
Method to calculate framewise displacement as per Jenkinson et al. 2002
Parameters
----------
in_file : string
matrix transformations from volume alignment file path if
motion_correct_tool is '3dvolreg', or FDRMS (*_rel.rms) output if
motion_correct_tool is 'mcflirt'.
motion_correct_tool : string
motion correction tool used, '3dvolreg' or 'mcflirt'.
center : ndarray
optional volume center for the calculation.
Returns
-------
out_file : string
Frame-wise displacement file path
"""
if center is None:
center = np.zeros((3, 1))
else:
center = np.asarray(center).reshape((3, 1))
if motion_correct_tool == '3dvolreg':
pm_ = np.genfromtxt(in_file)
pm = np.zeros((pm_.shape[0], pm_.shape[1] + 4))
pm[:, :12] = pm_
pm[:, 12:] = [0.0, 0.0, 0.0, 1.0]
# The default radius (as in FSL) of a sphere represents the brain
rmax = 80.0
T_rb_prev = pm[0].reshape(4, 4)
fd = np.zeros(pm.shape[0])
for i in range(1, pm.shape[0]):
T_rb = pm[i].reshape(4, 4)
M = np.dot(T_rb, np.linalg.inv(T_rb_prev)) - np.eye(4)
A = M[0:3, 0:3]
b = M[0:3, 3:4] + A @ center
fd[i] = np.sqrt(
(rmax * rmax / 5) * np.trace(np.dot(A.T, A)) + np.dot(b.T, b)
)
T_rb_prev = T_rb
elif motion_correct_tool == 'mcflirt':
rel_rms = np.loadtxt(in_file)
fd = np.append(0, rel_rms)
else:
raise ValueError(f"motion_correct_tool {motion_correct_tool} not supported")
out_file = os.path.join(os.getcwd(), 'FD_J.1D')
np.savetxt(out_file, fd, fmt='%.8f')
return out_file
def find_volume_center(img_file):
"""
Find the center of mass of a Nifti image volume
Parameters
----------
img_file : string (nifti file)
path to nifti volume image
Returns
-------
center : ndarray
volume center of mass vector
"""
img = nb.load(img_file)
dim = np.array(img.header["dim"][1:4])
pixdim = np.array(img.header["pixdim"][1:4])
# Calculation follows MCFLIRT
# https://github.com/fithisux/FSL/blob/7aa2932949129f5c61af912ea677d4dbda843895/src/mcflirt/mcflirt.cc#L479
center = 0.5 * (dim - 1) * pixdim
return center
[docs]def gen_motion_parameters(subject_id, scan_id, movement_parameters,
max_displacement, motion_correct_tool):
"""
Method to calculate all the movement parameters
Parameters
----------
subject_id : string
subject name or id
scan_id : string
scan name or id
max_displacement : string
path of file with maximum displacement (in mm) for brain voxels in each volume
movement_parameters : string
path of 1D file containing six movement/motion parameters(3 Translation,
3 Rotations) in different columns (roll pitch yaw dS dL dP)
Returns
-------
out_file : string
path to csv file containing various motion parameters
"""
mot = np.genfromtxt(movement_parameters).T
# Relative RMS of translation
rms = np.sqrt(mot[3] ** 2 + mot[4] ** 2 + mot[5] ** 2)
# remove any other information other than matrix from
# max displacement file. AFNI adds information to the file
if motion_correct_tool == '3dvolreg':
maxdisp = np.loadtxt(max_displacement)
elif motion_correct_tool == 'mcflirt':
maxdisp = np.loadtxt(max_displacement) # TODO: mcflirt outputs absdisp, instead of maxdisp
abs_relative = lambda v: np.abs(np.diff(v))
max_relative = lambda v: np.max(abs_relative(v))
avg_relative = lambda v: np.mean(abs_relative(v))
max_abs = lambda v: np.max(np.abs(v))
avg_abs = lambda v: np.mean(np.abs(v))
info = [
('Subject', subject_id),
('Scan', scan_id),
('Mean_Relative_RMS_Displacement', avg_relative(rms)),
('Max_Relative_RMS_Displacement', max_relative(rms)),
('Movements_gt_threshold', np.sum(abs_relative(rms) > 0.1)),
('Mean_Relative_Mean_Rotation',
avg_relative(np.abs(mot[0:3]).mean(axis=0))),
('Mean_Relative_Maxdisp', avg_relative(maxdisp)), # to be updated
('Max_Relative_Maxdisp', max_relative(maxdisp)), # to be updated
('Max_Abs_Maxdisp', max_abs(maxdisp)), # to be updated
('Max Relative_Roll', max_relative(mot[0])),
('Max_Relative_Pitch', max_relative(mot[1])),
('Max_Relative_Yaw', max_relative(mot[2])),
('Max_Relative_dS-I', max_relative(mot[3])),
('Max_Relative_dL-R', max_relative(mot[4])),
('Max_Relative_dP-A', max_relative(mot[5])),
('Mean_Relative_Roll', avg_relative(mot[0])),
('Mean_Relative_Pitch', avg_relative(mot[1])),
('Mean_Relative_Yaw', avg_relative(mot[2])),
('Mean_Relative_dS-I', avg_relative(mot[3])),
('Mean_Relative_dL-R', avg_relative(mot[4])),
('Mean_Relative_dP-A', avg_relative(mot[5])),
('Max_Abs_Roll', max_abs(mot[0])),
('Max_Abs_Pitch', max_abs(mot[1])),
