def test_set_spacing(self):
for img in self.imgs:
# set spacing from list
new_spacing_list = [6.9]*img.dimension
ants.set_spacing(img, new_spacing_list)
self.assertEqual(img.spacing, tuple(new_spacing_list))
# set spacing from tuple
new_spacing_tuple = tuple(new_spacing_list)
ants.set_spacing(img, new_spacing_tuple)
self.assertEqual(ants.get_spacing(img), new_spacing_tuple)
def volume(label_image):
# 计算标签文件中各个标签的体积
image = ants.image_read(label_image).numpy()
max_num = np.int(np.max(image))
Volume = np.zeros(max_num + 1)
for label in range(1, max_num + 1):
Volume[label] = np.sum(image == label)
[x, y, z] = ants.get_spacing(ants.from_numpy(image))
voxel_volume = x * y * z
return Volume * voxel_volume
def lung_extraction(image,
modality="proton",
antsxnet_cache_directory=None,
verbose=False):
"""
Perform proton or ct lung extraction using U-net.
Arguments
---------
image : ANTsImage
input image
modality : string
Modality image type. Options include "ct", "proton", "protonLobes",
"maskLobes", and "ventilation".
antsxnet_cache_directory : string
Destination directory for storing the downloaded template and model weights.
Since these can be resused, if is None, these data will be downloaded to a
~/.keras/ANTsXNet/.
verbose : boolean
Print progress to the screen.
Returns
-------
Dictionary of ANTs segmentation and probability images.
Example
-------
>>> output = lung_extraction(lung_image, modality="proton")
"""
from ..architectures import create_unet_model_2d
from ..architectures import create_unet_model_3d
from ..utilities import get_pretrained_network
from ..utilities import get_antsxnet_data
from ..utilities import pad_or_crop_image_to_size
if image.dimension != 3:
raise ValueError( "Image dimension must be 3." )
if antsxnet_cache_directory == None:
antsxnet_cache_directory = "ANTsXNet"
image_mods = [modality]
channel_size = len(image_mods)
weights_file_name = None
unet_model = None
if modality == "proton":
weights_file_name = get_pretrained_network("protonLungMri",
antsxnet_cache_directory=antsxnet_cache_directory)
classes = ("background", "left_lung", "right_lung")
number_of_classification_labels = len(classes)
reorient_template_file_name_path = get_antsxnet_data("protonLungTemplate",
antsxnet_cache_directory=antsxnet_cache_directory)
reorient_template = ants.image_read(reorient_template_file_name_path)
resampled_image_size = reorient_template.shape
unet_model = create_unet_model_3d((*resampled_image_size, channel_size),
number_of_outputs=number_of_classification_labels,
number_of_layers=4, number_of_filters_at_base_layer=16, dropout_rate=0.0,
convolution_kernel_size=(7, 7, 5), deconvolution_kernel_size=(7, 7, 5))
unet_model.load_weights(weights_file_name)
if verbose == True:
print("Lung extraction: normalizing image to the template.")
center_of_mass_template = ants.get_center_of_mass(reorient_template * 0 + 1)
center_of_mass_image = ants.get_center_of_mass(image * 0 + 1)
translation = np.asarray(center_of_mass_image) - np.asarray(center_of_mass_template)
xfrm = ants.create_ants_transform(transform_type="Euler3DTransform",
center=np.asarray(center_of_mass_template), translation=translation)
warped_image = ants.apply_ants_transform_to_image(xfrm, image, reorient_template)
batchX = np.expand_dims(warped_image.numpy(), axis=0)
batchX = np.expand_dims(batchX, axis=-1)
batchX = (batchX - batchX.mean()) / batchX.std()
predicted_data = unet_model.predict(batchX, verbose=int(verbose))
origin = warped_image.origin
spacing = warped_image.spacing
direction = warped_image.direction
probability_images_array = list()
for i in range(number_of_classification_labels):
probability_images_array.append(
ants.from_numpy(np.squeeze(predicted_data[0, :, :, :, i]),
origin=origin, spacing=spacing, direction=direction))
if verbose == True:
print("Lung extraction: renormalize probability mask to native space.")
for i in range(number_of_classification_labels):
probability_images_array[i] = ants.apply_ants_transform_to_image(
ants.invert_ants_transform(xfrm), probability_images_array[i], image)
image_matrix = ants.image_list_to_matrix(probability_images_array, image * 0 + 1)
segmentation_matrix = np.argmax(image_matrix, axis=0)
segmentation_image = ants.matrix_to_images(
np.expand_dims(segmentation_matrix, axis=0), image * 0 + 1)[0]
return_dict = {'segmentation_image' : segmentation_image,
'probability_images' : probability_images_array}
return(return_dict)
if modality == "protonLobes" or modality == "maskLobes":
reorient_template_file_name_path = get_antsxnet_data("protonLungTemplate",
antsxnet_cache_directory=antsxnet_cache_directory)
reorient_template = ants.image_read(reorient_template_file_name_path)
resampled_image_size = reorient_template.shape
spatial_priors_file_name_path = get_antsxnet_data("protonLobePriors",
antsxnet_cache_directory=antsxnet_cache_directory)
spatial_priors = ants.image_read(spatial_priors_file_name_path)
priors_image_list = ants.ndimage_to_list(spatial_priors)
channel_size = 1 + len(priors_image_list)
number_of_classification_labels = 1 + len(priors_image_list)
unet_model = create_unet_model_3d((*resampled_image_size, channel_size),
number_of_outputs=number_of_classification_labels, mode="classification",
number_of_filters_at_base_layer=16, number_of_layers=4,
convolution_kernel_size=(3, 3, 3), deconvolution_kernel_size=(2, 2, 2),
dropout_rate=0.0, weight_decay=0, additional_options=("attentionGating",))
if modality == "protonLobes":
penultimate_layer = unet_model.layers[-2].output
outputs2 = Conv3D(filters=1,
kernel_size=(1, 1, 1),
activation='sigmoid',
kernel_regularizer=regularizers.l2(0.0))(penultimate_layer)
unet_model = Model(inputs=unet_model.input, outputs=[unet_model.output, outputs2])
weights_file_name = get_pretrained_network("protonLobes",
antsxnet_cache_directory=antsxnet_cache_directory)
else:
weights_file_name = get_pretrained_network("maskLobes",
antsxnet_cache_directory=antsxnet_cache_directory)
unet_model.load_weights(weights_file_name)
if verbose == True:
print("Lung extraction: normalizing image to the template.")
center_of_mass_template = ants.get_center_of_mass(reorient_template * 0 + 1)
center_of_mass_image = ants.get_center_of_mass(image * 0 + 1)
translation = np.asarray(center_of_mass_image) - np.asarray(center_of_mass_template)
xfrm = ants.create_ants_transform(transform_type="Euler3DTransform",
center=np.asarray(center_of_mass_template), translation=translation)
warped_image = ants.apply_ants_transform_to_image(xfrm, image, reorient_template)
warped_array = warped_image.numpy()
if modality == "protonLobes":
warped_array = (warped_array - warped_array.mean()) / warped_array.std()
else:
warped_array[warped_array != 0] = 1
batchX = np.zeros((1, *warped_array.shape, channel_size))
batchX[0,:,:,:,0] = warped_array
for i in range(len(priors_image_list)):
batchX[0,:,:,:,i+1] = priors_image_list[i].numpy()
predicted_data = unet_model.predict(batchX, verbose=int(verbose))
origin = warped_image.origin
spacing = warped_image.spacing
direction = warped_image.direction
probability_images_array = list()
for i in range(number_of_classification_labels):
if modality == "protonLobes":
probability_images_array.append(
ants.from_numpy(np.squeeze(predicted_data[0][0, :, :, :, i]),
origin=origin, spacing=spacing, direction=direction))
else:
probability_images_array.append(
ants.from_numpy(np.squeeze(predicted_data[0, :, :, :, i]),
origin=origin, spacing=spacing, direction=direction))
if verbose == True:
print("Lung extraction: renormalize probability images to native space.")
