def _run(self):
dimension = self._fixed.sitk.GetDimension()
if self._use_fixed_mask:
fixed_itk_mask = self._fixed.itk_mask
else:
fixed_itk_mask = None
if self._use_moving_mask:
moving_itk_mask = self._moving.itk_mask
else:
moving_itk_mask = None
# Blur moving image with oriented Gaussian prior to the registration
if self._use_oriented_psf:
image_type = itk.Image[self._pixel_type, dimension]
# Get oriented Gaussian covariance matrix
cov_HR_coord = psf.PSF(
).get_covariance_matrix_in_reconstruction_space(
self._fixed, self._moving)
itk_gaussian_interpolator = itk.OrientedGaussianInterpolateImageFunction[
image_type, self._pixel_type].New()
itk_gaussian_interpolator.SetCovariance(cov_HR_coord.flatten())
itk_gaussian_interpolator.SetAlpha(self._alpha_cut)
else:
itk_gaussian_interpolator = None
self._registration_method = \
simplereg.wrap_itk_registration.WrapItkRegistration(
dimension=dimension,
fixed_itk=self._fixed.itk,
moving_itk=self._moving.itk,
fixed_itk_mask=fixed_itk_mask,
moving_itk_mask=moving_itk_mask,
registration_type=self._registration_type,
interpolator=self._interpolator,
metric=self._metric,
# metric_params=self._metric_params,
optimizer=self._optimizer,
optimizer_params=self._optimizer_params,
initializer_type=self._initializer_type,
use_multiresolution_framework=self._use_multiresolution_framework,
# optimizer_scales=self._scales_estimator,
shrink_factors=self._shrink_factors,
smoothing_sigmas=self._smoothing_sigmas,
verbose=self._use_verbose,
itk_oriented_gaussian_interpolate_image_filter=itk_gaussian_interpolator,
)
self._registration_method.run()
self._registration_transform_sitk = \
self._registration_method.get_registration_transform_sitk()
def __init__(self,
deconvolution_mode="full_3D",
predefined_covariance=None,
alpha_cut=3,
image_type=itk.Image.D3,
default_pixel_type=0.0,
):
self._deconvolution_mode = deconvolution_mode
# In case only diagonal entries are given, create diagonal matrix
if predefined_covariance is not None:
if predefined_covariance.size is 3:
self._predefined_covariance = np.diag(
np.array(predefined_covariance))
else:
self._predefined_covariance = np.array(predefined_covariance)
self._psf = psf.PSF()
# Allocate and initialize Oriented Gaussian Interpolate Image Filter
self._filter_oriented_gaussian = \
itk.OrientedGaussianInterpolateImageFilter[
image_type, image_type].New()
self._filter_oriented_gaussian.SetDefaultPixelValue(default_pixel_type)
self._filter_oriented_gaussian.SetAlpha(alpha_cut)
# Allocate and initialize Adjoint Oriented Gaussian Interpolate Image
# Filter
self._filter_adjoint_oriented_gaussian = \
itk.AdjointOrientedGaussianInterpolateImageFilter[
image_type, image_type].New()
self._filter_adjoint_oriented_gaussian.SetDefaultPixelValue(
default_pixel_type)
self._filter_adjoint_oriented_gaussian.SetAlpha(alpha_cut)
# Allocate and initialize masking image filter
self._masking = itk.MultiplyImageFilter[
image_type, image_type, image_type].New()
self._get_covariance = {
"full_3D": self._get_covariance_full_3d,
"only_in_plane": self._get_covariance_only_in_plane,
"predefined_covariance": self._get_covariance_predefined,
}
def _get_interpolator(self, stack_slice):
if self._interpolator == "OrientedGaussian":
# Get oriented PSF covariance matrix
cov = psf.PSF().get_covariance_matrix_in_reconstruction_space(
stack_slice, self._reference)
# Specify oriented Gaussian interpolator
interpolator_itk = itk.OrientedGaussianInterpolateImageFunction[
self._image_type, self._pixel_type].New()
interpolator_itk.SetCovariance(cov.flatten())
interpolator_itk.SetAlpha(self._alpha_cut)
else:
interpolator_itk = eval(
"itk.%sInterpolateImageFunction"
"[self._image_type, self._pixel_type].New()" %