('Max_Abs_Yaw', max_abs(mot[2])),
('Max_Abs_dS-I', max_abs(mot[3])),
('Max_Abs_dL-R', max_abs(mot[4])),
('Max_Abs_dP-A', max_abs(mot[5])),
('Mean_Abs_Roll', avg_abs(mot[0])),
('Mean_Abs_Pitch', avg_abs(mot[1])),
('Mean_Abs_Yaw', avg_abs(mot[2])),
('Mean_Abs_dS-I', avg_abs(mot[3])),
('Mean_Abs_dL-R', avg_abs(mot[4])),
('Mean_Abs_dP-A', avg_abs(mot[5])),
]
out_file = os.path.join(os.getcwd(), 'motion_parameters.txt')
with open(out_file, 'w') as f:
f.write(','.join(t for t, v in info))
f.write('\n')
f.write(','.join(
v if type(v) == str else '{0:.6f}'.format(v) for t, v in info))
f.write('\n')
return out_file
[docs]def gen_power_parameters(subject_id, scan_id, fdp=None, fdj=None, dvars=None,
motion_correct_tool='3dvolreg'):
"""
Method to generate Power parameters for scrubbing
Parameters
----------
subject_id : string
subject name or id
scan_id : string
scan name or id
FDP_1D : string
framewise displacement(FD as per power et al., 2012) file path
FDJ_1D : string
framewise displacement(FD as per jenkinson et al., 2002) file path
threshold : float
scrubbing threshold set in the configuration
by default the value is set to 1.0
DVARS : string
path to numpy file containing DVARS
Returns
-------
out_file : string (csv file)
path to csv file containing all the pow parameters
"""
fdp_data = np.loadtxt(fdp)
dvars_data = np.loadtxt(dvars)
# Mean (across time/frames) of the absolute values
# for Framewise Displacement (FD)
meanFD_Power = np.mean(fdp_data)
# Mean DVARS
meanDVARS = np.mean(dvars_data)
if motion_correct_tool == '3dvolreg':
fdj_data = np.loadtxt(fdj)
# Mean FD Jenkinson
meanFD_Jenkinson = np.mean(fdj_data)
# Root mean square (RMS; across time/frames)
# of the absolute values for FD
rmsFDJ = np.sqrt(np.mean(fdj_data))
# Mean of the top quartile of FD is $FDquartile
quat = int(len(fdj_data) / 4)
FDJquartile = np.mean(np.sort(fdj_data)[::-1][:quat])
info = [
('Subject', subject_id),
('Scan', scan_id),
('MeanFD_Power', meanFD_Power),
('MeanFD_Jenkinson', meanFD_Jenkinson),
('rootMeanSquareFD', rmsFDJ),
('FDquartile(top1/4thFD)', FDJquartile),
('MeanDVARS', meanDVARS),
]
elif motion_correct_tool == 'mcflirt':
info = [
('Subject', subject_id),
('Scan', scan_id),
('MeanFD_Power', meanFD_Power),
('MeanDVARS', meanDVARS),
]
out_file = os.path.join(os.getcwd(), 'pow_params.txt')
with open(out_file, 'w') as f:
f.write(','.join(t for t, v in info))
f.write('\n')
f.write(','.join(
v if type(v) == str else '{0:.4f}'.format(v) for t, v in info))
f.write('\n')
return out_file
def DVARS_strip_t0(file_1D):
x = np.loadtxt(file_1D)
x = x[1:]
np.savetxt('dvars_strip.1D', x)
return os.path.abspath('dvars_strip.1D')
class ImageTo1DInputSpec(AFNICommandInputSpec):
in_file = File(desc='input file to 3dTto1D',
argstr='-input %s',
position=1,
mandatory=True,
exists=True,
copyfile=False)
mask = File(desc='-mask dset = use dset as mask to include/exclude voxels',
argstr='-mask %s',
position=2,
exists=True)
out_file = File(name_template="%s_3DtoT1.1D", desc='output 1D file name',
argstr='-prefix %s', name_source="in_file", keep_extension=True)
_methods = [
'enorm', 'dvars',
'rms', 'srms', 's_srms',
'mdiff', 'smdiff',
'4095_count', '4095_frac', '4095_warn',
]
method = traits.Enum(
*_methods,
argstr='-method %s'
)
class ImageTo1DOutputSpec(TraitedSpec):
out_file = File(desc='output 1D file name')
[docs]class ImageTo1D(AFNICommand):
_cmd = '3dTto1D'
input_spec = ImageTo1DInputSpec
output_spec = ImageTo1DOutputSpec
[docs]def calculate_DVARS(func_brain, mask):
"""
Method to calculate DVARS as per power's method
Parameters
----------
func_brain : string (nifti file)
path to motion correct functional data
mask : string (nifti file)
path to brain only mask for functional data
Returns
-------
out_file : string (numpy mat file)
path to file containing array of DVARS calculation for each voxel
"""
rest_data = nb.load(func_brain).get_data().astype(np.float32)
mask_data = nb.load(mask).get_data().astype('bool')
# square of relative intensity value for each voxel across every timepoint
data = np.square(np.diff(rest_data, axis=3))
# applying mask, getting the data in the brain only
data = data[mask_data]
# square root and mean across all timepoints inside mask
dvars = np.sqrt(np.mean(data, axis=0))
out_file = os.path.join(os.getcwd(), 'DVARS.txt')
np.savetxt(out_file, dvars)
return out_file