for i in range(number_of_classification_labels):
probability_images_array[i] = ants.apply_ants_transform_to_image(
ants.invert_ants_transform(xfrm), probability_images_array[i], image)
image_matrix = ants.image_list_to_matrix(probability_images_array, image * 0 + 1)
segmentation_matrix = np.argmax(image_matrix, axis=0)
segmentation_image = ants.matrix_to_images(
np.expand_dims(segmentation_matrix, axis=0), image * 0 + 1)[0]
if modality == "protonLobes":
whole_lung_mask = ants.from_numpy(np.squeeze(predicted_data[1][0, :, :, :, 0]),
origin=origin, spacing=spacing, direction=direction)
whole_lung_mask = ants.apply_ants_transform_to_image(
ants.invert_ants_transform(xfrm), whole_lung_mask, image)
return_dict = {'segmentation_image' : segmentation_image,
'probability_images' : probability_images_array,
'whole_lung_mask_image' : whole_lung_mask}
return(return_dict)
else:
return_dict = {'segmentation_image' : segmentation_image,
'probability_images' : probability_images_array}
return(return_dict)
elif modality == "ct":
################################
#
# Preprocess image
#
################################
if verbose == True:
print("Preprocess CT image.")
def closest_simplified_direction_matrix(direction):
closest = np.floor(np.abs(direction) + 0.5)
closest[direction < 0] *= -1.0
return closest
simplified_direction = closest_simplified_direction_matrix(image.direction)
reference_image_size = (128, 128, 128)
ct_preprocessed = ants.resample_image(image, reference_image_size, use_voxels=True, interp_type=0)
ct_preprocessed[ct_preprocessed < -1000] = -1000
ct_preprocessed[ct_preprocessed > 400] = 400
ct_preprocessed.set_direction(simplified_direction)
ct_preprocessed.set_origin((0, 0, 0))
ct_preprocessed.set_spacing((1, 1, 1))
################################
#
# Reorient image
#
################################
reference_image = ants.make_image(reference_image_size,
voxval=0,
spacing=(1, 1, 1),
origin=(0, 0, 0),
direction=np.identity(3))
center_of_mass_reference = np.floor(ants.get_center_of_mass(reference_image * 0 + 1))
center_of_mass_image = np.floor(ants.get_center_of_mass(ct_preprocessed * 0 + 1))
translation = np.asarray(center_of_mass_image) - np.asarray(center_of_mass_reference)
xfrm = ants.create_ants_transform(transform_type="Euler3DTransform",
center=np.asarray(center_of_mass_reference), translation=translation)
ct_preprocessed = ((ct_preprocessed - ct_preprocessed.min()) /
(ct_preprocessed.max() - ct_preprocessed.min()))
ct_preprocessed_warped = ants.apply_ants_transform_to_image(
xfrm, ct_preprocessed, reference_image, interpolation="nearestneighbor")
ct_preprocessed_warped = ((ct_preprocessed_warped - ct_preprocessed_warped.min()) /
(ct_preprocessed_warped.max() - ct_preprocessed_warped.min())) - 0.5
################################
#
# Build models and load weights
#
################################
if verbose == True:
print("Build model and load weights.")
weights_file_name = get_pretrained_network("lungCtWithPriorsSegmentationWeights",
antsxnet_cache_directory=antsxnet_cache_directory)
classes = ("background", "left lung", "right lung", "airways")
number_of_classification_labels = len(classes)
luna16_priors = ants.ndimage_to_list(ants.image_read(get_antsxnet_data("luna16LungPriors")))
for i in range(len(luna16_priors)):
luna16_priors[i] = ants.resample_image(luna16_priors[i], reference_image_size, use_voxels=True)
channel_size = len(luna16_priors) + 1
unet_model = create_unet_model_3d((*reference_image_size, channel_size),
number_of_outputs=number_of_classification_labels, mode="classification",
number_of_layers=4, number_of_filters_at_base_layer=16, dropout_rate=0.0,
convolution_kernel_size=(3, 3, 3), deconvolution_kernel_size=(2, 2, 2),
weight_decay=1e-5, additional_options=("attentionGating",))
unet_model.load_weights(weights_file_name)
################################
#
# Do prediction and normalize to native space
#
################################
if verbose == True:
print("Prediction.")
batchX = np.zeros((1, *reference_image_size, channel_size))
batchX[:,:,:,:,0] = ct_preprocessed_warped.numpy()
for i in range(len(luna16_priors)):
batchX[:,:,:,:,i+1] = luna16_priors[i].numpy() - 0.5
predicted_data = unet_model.predict(batchX, verbose=verbose)
probability_images = list()
for i in range(number_of_classification_labels):
if verbose == True:
print("Reconstructing image", classes[i])
probability_image = ants.from_numpy(np.squeeze(predicted_data[:,:,:,:,i]),
origin=ct_preprocessed_warped.origin, spacing=ct_preprocessed_warped.spacing,
direction=ct_preprocessed_warped.direction)
probability_image = ants.apply_ants_transform_to_image(
ants.invert_ants_transform(xfrm), probability_image, ct_preprocessed)
probability_image = ants.resample_image(probability_image,
resample_params=image.shape, use_voxels=True, interp_type=0)
probability_image = ants.copy_image_info(image, probability_image)
probability_images.append(probability_image)
image_matrix = ants.image_list_to_matrix(probability_images, image * 0 + 1)
segmentation_matrix = np.argmax(image_matrix, axis=0)
segmentation_image = ants.matrix_to_images(
np.expand_dims(segmentation_matrix, axis=0), image * 0 + 1)[0]
return_dict = {'segmentation_image' : segmentation_image,
'probability_images' : probability_images}
return(return_dict)
elif modality == "ventilation":
################################
#
# Preprocess image
#
################################
if verbose == True:
print("Preprocess ventilation image.")
template_size = (256, 256)
image_modalities = ("Ventilation",)
channel_size = len(image_modalities)
preprocessed_image = (image - image.mean()) / image.std()
ants.set_direction(preprocessed_image, np.identity(3))
################################
#
# Build models and load weights
#
################################
unet_model = create_unet_model_2d((*template_size, channel_size),
number_of_outputs=1, mode='sigmoid',
number_of_layers=4, number_of_filters_at_base_layer=32, dropout_rate=0.0,
convolution_kernel_size=(3, 3), deconvolution_kernel_size=(2, 2),
weight_decay=0)
if verbose == True:
print("Whole lung mask: retrieving model weights.")