self._interpolator)
return interpolator_itk
def _run_registration_inplane_similarity_3D(self, id, endl=" \\\n"):
if self._fixed is None or self._moving is None:
raise ValueError("Error: Fixed and moving image not specified")
if self._use_verbose:
verbose = "1"
else:
verbose = "0"
# Clean output directory first
os.system("rm -rf " + self._dir_tmp + "*")
moving_str = "RegistrationITK_moving_" + id + self._moving.get_filename(
)
fixed_str = "RegistrationITK_fixed_" + id + self._fixed.get_filename()
moving_mask_str = "RegistrationITK_moving_mask_" + id + self._moving.get_filename(
)
fixed_mask_str = "RegistrationITK_fixed_mask_" + id + self._fixed.get_filename(
)
registration_transform_str = "RegistrationITK_transform_" + id + \
self._fixed.get_filename() + "_" + self._moving.get_filename()
# Write images to HDD
# if not os.path.isfile(self._dir_tmp + moving_str + ".nii.gz"):
sitkh.write_nifti_image_sitk(self._moving.sitk,
self._dir_tmp + moving_str + ".nii.gz")
# if not os.path.isfile(self._dir_tmp + fixed_str + ".nii.gz"):
sitkh.write_nifti_image_sitk(self._fixed.sitk,
self._dir_tmp + fixed_str + ".nii.gz")
# if not os.path.isfile(self._dir_tmp + moving_mask_str + ".nii.gz")
# and self._use_moving_mask:
sitkh.write_nifti_image_sitk(
self._moving.sitk_mask,
self._dir_tmp + moving_mask_str + ".nii.gz")
# if not os.path.isfile(self._dir_tmp + fixed_mask_str + ".nii.gz") and
# self._use_fixed_mask:
sitkh.write_nifti_image_sitk(
self._fixed.sitk_mask, self._dir_tmp + fixed_mask_str + ".nii.gz")
# Prepare command for execution
cmd = DIR_CPP_BUILD + "/bin/itkInplaneSimilarity3DReg" + endl
cmd += "--f " + self._dir_tmp + fixed_str + ".nii.gz" + endl
cmd += "--m " + self._dir_tmp + moving_str + ".nii.gz" + endl
if self._use_fixed_mask:
cmd += "--fmask " + self._dir_tmp + fixed_mask_str + ".nii.gz" + endl
if self._use_moving_mask:
cmd += "--mmask " + self._dir_tmp + moving_mask_str + ".nii.gz" + endl
cmd += "--tout " + self._dir_tmp + registration_transform_str + ".txt" + endl
cmd += "--useAffine " + \
str(int(self._registration_type is "Affine")) + endl
cmd += "--useMultires " + \
str(int(self._use_multiresolution_framework)) + endl
cmd += "--metric " + self._metric + endl
cmd += "--scalesEst " + self._scales_estimator + endl
cmd += "--interpolator " + self._interpolator + endl
cmd += "--ANTSrad " + str(self._ANTSradius) + endl
# cmd += "--translationScale " + str(self._translation_scale) + endl
cmd += "--verbose " + verbose + endl
# Compute oriented Gaussian PSF if desired
if self._interpolator in ["OrientedGaussian"]:
if self._cov is None:
# Get oriented Gaussian covariance matrix
cov_HR_coord = psf.PSF().\
get_covariance_matrix_in_reconstruction_space(
self._moving, self._fixed).flatten()
else:
cov_HR_coord = self._cov.flatten()
cmd += "--cov " + "'" + ' '.join(cov_HR_coord.astype("|S12")) + "'"
# if self._use_verbose:
ph.execute_command(cmd)
# Read transformation file
params_all = np.loadtxt(self._dir_tmp + registration_transform_str +
".txt")
## (center_x, center_y, center_z, direction_fixed_image_flattened_0, ..., direction_fixed_image_flattened_8)
self._parameters_fixed = params_all[0:-7]
## (versor_0, versor_1, versor_2, translation_x, translation_y, translation_z, scale)
self._parameters = params_all[-7:]
# Get affine registration transform T(x) = R D Lambda D^{-1} (x-c) + t
# + c
self._registration_transform_sitk = self._get_affine_transform_from_similarity_registration(
)
def _run_registration_rigid_affine(self, id, endl=" \\\n"):
if self._fixed is None or self._moving is None:
raise ValueError("Error: Fixed and moving image not specified")
if self._use_verbose:
verbose = "1"
else:
verbose = "0"