weights_file_name = get_pretrained_network("wholeLungMaskFromVentilation",
antsxnet_cache_directory=antsxnet_cache_directory)
unet_model.load_weights(weights_file_name)
################################
#
# Extract slices
#
################################
spacing = ants.get_spacing(preprocessed_image)
dimensions_to_predict = (spacing.index(max(spacing)),)
total_number_of_slices = 0
for d in range(len(dimensions_to_predict)):
total_number_of_slices += preprocessed_image.shape[dimensions_to_predict[d]]
batchX = np.zeros((total_number_of_slices, *template_size, channel_size))
slice_count = 0
for d in range(len(dimensions_to_predict)):
number_of_slices = preprocessed_image.shape[dimensions_to_predict[d]]
if verbose == True:
print("Extracting slices for dimension ", dimensions_to_predict[d], ".")
for i in range(number_of_slices):
ventilation_slice = pad_or_crop_image_to_size(ants.slice_image(preprocessed_image, dimensions_to_predict[d], i), template_size)
batchX[slice_count,:,:,0] = ventilation_slice.numpy()
slice_count += 1
################################
#
# Do prediction and then restack into the image
#
################################
if verbose == True:
print("Prediction.")
prediction = unet_model.predict(batchX, verbose=verbose)
permutations = list()
permutations.append((0, 1, 2))
permutations.append((1, 0, 2))
permutations.append((1, 2, 0))
probability_image = ants.image_clone(image) * 0
current_start_slice = 0
for d in range(len(dimensions_to_predict)):
current_end_slice = current_start_slice + preprocessed_image.shape[dimensions_to_predict[d]] - 1
which_batch_slices = range(current_start_slice, current_end_slice)
prediction_per_dimension = prediction[which_batch_slices,:,:,0]
prediction_array = np.transpose(np.squeeze(prediction_per_dimension), permutations[dimensions_to_predict[d]])
prediction_image = ants.copy_image_info(image,
pad_or_crop_image_to_size(ants.from_numpy(prediction_array),
image.shape))
probability_image = probability_image + (prediction_image - probability_image) / (d + 1)
current_start_slice = current_end_slice + 1
return(probability_image)
else:
return ValueError("Unrecognized modality.")
# Extracting canonical functional network maps
## preprocessing
csfAndWM = ( ants.threshold_image( boldseg, 1, 1 ) +
ants.threshold_image( boldseg, 3, 3 ) ).morphology("erode",1)
bold = ants.image_read( boldfnsR )
boldList = ants.ndimage_to_list( bold )
avgBold = ants.get_average_of_timeseries( bold, range( 5 ) )
boldUndTX = ants.registration( und, avgBold, "SyN", regIterations = (15,4),
synMetric = "CC", synSampling = 2, verbose = False )
boldUndTS = ants.apply_transforms( und, bold, boldUndTX['fwdtransforms'], imagetype = 3 )
motCorr = ants.motion_correction( boldUndTS, avgBold,
type_of_transform="Rigid", verbose = True )
tr = ants.get_spacing( bold )[3]
highMotionTimes = np.where( motCorr['FD'] >= 0.5 )
goodtimes = np.where( motCorr['FD'] < 0.5 )
avgBold = ants.get_average_of_timeseries( motCorr['motion_corrected'], range( 5 ) )
#######################
nt = len(motCorr['FD'])
plt.plot( range( nt ), motCorr['FD'] )
plt.show()
#################################################
mycompcor = ants.compcor( motCorr['motion_corrected'],
filter_type='polynomial', degree=4 )
##########
smth = ( 1.0, 1.0, 1.0, 2.0 ) # this is for sigmaInPhysicalCoordinates = F
simg = ants.smooth_image( motCorr['motion_corrected'], smth, sigma_in_physical_coordinates = False )
gmseg = ants.threshold_image( boldseg, 2, 2 )
outputPrefix = outputDirectory + 'antspy'
numberOfOuterIterations = 5
image = ants.image_read( dataDirectory + 'KKI2009-01-MPRAGE_slice150.nii.gz', dimension = 2 )
mask = ants.image_read( dataDirectory + 'KKI2009-01-MPRAGE_slice150_mask.nii.gz', dimension = 2 )
weightMask = None
for i in range( numberOfOuterIterations ):
print( "*************** N4 <---> Atropos iteration ", i, " ******************\n" )
n4Results = ants.n4_bias_field_correction( image, mask = mask,
weight_mask = weightMask, verbose = True )
image = n4Results
atroposResults = ants.atropos( a = image, x = mask )
onesImage = ants.make_image( image.shape, voxval = 0,
spacing = ants.get_spacing( image ),
origin = ants.get_origin( image ),
direction = ants.get_direction( image ) )
weightMask = atroposResults['probabilityimages'][1] *\
( onesImage - atroposResults['probabilityimages'][0] ) *\
( onesImage - atroposResults['probabilityimages'][2] ) +\
atroposResults['probabilityimages'][2] *\
( onesImage - atroposResults['probabilityimages'][1] ) *\
( onesImage - atroposResults['probabilityimages'][0] )
ants.image_write( image, outputPrefix + "N4Corrected.nii.gz" )
ants.image_write( atroposResults['segmentation'], outputPrefix + "AtroposSegmentation.nii.gz" )
for i in range( len( atroposResults['probabilityimages'] ) ):
ants.image_write( atroposResults['probabilityimages'][i],
outputPrefix + "AtroposSegmentationProbability" + str( i ) + ".nii.gz" )
def el_bicho(ventilation_image,
mask,
use_coarse_slices_only=True,
antsxnet_cache_directory=None,
verbose=False):
"""
Perform functional lung segmentation using hyperpolarized gases.
https://pubmed.ncbi.nlm.nih.gov/30195415/
Arguments
---------
ventilation_image : ANTsImage
input ventilation image.
mask : ANTsImage
input mask.
use_coarse_slices_only : boolean
If True, apply network only in the dimension of greatest slice thickness.
If False, apply to all dimensions and average the results.
antsxnet_cache_directory : string
Destination directory for storing the downloaded template and model weights.
Since these can be resused, if is None, these data will be downloaded to a
~/.keras/ANTsXNet/.
verbose : boolean
Print progress to the screen.
Returns
-------
Ventilation segmentation and corresponding probability images
Example
-------
>>> image = ants.image_read("ventilation.nii.gz")
>>> mask = ants.image_read("mask.nii.gz")
>>> lung_seg = el_bicho(image, mask, use_coarse_slices=True, verbose=False)
"""
from ..architectures import create_unet_model_2d
from ..utilities import get_pretrained_network
from ..utilities import pad_or_crop_image_to_size
if ventilation_image.dimension != 3:
raise ValueError("Image dimension must be 3.")
if ventilation_image.shape != mask.shape:
raise ValueError(
"Ventilation image and mask size are not the same size.")
if antsxnet_cache_directory == None:
antsxnet_cache_directory = "ANTsXNet"
################################
#
# Preprocess image
#
################################
template_size = (256, 256)
classes = (0, 1, 2, 3, 4)
number_of_classification_labels = len(classes)
image_modalities = ("Ventilation", "Mask")
channel_size = len(image_modalities)
preprocessed_image = (ventilation_image -
ventilation_image.mean()) / ventilation_image.std()
ants.set_direction(preprocessed_image, np.identity(3))
mask_identity = ants.image_clone(mask)
ants.set_direction(mask_identity, np.identity(3))
################################
#
# Build models and load weights
#
################################
unet_model = create_unet_model_2d(
(*template_size, channel_size),
number_of_outputs=number_of_classification_labels,
number_of_layers=4,
number_of_filters_at_base_layer=32,
dropout_rate=0.0,
convolution_kernel_size=(3, 3),
deconvolution_kernel_size=(2, 2),
weight_decay=1e-5,
additional_options=("attentionGating"))
if verbose == True:
print("El Bicho: retrieving model weights.")