moving_str = "RegistrationITK_moving_" + id + self._moving.get_filename(
)
fixed_str = "RegistrationITK_fixed_" + id + self._fixed.get_filename()
moving_mask_str = "RegistrationITK_moving_mask_" + id + self._moving.get_filename(
)
fixed_mask_str = "RegistrationITK_fixed_mask_" + id + self._fixed.get_filename(
)
registration_transform_str = "RegistrationITK_transform_" + id + \
self._fixed.get_filename() + "_" + self._moving.get_filename()
# Write images to HDD
# if not os.path.isfile(self._dir_tmp + moving_str + ".nii.gz"):
sitkh.write_nifti_image_sitk(self._moving.sitk,
self._dir_tmp + moving_str + ".nii.gz")
# if not os.path.isfile(self._dir_tmp + fixed_str + ".nii.gz"):
sitkh.write_nifti_image_sitk(self._fixed.sitk,
self._dir_tmp + fixed_str + ".nii.gz")
# if not os.path.isfile(self._dir_tmp + moving_mask_str + ".nii.gz")
# and self._use_moving_mask:
sitkh.write_nifti_image_sitk(
self._moving.sitk_mask,
self._dir_tmp + moving_mask_str + ".nii.gz")
# if not os.path.isfile(self._dir_tmp + fixed_mask_str + ".nii.gz") and
# self._use_fixed_mask:
sitkh.write_nifti_image_sitk(
self._fixed.sitk_mask, self._dir_tmp + fixed_mask_str + ".nii.gz")
# Prepare command for execution
# cmd = "/Users/mebner/UCL/UCL/Software/Volumetric\ Reconstruction/build/cpp/bin/itkReg "
cmd = DIR_CPP_BUILD + "/bin/itkReg" + endl
cmd += "--f " + self._dir_tmp + fixed_str + ".nii.gz" + endl
cmd += "--m " + self._dir_tmp + moving_str + ".nii.gz" + endl
if self._use_fixed_mask:
cmd += "--fmask " + self._dir_tmp + fixed_mask_str + ".nii.gz" + endl
if self._use_moving_mask:
cmd += "--mmask " + self._dir_tmp + moving_mask_str + ".nii.gz" + endl
cmd += "--tout " + self._dir_tmp + registration_transform_str + ".txt" + endl
cmd += "--useAffine " + \
str(int(self._registration_type is "Affine")) + endl
cmd += "--useMultires " + \
str(int(self._use_multiresolution_framework)) + endl
cmd += "--metric " + self._metric + endl
cmd += "--scalesEst " + self._scales_estimator + endl
cmd += "--interpolator " + self._interpolator + endl
cmd += "--ANTSrad " + str(self._ANTSradius) + endl
cmd += "--verbose " + verbose + endl
# Compute oriented Gaussian PSF if desired
if self._interpolator in ["OrientedGaussian"]:
if self._cov is None:
# Get oriented Gaussian covariance matrix
cov_HR_coord = psf.PSF().\
get_covariance_matrix_in_reconstruction_space(
self._moving, self._fixed).flatten()
else:
cov_HR_coord = self._cov.flatten()
cmd += "--cov " + "'" + ' '.join(cov_HR_coord.astype("|S12")) + "'"
ph.execute_command(cmd, verbose=0)
# Read transformation file
params_all = np.loadtxt(self._dir_tmp + registration_transform_str +
".txt")
if self._registration_type in ["Rigid"]:
self._parameters_fixed = params_all[0:4]
self._parameters = params_all[4:]
self._registration_transform_sitk = sitk.Euler3DTransform()
else:
self._parameters_fixed = params_all[0:3]
self._parameters = params_all[3:]
self._registration_transform_sitk = sitk.AffineTransform(3)
self._registration_transform_sitk.SetParameters(self._parameters)
self._registration_transform_sitk.SetFixedParameters(
self._parameters_fixed)
def test_run_simulation_view(self):
filename_HR_volume = "HR_volume_postmortem"
HR_volume = st.Stack.from_filename(
os.path.join(self.dir_test_data, filename_HR_volume + ".nii.gz"),
os.path.join(self.dir_test_data,
filename_HR_volume + "_mask.nii.gz"))
# 1) Test for Nearest Neighbor Interpolator
slice_acquistion = sa.SliceAcqusition(HR_volume)
slice_acquistion.set_interpolator_type("NearestNeighbor")
slice_acquistion.set_motion_type("NoMotion")
slice_acquistion.run_simulation_view_1()
slice_acquistion.run_simulation_view_2()
slice_acquistion.run_simulation_view_3()
stacks_simulated = slice_acquistion.get_simulated_stacks()