weights_file_name = get_pretrained_network(
"elBicho", antsxnet_cache_directory=antsxnet_cache_directory)
unet_model.load_weights(weights_file_name)
################################
#
# Extract slices
#
################################
spacing = ants.get_spacing(preprocessed_image)
dimensions_to_predict = (spacing.index(max(spacing)), )
if use_coarse_slices_only == False:
dimensions_to_predict = list(range(3))
total_number_of_slices = 0
for d in range(len(dimensions_to_predict)):
total_number_of_slices += preprocessed_image.shape[
dimensions_to_predict[d]]
batchX = np.zeros((total_number_of_slices, *template_size, channel_size))
slice_count = 0
for d in range(len(dimensions_to_predict)):
number_of_slices = preprocessed_image.shape[dimensions_to_predict[d]]
if verbose == True:
print("Extracting slices for dimension ", dimensions_to_predict[d],
".")
for i in range(number_of_slices):
ventilation_slice = pad_or_crop_image_to_size(
ants.slice_image(preprocessed_image, dimensions_to_predict[d],
i), template_size)
batchX[slice_count, :, :, 0] = ventilation_slice.numpy()
mask_slice = pad_or_crop_image_to_size(
ants.slice_image(mask_identity, dimensions_to_predict[d], i),
template_size)
batchX[slice_count, :, :, 1] = mask_slice.numpy()
slice_count += 1
################################
#
# Do prediction and then restack into the image
#
################################
if verbose == True:
print("Prediction.")
prediction = unet_model.predict(batchX, verbose=verbose)
permutations = list()
permutations.append((0, 1, 2))
permutations.append((1, 0, 2))
permutations.append((1, 2, 0))
probability_images = list()
for l in range(number_of_classification_labels):
probability_images.append(ants.image_clone(mask) * 0)
current_start_slice = 0
for d in range(len(dimensions_to_predict)):
current_end_slice = current_start_slice + preprocessed_image.shape[
dimensions_to_predict[d]] - 1
which_batch_slices = range(current_start_slice, current_end_slice)
for l in range(number_of_classification_labels):
prediction_per_dimension = prediction[which_batch_slices, :, :, l]
prediction_array = np.transpose(
np.squeeze(prediction_per_dimension),
permutations[dimensions_to_predict[d]])
prediction_image = ants.copy_image_info(
ventilation_image,
pad_or_crop_image_to_size(ants.from_numpy(prediction_array),
ventilation_image.shape))
probability_images[l] = probability_images[l] + (
prediction_image - probability_images[l]) / (d + 1)
current_start_slice = current_end_slice + 1
################################
#
# Convert probability images to segmentation
#
################################
image_matrix = ants.image_list_to_matrix(
probability_images[1:(len(probability_images))], mask * 0 + 1)
background_foreground_matrix = np.stack([
ants.image_list_to_matrix([probability_images[0]], mask * 0 + 1),
np.expand_dims(np.sum(image_matrix, axis=0), axis=0)
])
foreground_matrix = np.argmax(background_foreground_matrix, axis=0)
segmentation_matrix = (np.argmax(image_matrix, axis=0) +
1) * foreground_matrix
segmentation_image = ants.matrix_to_images(
np.expand_dims(segmentation_matrix, axis=0), mask * 0 + 1)[0]
return_dict = {
'segmentation_image': segmentation_image,
'probability_images': probability_images
}
return (return_dict)
def ew_david(flair,
t1,
do_preprocessing=True,
do_slicewise=True,
antsxnet_cache_directory=None,
verbose=False):
"""
Perform White matter hypterintensity probabilistic segmentation
using deep learning
Preprocessing on the training data consisted of:
* n4 bias correction,
* brain extraction, and
* affine registration to MNI.
The input T1 should undergo the same steps. If the input T1 is the raw
T1, these steps can be performed by the internal preprocessing, i.e. set
\code{doPreprocessing = TRUE}
Arguments
---------
flair : ANTsImage
input 3-D FLAIR brain image (not skull-stripped).
t1 : ANTsImage
input 3-D T1 brain image (not skull-stripped).
do_preprocessing : boolean
perform n4 bias correction?
do_slicewise : boolean
apply 2-D modal along direction of maximal slice thickness.
verbose : boolean
Print progress to the screen.
Returns
-------
WMH segmentation probability image
Example
-------
>>> image = ants.image_read("flair.nii.gz")
>>> probability_mask = sysu_media_wmh_segmentation(image)
"""
from ..architectures import create_unet_model_2d
from ..architectures import create_unet_model_3d
from ..utilities import get_pretrained_network
from ..utilities import preprocess_brain_image
from ..utilities import extract_image_patches
from ..utilities import reconstruct_image_from_patches
from ..utilities import pad_or_crop_image_to_size
if flair.dimension != 3:
raise ValueError( "Image dimension must be 3." )
if t1.dimension != 3:
raise ValueError( "Image dimension must be 3." )
if antsxnet_cache_directory == None:
antsxnet_cache_directory = "ANTsXNet"
if do_slicewise == False:
################################
#
# Preprocess images
#
################################
t1_preprocessed = t1
t1_preprocessing = None
if do_preprocessing == True:
t1_preprocessing = preprocess_brain_image(t1,
truncate_intensity=(0.01, 0.99),
do_brain_extraction=True,
template="croppedMni152",
template_transform_type="AffineFast",
do_bias_correction=True,
do_denoising=False,
antsxnet_cache_directory=antsxnet_cache_directory,
verbose=verbose)
t1_preprocessed = t1_preprocessing["preprocessed_image"] * t1_preprocessing['brain_mask']
flair_preprocessed = flair
if do_preprocessing == True:
flair_preprocessing = preprocess_brain_image(flair,
truncate_intensity=(0.01, 0.99),
do_brain_extraction=False,
do_bias_correction=True,
do_denoising=False,
antsxnet_cache_directory=antsxnet_cache_directory,
verbose=verbose)
flair_preprocessed = ants.apply_transforms(fixed=t1_preprocessed,
moving=flair_preprocessing["preprocessed_image"],
transformlist=t1_preprocessing['template_transforms']['fwdtransforms'])
flair_preprocessed = flair_preprocessed * t1_preprocessing['brain_mask']
################################
#
# Build model and load weights
#
################################
patch_size = (112, 112, 112)
stride_length = (t1_preprocessed.shape[0] - patch_size[0],
t1_preprocessed.shape[1] - patch_size[1],
t1_preprocessed.shape[2] - patch_size[2])
classes = ("background", "wmh" )
number_of_classification_labels = len(classes)
labels = (0, 1)
image_modalities = ("T1", "FLAIR")
channel_size = len(image_modalities)
unet_model = create_unet_model_3d((*patch_size, channel_size),
number_of_outputs = number_of_classification_labels,
number_of_layers = 4, number_of_filters_at_base_layer = 16, dropout_rate = 0.0,
convolution_kernel_size = (3, 3, 3), deconvolution_kernel_size = (2, 2, 2),
weight_decay = 1e-5, nn_unet_activation_style=False, add_attention_gating=True)
weights_file_name = get_pretrained_network("ewDavidWmhSegmentationWeights",
antsxnet_cache_directory=antsxnet_cache_directory)
unet_model.load_weights(weights_file_name)
################################
#
# Do prediction and normalize to native space
#
################################
if verbose == True:
print("ew_david: prediction.")