for i in range(0, len(stacks_simulated)):
HR_volume_resampled_sitk = sitk.Resample(
HR_volume.sitk, stacks_simulated[i].sitk,
sitk.Euler3DTransform(), sitk.sitkNearestNeighbor, 0.0,
stacks_simulated[i].sitk.GetPixelIDValue())
self.assertEqual(
np.round(np.linalg.norm(
sitk.GetArrayFromImage(stacks_simulated[i].sitk -
HR_volume_resampled_sitk)),
decimals=self.accuracy), 0)
# 2) Test for Linear Interpolator
slice_acquistion = sa.SliceAcqusition(HR_volume)
slice_acquistion.set_interpolator_type("Linear")
slice_acquistion.set_motion_type("NoMotion")
slice_acquistion.run_simulation_view_1()
slice_acquistion.run_simulation_view_2()
slice_acquistion.run_simulation_view_3()
stacks_simulated = slice_acquistion.get_simulated_stacks()
for i in range(0, len(stacks_simulated)):
HR_volume_resampled_sitk = sitk.Resample(
HR_volume.sitk, stacks_simulated[i].sitk,
sitk.Euler3DTransform(), sitk.sitkLinear, 0.0,
stacks_simulated[i].sitk.GetPixelIDValue())
self.assertEqual(
np.round(np.linalg.norm(
sitk.GetArrayFromImage(stacks_simulated[i].sitk -
HR_volume_resampled_sitk)),
decimals=self.accuracy), 0)
# 3) Test for Oriented Gaussian Interpolator
alpha_cut = 3
slice_acquistion = sa.SliceAcqusition(HR_volume)
slice_acquistion.set_interpolator_type("OrientedGaussian")
slice_acquistion.set_motion_type("NoMotion")
slice_acquistion.set_alpha_cut(alpha_cut)
slice_acquistion.run_simulation_view_1()
slice_acquistion.run_simulation_view_2()
slice_acquistion.run_simulation_view_3()
stacks_simulated = slice_acquistion.get_simulated_stacks()
resampler = itk.ResampleImageFilter[image_type, image_type].New()
resampler.SetDefaultPixelValue(0.0)
resampler.SetInput(HR_volume.itk)
interpolator = itk.OrientedGaussianInterpolateImageFunction[
image_type, pixel_type].New()
interpolator.SetAlpha(alpha_cut)
resampler.SetInterpolator(interpolator)
PSF = psf.PSF()
for i in range(0, len(stacks_simulated)):
resampler.SetOutputParametersFromImage(stacks_simulated[i].itk)
# Set covariance based on oblique PSF
Cov_HR_coord = PSF.get_covariance_matrix_in_reconstruction_space(
stacks_simulated[i], HR_volume)
interpolator.SetCovariance(Cov_HR_coord.flatten())
resampler.UpdateLargestPossibleRegion()
resampler.Update()
HR_volume_resampled_itk = resampler.GetOutput()
HR_volume_resampled_sitk = sitkh.get_sitk_from_itk_image(
HR_volume_resampled_itk)
self.assertEqual(
np.round(np.linalg.norm(
sitk.GetArrayFromImage(stacks_simulated[i].sitk -
HR_volume_resampled_sitk)),
decimals=self.accuracy), 0)
def test_ground_truth_affine_transforms_with_motion_OrientedGaussian(self):
filename_HR_volume = "HR_volume_postmortem"
HR_volume = st.Stack.from_filename(
os.path.join(self.dir_test_data, filename_HR_volume + ".nii.gz"),
os.path.join(self.dir_test_data,
filename_HR_volume + "_mask.nii.gz"))
alpha_cut = 3
# Run simulation for Oriented Gaussian interpolation, hence more
# "realistic" case
slice_acquistion = sa.SliceAcqusition(HR_volume)
slice_acquistion.set_interpolator_type("OrientedGaussian")
# slice_acquistion.set_interpolator_type("NearestNeighbor")
# slice_acquistion.set_interpolator_type("Linear")
slice_acquistion.set_motion_type("Random")
# slice_acquistion.set_motion_type("NoMotion")
slice_acquistion.set_alpha_cut(alpha_cut)
slice_acquistion.run_simulation_view_1()
slice_acquistion.run_simulation_view_2()
slice_acquistion.run_simulation_view_3()
# Get simulated stack of slices + corresponding ground truth affine
# transforms indicating the correct acquisition of the slice
# within the volume
stacks_simulated = slice_acquistion.get_simulated_stacks()
affine_transforms_ground_truth, rigid_motion_transforms_ground_truth = slice_acquistion.get_ground_truth_data(
)