batchX = np.zeros((8, *patch_size, channel_size))
t1_preprocessed = (t1_preprocessed - t1_preprocessed.mean()) / t1_preprocessed.std()
t1_patches = extract_image_patches(t1_preprocessed, patch_size=patch_size,
max_number_of_patches="all", stride_length=stride_length,
return_as_array=True)
batchX[:,:,:,:,0] = t1_patches
flair_preprocessed = (flair_preprocessed - flair_preprocessed.mean()) / flair_preprocessed.std()
flair_patches = extract_image_patches(flair_preprocessed, patch_size=patch_size,
max_number_of_patches="all", stride_length=stride_length,
return_as_array=True)
batchX[:,:,:,:,1] = flair_patches
predicted_data = unet_model.predict(batchX, verbose=verbose)
probability_images = list()
for i in range(len(labels)):
print("Reconstructing image", classes[i])
reconstructed_image = reconstruct_image_from_patches(predicted_data[:,:,:,:,i],
domain_image=t1_preprocessed, stride_length=stride_length)
if do_preprocessing == True:
probability_images.append(ants.apply_transforms(fixed=t1,
moving=reconstructed_image,
transformlist=t1_preprocessing['template_transforms']['invtransforms'],
whichtoinvert=[True], interpolator="linear", verbose=verbose))
else:
probability_images.append(reconstructed_image)
return(probability_images[1])
else: # do_slicewise
################################
#
# Preprocess images
#
################################
t1_preprocessed = t1
t1_preprocessing = None
if do_preprocessing == True:
t1_preprocessing = preprocess_brain_image(t1,
truncate_intensity=(0.01, 0.99),
do_brain_extraction=False,
do_bias_correction=True,
do_denoising=False,
antsxnet_cache_directory=antsxnet_cache_directory,
verbose=verbose)
t1_preprocessed = t1_preprocessing["preprocessed_image"]
flair_preprocessed = flair
if do_preprocessing == True:
flair_preprocessing = preprocess_brain_image(flair,
truncate_intensity=(0.01, 0.99),
do_brain_extraction=False,
do_bias_correction=True,
do_denoising=False,
antsxnet_cache_directory=antsxnet_cache_directory,
verbose=verbose)
flair_preprocessed = flair_preprocessing["preprocessed_image"]
resampling_params = list(ants.get_spacing(flair_preprocessed))
do_resampling = False
for d in range(len(resampling_params)):
if resampling_params[d] < 0.8:
resampling_params[d] = 1.0
do_resampling = True
resampling_params = tuple(resampling_params)
if do_resampling:
flair_preprocessed = ants.resample_image(flair_preprocessed, resampling_params, use_voxels=False, interp_type=0)
t1_preprocessed = ants.resample_image(t1_preprocessed, resampling_params, use_voxels=False, interp_type=0)
flair_preprocessed = (flair_preprocessed - flair_preprocessed.mean()) / flair_preprocessed.std()
t1_preprocessed = (t1_preprocessed - t1_preprocessed.mean()) / t1_preprocessed.std()
################################
#
# Build model and load weights
#
################################
template_size = (256, 256)
classes = ("background", "wmh" )
number_of_classification_labels = len(classes)
labels = (0, 1)
image_modalities = ("T1", "FLAIR")
channel_size = len(image_modalities)
unet_model = create_unet_model_2d((*template_size, channel_size),
number_of_outputs = number_of_classification_labels,
number_of_layers = 4, number_of_filters_at_base_layer = 32, dropout_rate = 0.0,
convolution_kernel_size = (3, 3), deconvolution_kernel_size = (2, 2),
weight_decay = 1e-5, nn_unet_activation_style=True, add_attention_gating=True)
if verbose == True:
print("ewDavid: retrieving model weights.")
weights_file_name = get_pretrained_network("ewDavidWmhSegmentationSlicewiseWeights",
antsxnet_cache_directory=antsxnet_cache_directory)
unet_model.load_weights(weights_file_name)
################################
#
# Extract slices
#
################################
use_coarse_slices_only = True
spacing = ants.get_spacing(flair_preprocessed)
dimensions_to_predict = (spacing.index(max(spacing)),)
if use_coarse_slices_only == False:
dimensions_to_predict = list(range(3))
total_number_of_slices = 0
for d in range(len(dimensions_to_predict)):
total_number_of_slices += flair_preprocessed.shape[dimensions_to_predict[d]]
batchX = np.zeros((total_number_of_slices, *template_size, channel_size))
slice_count = 0
for d in range(len(dimensions_to_predict)):
number_of_slices = flair_preprocessed.shape[dimensions_to_predict[d]]
if verbose == True:
print("Extracting slices for dimension ", dimensions_to_predict[d], ".")
for i in range(number_of_slices):
flair_slice = pad_or_crop_image_to_size(ants.slice_image(flair_preprocessed, dimensions_to_predict[d], i), template_size)
batchX[slice_count,:,:,0] = flair_slice.numpy()
t1_slice = pad_or_crop_image_to_size(ants.slice_image(t1_preprocessed, dimensions_to_predict[d], i), template_size)
batchX[slice_count,:,:,1] = t1_slice.numpy()
slice_count += 1
################################
#
# Do prediction and then restack into the image
#
################################
if verbose == True:
print("Prediction.")
prediction = unet_model.predict(batchX, verbose=verbose)
permutations = list()
permutations.append((0, 1, 2))
permutations.append((1, 0, 2))
permutations.append((1, 2, 0))
prediction_image_average = ants.image_clone(flair_preprocessed) * 0
current_start_slice = 0
for d in range(len(dimensions_to_predict)):
current_end_slice = current_start_slice + flair_preprocessed.shape[dimensions_to_predict[d]] - 1
which_batch_slices = range(current_start_slice, current_end_slice)
prediction_per_dimension = prediction[which_batch_slices,:,:,1]
prediction_array = np.transpose(np.squeeze(prediction_per_dimension), permutations[dimensions_to_predict[d]])
prediction_image = ants.copy_image_info(flair_preprocessed,
pad_or_crop_image_to_size(ants.from_numpy(prediction_array),
flair_preprocessed.shape))
prediction_image_average = prediction_image_average + (prediction_image - prediction_image_average) / (d + 1)
current_start_slice = current_end_slice + 1
if do_resampling:
prediction_image_average = ants.resample_image_to_target(prediction_image_average, flair)
return(prediction_image_average)
def test_get_spacing(self):
for img in self.imgs:
spacing = ants.get_spacing(img)
self.assertTrue(isinstance(spacing, tuple))
self.assertEqual(len(ants.get_spacing(img)), img.dimension)
def ew_david(flair,
t1,
do_preprocessing=True,
which_model="sysu",
which_axes=2,
number_of_simulations=0,
sd_affine=0.01,
antsxnet_cache_directory=None,
verbose=False):
"""
Perform White matter hyperintensity probabilistic segmentation
using deep learning
Preprocessing on the training data consisted of:
* n4 bias correction,
* intensity truncation,
* brain extraction, and
* affine registration to MNI.
The input T1 should undergo the same steps. If the input T1 is the raw
T1, these steps can be performed by the internal preprocessing, i.e. set
\code{do_preprocessing = True}
Arguments
---------
flair : ANTsImage
input 3-D FLAIR brain image (not skull-stripped).
t1 : ANTsImage
input 3-D T1 brain image (not skull-stripped).
do_preprocessing : boolean
perform n4 bias correction, intensity truncation, brain extraction.
which_model : string
one of:
* "sysu" -- same as the original sysu network (without site specific preprocessing),
* "sysu-ri" -- same as "sysu" but using ranked intensity scaling for input images,
* "sysuWithAttention" -- "sysu" with attention gating,
* "sysuWithAttentionAndSite" -- "sysu" with attention gating with site branch (see "sysuWithSite"),
* "sysuPlus" -- "sysu" with attention gating and nn-Unet activation,
* "sysuPlusSeg" -- "sysuPlus" with deep_atropos segmentation in an additional channel, and
* "sysuWithSite" -- "sysu" with global pooling on encoding channels to predict "site".
* "sysuPlusSegWithSite" -- "sysuPlusSeg" combined with "sysuWithSite"
In addition to both modalities, all models have T1-only and flair-only variants except
for "sysuPlusSeg" (which only has a T1-only variant) or "sysu-ri" (which has neither single
modality variant).
which_axes : string or scalar or tuple/vector
apply 2-D model to 1 or more axes. In addition to a scalar
or vector, e.g., which_axes = (0, 2), one can use "max" for the
axis with maximum anisotropy (default) or "all" for all axes.
number_of_simulations : integer
Number of random affine perturbations to transform the input.
sd_affine : float
Define the standard deviation of the affine transformation parameter.
antsxnet_cache_directory : string
Destination directory for storing the downloaded template and model weights.
Since these can be resused, if is None, these data will be downloaded to a
~/.keras/ANTsXNet/.
verbose : boolean
Print progress to the screen.
Returns
-------
WMH segmentation probability image
Example
-------
>>> image = ants.image_read("flair.nii.gz")
>>> probability_mask = sysu_media_wmh_segmentation(image)
"""
from ..architectures import create_unet_model_2d
from ..utilities import deep_atropos
from ..utilities import get_pretrained_network
from ..utilities import preprocess_brain_image
from ..utilities import randomly_transform_image_data
from ..utilities import pad_or_crop_image_to_size
do_t1_only = False
do_flair_only = False
if flair is None and t1 is not None:
do_t1_only = True
elif flair is not None and t1 is None:
do_flair_only = True
use_t1_segmentation = False
if "Seg" in which_model:
if do_flair_only:
raise ValueError("Segmentation requires T1.")
else:
use_t1_segmentation = True
if use_t1_segmentation and do_preprocessing == False:
raise ValueError(
"Using the t1 segmentation requires do_preprocessing=True.")
if antsxnet_cache_directory == None:
antsxnet_cache_directory = "ANTsXNet"
do_slicewise = True
if do_slicewise == False:
raise ValueError("Not available.")