resampler = itk.ResampleImageFilter[image_type, image_type].New()
resampler.SetDefaultPixelValue(0.0)
resampler.SetInput(HR_volume.itk)
interpolator = itk.OrientedGaussianInterpolateImageFunction[
image_type, pixel_type].New()
interpolator.SetAlpha(alpha_cut)
# interpolator = itk.LinearInterpolateImageFunction[image_type, pixel_type].New()
# interpolator = itk.NearestNeighborInterpolateImageFunction[image_type, pixel_type].New()
resampler.SetInterpolator(interpolator)
PSF = psf.PSF()
for i in range(0, len(stacks_simulated)):
stack = stacks_simulated[i]
slices = stack.get_slices()
N_slices = stack.get_number_of_slices()
for j in range(0, N_slices):
# print("Stack %s: Slice %s/%s" %(i,j,N_slices-1))
slice = slices[j]
slice.update_motion_correction(
rigid_motion_transforms_ground_truth[i][j])
# Get covariance based on oblique PSF
Cov_HR_coord = PSF.get_covariance_matrix_in_reconstruction_space(
slice, HR_volume)
# Update resampler
interpolator.SetCovariance(Cov_HR_coord.flatten())
resampler.SetOutputParametersFromImage(slice.itk)
resampler.UpdateLargestPossibleRegion()
resampler.Update()
HR_volume_resampled_slice_itk = resampler.GetOutput()
HR_volume_resampled_slice_sitk = sitkh.get_sitk_from_itk_image(
HR_volume_resampled_slice_itk)
# HR_volume_resampled_slice_sitk = sitk.Resample(
# HR_volume.sitk, slice.sitk, sitk.Euler3DTransform(), sitk.sitkNearestNeighbor, 0.0, slice.sitk.GetPixelIDValue()
# )
norm_diff = np.linalg.norm(
sitk.GetArrayFromImage(slice.sitk -
HR_volume_resampled_slice_sitk))
# try:
self.assertEqual(np.round(norm_diff, decimals=self.accuracy),
0)
def _run(self):
if self._use_fixed_mask:
fixed_sitk_mask = self._fixed.sitk_mask
else:
fixed_sitk_mask = None
if self._use_moving_mask:
moving_sitk_mask = self._moving.sitk_mask
else:
moving_sitk_mask = None
# Blur moving image with oriented Gaussian prior to the registration
if self._use_oriented_psf:
# Get oriented Gaussian covariance matrix
cov_HR_coord = psf.PSF(
).get_covariance_matrix_in_reconstruction_space(
self._fixed, self._moving)
# Create recursive YVV Gaussianfilter
image_type = itk.Image[itk.D, self._fixed.sitk.GetDimension()]
gaussian_yvv = itk.SmoothingRecursiveYvvGaussianImageFilter[
image_type, image_type].New()
# Feed Gaussian filter with axis aligned covariance matrix
sigma_axis_aligned = np.sqrt(np.diagonal(cov_HR_coord))
print("Oriented PSF blurring with (axis aligned) sigma = " +
str(sigma_axis_aligned))
print("\t(Based on computed covariance matrix = ")
for i in range(0, 3):
print("\t\t" + str(cov_HR_coord[i, :]))
print("\twith square root of diagonal " +
str(np.diagonal(cov_HR_coord)) + ")")
gaussian_yvv.SetInput(self._moving.itk)
gaussian_yvv.SetSigmaArray(sigma_axis_aligned)
gaussian_yvv.Update()
moving_itk = gaussian_yvv.GetOutput()
moving_itk.DisconnectPipeline()
moving_sitk = sitkh.get_sitk_from_itk_image(moving_itk)
else:
moving_sitk = self._moving.sitk
self._registration_method = \
simplereg.simple_itk_registration.SimpleItkRegistration(
fixed_sitk=self._fixed.sitk,
moving_sitk=moving_sitk,
fixed_sitk_mask=fixed_sitk_mask,
moving_sitk_mask=moving_sitk_mask,
registration_type=self._registration_type,
interpolator=self._interpolator,
metric=self._metric,
metric_params=self._metric_params,
optimizer=self._optimizer,
optimizer_params=self._optimizer_params,
initializer_type=self._initializer_type,
use_multiresolution_framework=self._use_multiresolution_framework,
optimizer_scales=self._scales_estimator,
shrink_factors=self._shrink_factors,
smoothing_sigmas=self._smoothing_sigmas,
verbose=self._use_verbose,
)
self._registration_method.run()
self._registration_transform_sitk = \
self._registration_method.get_registration_transform_sitk()