# ################################
# #
# # Preprocess images
# #
# ################################
# t1_preprocessed = t1
# t1_preprocessing = None
# if do_preprocessing == True:
# t1_preprocessing = preprocess_brain_image(t1,
# truncate_intensity=(0.01, 0.99),
# brain_extraction_modality="t1",
# template="croppedMni152",
# template_transform_type="antsRegistrationSyNQuickRepro[a]",
# do_bias_correction=True,
# do_denoising=False,
# antsxnet_cache_directory=antsxnet_cache_directory,
# verbose=verbose)
# t1_preprocessed = t1_preprocessing["preprocessed_image"] * t1_preprocessing['brain_mask']
# flair_preprocessed = flair
# if do_preprocessing == True:
# flair_preprocessing = preprocess_brain_image(flair,
# truncate_intensity=(0.01, 0.99),
# brain_extraction_modality="t1",
# do_bias_correction=True,
# do_denoising=False,
# antsxnet_cache_directory=antsxnet_cache_directory,
# verbose=verbose)
# flair_preprocessed = ants.apply_transforms(fixed=t1_preprocessed,
# moving=flair_preprocessing["preprocessed_image"],
# transformlist=t1_preprocessing['template_transforms']['fwdtransforms'])
# flair_preprocessed = flair_preprocessed * t1_preprocessing['brain_mask']
# ################################
# #
# # Build model and load weights
# #
# ################################
# patch_size = (112, 112, 112)
# stride_length = (t1_preprocessed.shape[0] - patch_size[0],
# t1_preprocessed.shape[1] - patch_size[1],
# t1_preprocessed.shape[2] - patch_size[2])
# classes = ("background", "wmh" )
# number_of_classification_labels = len(classes)
# labels = (0, 1)
# image_modalities = ("T1", "FLAIR")
# channel_size = len(image_modalities)
# unet_model = create_unet_model_3d((*patch_size, channel_size),
# number_of_outputs = number_of_classification_labels,
# number_of_layers = 4, number_of_filters_at_base_layer = 16, dropout_rate = 0.0,
# convolution_kernel_size = (3, 3, 3), deconvolution_kernel_size = (2, 2, 2),
# weight_decay = 1e-5, additional_options=("attentionGating"))
# weights_file_name = get_pretrained_network("ewDavidWmhSegmentationWeights",
# antsxnet_cache_directory=antsxnet_cache_directory)
# unet_model.load_weights(weights_file_name)
# ################################
# #
# # Do prediction and normalize to native space
# #
# ################################
# if verbose == True:
# print("ew_david: prediction.")
# batchX = np.zeros((8, *patch_size, channel_size))
# t1_preprocessed = (t1_preprocessed - t1_preprocessed.mean()) / t1_preprocessed.std()
# t1_patches = extract_image_patches(t1_preprocessed, patch_size=patch_size,
# max_number_of_patches="all", stride_length=stride_length,
# return_as_array=True)
# batchX[:,:,:,:,0] = t1_patches
# flair_preprocessed = (flair_preprocessed - flair_preprocessed.mean()) / flair_preprocessed.std()
# flair_patches = extract_image_patches(flair_preprocessed, patch_size=patch_size,
# max_number_of_patches="all", stride_length=stride_length,
# return_as_array=True)
# batchX[:,:,:,:,1] = flair_patches
# predicted_data = unet_model.predict(batchX, verbose=verbose)
# probability_images = list()
# for i in range(len(labels)):
# print("Reconstructing image", classes[i])
# reconstructed_image = reconstruct_image_from_patches(predicted_data[:,:,:,:,i],
# domain_image=t1_preprocessed, stride_length=stride_length)
# if do_preprocessing == True:
# probability_images.append(ants.apply_transforms(fixed=t1,
# moving=reconstructed_image,
# transformlist=t1_preprocessing['template_transforms']['invtransforms'],
# whichtoinvert=[True], interpolator="linear", verbose=verbose))
# else:
# probability_images.append(reconstructed_image)
# return(probability_images[1])
else: # do_slicewise
################################
#
# Preprocess images
#
################################
use_rank_intensity_scaling = False
if "-ri" in which_model:
use_rank_intensity_scaling = True
t1_preprocessed = None
t1_preprocessing = None
brain_mask = None
if t1 is not None:
if do_preprocessing == True:
t1_preprocessing = preprocess_brain_image(
t1,
truncate_intensity=(0.01, 0.995),
brain_extraction_modality="t1",
do_bias_correction=False,
do_denoising=False,
antsxnet_cache_directory=antsxnet_cache_directory,
verbose=verbose)
brain_mask = ants.threshold_image(
t1_preprocessing["brain_mask"], 0.5, 1, 1, 0)
t1_preprocessed = t1_preprocessing["preprocessed_image"]
t1_segmentation = None
if use_t1_segmentation:
atropos_seg = deep_atropos(t1,
do_preprocessing=True,
verbose=verbose)
t1_segmentation = atropos_seg['segmentation_image']
flair_preprocessed = None
if flair is not None:
flair_preprocessed = flair
if do_preprocessing == True:
if brain_mask is None:
flair_preprocessing = preprocess_brain_image(
flair,
truncate_intensity=(0.01, 0.995),
brain_extraction_modality="flair",
do_bias_correction=False,
do_denoising=False,
antsxnet_cache_directory=antsxnet_cache_directory,
verbose=verbose)
brain_mask = ants.threshold_image(
flair_preprocessing["brain_mask"], 0.5, 1, 1, 0)
else:
flair_preprocessing = preprocess_brain_image(
flair,
truncate_intensity=None,
brain_extraction_modality=None,
do_bias_correction=False,
do_denoising=False,
antsxnet_cache_directory=antsxnet_cache_directory,
verbose=verbose)
flair_preprocessed = flair_preprocessing["preprocessed_image"]
if t1_preprocessed is not None:
t1_preprocessed = t1_preprocessed * brain_mask
if flair_preprocessed is not None:
flair_preprocessed = flair_preprocessed * brain_mask
if t1_preprocessed is not None:
resampling_params = list(ants.get_spacing(t1_preprocessed))
else:
resampling_params = list(ants.get_spacing(flair_preprocessed))
do_resampling = False
for d in range(len(resampling_params)):
if resampling_params[d] < 0.8:
resampling_params[d] = 1.0
do_resampling = True
resampling_params = tuple(resampling_params)
if do_resampling:
if flair_preprocessed is not None:
flair_preprocessed = ants.resample_image(flair_preprocessed,
resampling_params,
use_voxels=False,
interp_type=0)
if t1_preprocessed is not None:
t1_preprocessed = ants.resample_image(t1_preprocessed,
resampling_params,
use_voxels=False,
interp_type=0)
if t1_segmentation is not None:
t1_segmentation = ants.resample_image(t1_segmentation,
resampling_params,
use_voxels=False,
interp_type=1)
if brain_mask is not None:
brain_mask = ants.resample_image(brain_mask,
resampling_params,
use_voxels=False,
interp_type=1)
################################
#
# Build model and load weights
#
################################
template_size = (208, 208)
image_modalities = ("T1", "FLAIR")
if do_flair_only:
image_modalities = ("FLAIR", )
elif do_t1_only:
image_modalities = ("T1", )
if use_t1_segmentation:
image_modalities = (*image_modalities, "T1Seg")
channel_size = len(image_modalities)
unet_model = None
if which_model == "sysu" or which_model == "sysu-ri":
unet_model = create_unet_model_2d(
(*template_size, channel_size),
number_of_outputs=1,
mode="sigmoid",
number_of_filters=(64, 96, 128, 256, 512),
dropout_rate=0.0,
convolution_kernel_size=(3, 3),
deconvolution_kernel_size=(2, 2),
weight_decay=0,
additional_options=("initialConvolutionKernelSize[5]", ))
elif which_model == "sysuWithAttention":
unet_model = create_unet_model_2d(
(*template_size, channel_size),
number_of_outputs=1,
mode="sigmoid",
number_of_filters=(64, 96, 128, 256, 512),
dropout_rate=0.0,
convolution_kernel_size=(3, 3),
deconvolution_kernel_size=(2, 2),
weight_decay=0,
additional_options=("attentionGating",
"initialConvolutionKernelSize[5]"))
elif which_model == "sysuWithAttentionAndSite":
unet_model = create_unet_model_2d(
(*template_size, channel_size),
number_of_outputs=1,
mode="sigmoid",
scalar_output_size=3,
scalar_output_activation="softmax",
number_of_filters=(64, 96, 128, 256, 512),
dropout_rate=0.0,
convolution_kernel_size=(3, 3),
deconvolution_kernel_size=(2, 2),
weight_decay=0,
additional_options=("attentionGating",
"initialConvolutionKernelSize[5]"))
elif which_model == "sysuWithSite":
unet_model = create_unet_model_2d(
(*template_size, channel_size),
number_of_outputs=1,
mode="sigmoid",
scalar_output_size=3,
scalar_output_activation="softmax",
number_of_filters=(64, 96, 128, 256, 512),
dropout_rate=0.0,
convolution_kernel_size=(3, 3),
deconvolution_kernel_size=(2, 2),
weight_decay=0,
additional_options=("initialConvolutionKernelSize[5]", ))
elif which_model == "sysuPlusSegWithSite":
unet_model = create_unet_model_2d(
(*template_size, channel_size),
number_of_outputs=1,
mode="sigmoid",
scalar_output_size=3,
scalar_output_activation="softmax",
number_of_filters=(64, 96, 128, 256, 512),
dropout_rate=0.0,
convolution_kernel_size=(3, 3),
deconvolution_kernel_size=(2, 2),
weight_decay=0,
additional_options=("nnUnetActivationStyle", "attentionGating",
"initialConvolutionKernelSize[5]"))
else:
unet_model = create_unet_model_2d(
(*template_size, channel_size),
number_of_outputs=1,
mode="sigmoid",
number_of_filters=(64, 96, 128, 256, 512),
dropout_rate=0.0,
convolution_kernel_size=(3, 3),
deconvolution_kernel_size=(2, 2),
weight_decay=0,
additional_options=("nnUnetActivationStyle", "attentionGating",
"initialConvolutionKernelSize[5]"))
if verbose == True:
print("ewDavid: retrieving model weights.")
weights_file_name = None
if which_model == "sysu" and flair is not None and t1 is not None:
weights_file_name = get_pretrained_network(
"ewDavidSysu",
antsxnet_cache_directory=antsxnet_cache_directory)
elif which_model == "sysu-ri" and flair is not None and t1 is not None:
weights_file_name = get_pretrained_network(
"ewDavidSysuRankedIntensity",
antsxnet_cache_directory=antsxnet_cache_directory)
elif which_model == "sysu" and flair is None and t1 is not None:
weights_file_name = get_pretrained_network(
"ewDavidSysuT1Only",
antsxnet_cache_directory=antsxnet_cache_directory)
elif which_model == "sysu" and flair is not None and t1 is None:
weights_file_name = get_pretrained_network(
"ewDavidSysuFlairOnly",
antsxnet_cache_directory=antsxnet_cache_directory)
elif which_model == "sysuWithAttention" and flair is not None and t1 is not None:
weights_file_name = get_pretrained_network(
"ewDavidSysuWithAttention",
antsxnet_cache_directory=antsxnet_cache_directory)
elif which_model == "sysuWithAttention" and flair is None and t1 is not None:
weights_file_name = get_pretrained_network(
"ewDavidSysuWithAttentionT1Only",
antsxnet_cache_directory=antsxnet_cache_directory)
elif which_model == "sysuWithAttention" and flair is not None and t1 is None:
weights_file_name = get_pretrained_network(
"ewDavidSysuWithAttentionFlairOnly",
antsxnet_cache_directory=antsxnet_cache_directory)
elif which_model == "sysuWithAttentionAndSite" and flair is not None and t1 is not None:
weights_file_name = get_pretrained_network(
"ewDavidSysuWithAttentionAndSite",
antsxnet_cache_directory=antsxnet_cache_directory)
elif which_model == "sysuWithAttentionAndSite" and flair is None and t1 is not None:
weights_file_name = get_pretrained_network(
"ewDavidSysuWithAttentionAndSiteT1Only",
antsxnet_cache_directory=antsxnet_cache_directory)
elif which_model == "sysuWithAttentionAndSite" and flair is not None and t1 is None:
weights_file_name = get_pretrained_network(
"ewDavidSysuWithAttentionAndSiteFlairOnly",
antsxnet_cache_directory=antsxnet_cache_directory)
elif which_model == "sysuPlus" and flair is not None and t1 is not None:
weights_file_name = get_pretrained_network(
"ewDavidSysuPlus",
antsxnet_cache_directory=antsxnet_cache_directory)
elif which_model == "sysuPlus" and flair is None and t1 is not None:
weights_file_name = get_pretrained_network(
"ewDavidSysuPlusT1Only",
antsxnet_cache_directory=antsxnet_cache_directory)
elif which_model == "sysuPlus" and flair is not None and t1 is None:
weights_file_name = get_pretrained_network(
"ewDavidSysuPlusFlairOnly",
antsxnet_cache_directory=antsxnet_cache_directory)
elif which_model == "sysuPlusSeg" and flair is not None and t1 is not None:
weights_file_name = get_pretrained_network(
"ewDavidSysuPlusSeg",
antsxnet_cache_directory=antsxnet_cache_directory)
elif which_model == "sysuPlusSeg" and flair is None and t1 is not None:
weights_file_name = get_pretrained_network(
"ewDavidSysuPlusSegT1Only",
antsxnet_cache_directory=antsxnet_cache_directory)
elif which_model == "sysuPlusSegWithSite" and flair is not None and t1 is not None:
weights_file_name = get_pretrained_network(
"ewDavidSysuPlusSegWithSite",
antsxnet_cache_directory=antsxnet_cache_directory)
elif which_model == "sysuPlusSegWithSite" and flair is None and t1 is not None:
weights_file_name = get_pretrained_network(
"ewDavidSysuPlusSegWithSiteT1Only",
antsxnet_cache_directory=antsxnet_cache_directory)
elif which_model == "sysuWithSite" and flair is not None and t1 is not None:
weights_file_name = get_pretrained_network(
"ewDavidSysuWithSite",
antsxnet_cache_directory=antsxnet_cache_directory)
elif which_model == "sysuWithSite" and flair is None and t1 is not None:
weights_file_name = get_pretrained_network(
"ewDavidSysuWithSiteT1Only",
antsxnet_cache_directory=antsxnet_cache_directory)
elif which_model == "sysuWithSite" and flair is not None and t1 is None:
weights_file_name = get_pretrained_network(
"ewDavidSysuWithSiteFlairOnly",
antsxnet_cache_directory=antsxnet_cache_directory)
else:
raise ValueError(
"Incorrect model specification or image combination.")
unet_model.load_weights(weights_file_name)
################################
#
# Data augmentation and extract slices
#
################################
wmh_probability_image = None
if t1 is not None:
wmh_probability_image = ants.image_clone(t1_preprocessed) * 0
else:
wmh_probability_image = ants.image_clone(flair_preprocessed) * 0
wmh_site = np.array([0, 0, 0])
data_augmentation = None
if number_of_simulations > 0:
if do_flair_only:
data_augmentation = randomly_transform_image_data(
reference_image=flair_preprocessed,
input_image_list=[[flair_preprocessed]],
number_of_simulations=number_of_simulations,
transform_type='affine',
sd_affine=sd_affine,
input_image_interpolator='linear')
elif do_t1_only:
if use_t1_segmentation:
data_augmentation = randomly_transform_image_data(
reference_image=t1_preprocessed,
input_image_list=[[t1_preprocessed]],
segmentation_image_list=[t1_segmentation],
number_of_simulations=number_of_simulations,
transform_type='affine',
sd_affine=sd_affine,
input_image_interpolator='linear',
segmentation_image_interpolator='nearestNeighbor')
else:
data_augmentation = randomly_transform_image_data(
reference_image=t1_preprocessed,
input_image_list=[[t1_preprocessed]],
number_of_simulations=number_of_simulations,
transform_type='affine',
sd_affine=sd_affine,
input_image_interpolator='linear')
else:
if use_t1_segmentation:
data_augmentation = randomly_transform_image_data(
reference_image=t1_preprocessed,
input_image_list=[[
flair_preprocessed, t1_preprocessed
]],
segmentation_image_list=[t1_segmentation],
number_of_simulations=number_of_simulations,
transform_type='affine',
sd_affine=sd_affine,
input_image_interpolator='linear',
segmentation_image_interpolator='nearestNeighbor')
else:
data_augmentation = randomly_transform_image_data(
reference_image=t1_preprocessed,
input_image_list=[[
flair_preprocessed, t1_preprocessed
]],
number_of_simulations=number_of_simulations,
transform_type='affine',
sd_affine=sd_affine,
input_image_interpolator='linear')
dimensions_to_predict = list((0, ))
if which_axes == "max":
spacing = ants.get_spacing(wmh_probability_image)
dimensions_to_predict = (spacing.index(max(spacing)), )
elif which_axes == "all":
dimensions_to_predict = list(range(3))
else:
if isinstance(which_axes, int):
dimensions_to_predict = list((which_axes, ))
else:
dimensions_to_predict = list(which_axes)
total_number_of_slices = 0
for d in range(len(dimensions_to_predict)):
total_number_of_slices += wmh_probability_image.shape[
dimensions_to_predict[d]]
batchX = np.zeros(
(total_number_of_slices, *template_size, channel_size))
for n in range(number_of_simulations + 1):
batch_flair = flair_preprocessed
batch_t1 = t1_preprocessed
batch_t1_segmentation = t1_segmentation
batch_brain_mask = brain_mask
if n > 0:
if do_flair_only:
batch_flair = data_augmentation['simulated_images'][n -
1][0]
batch_brain_mask = ants.apply_ants_transform_to_image(
data_augmentation['simulated_transforms'][n - 1],
brain_mask,
flair_preprocessed,
interpolation="nearestneighbor")
elif do_t1_only:
batch_t1 = data_augmentation['simulated_images'][n - 1][0]
batch_brain_mask = ants.apply_ants_transform_to_image(
data_augmentation['simulated_transforms'][n - 1],
brain_mask,
t1_preprocessed,
interpolation="nearestneighbor")
else:
batch_flair = data_augmentation['simulated_images'][n -
1][0]
batch_t1 = data_augmentation['simulated_images'][n - 1][1]
batch_brain_mask = ants.apply_ants_transform_to_image(
data_augmentation['simulated_transforms'][n - 1],
brain_mask,
flair_preprocessed,
interpolation="nearestneighbor")
if use_t1_segmentation:
batch_t1_segmentation = data_augmentation[
'simulated_segmentation_images'][n - 1]
if use_rank_intensity_scaling:
if batch_t1 is not None:
batch_t1 = ants.rank_intensity(batch_t1,
batch_brain_mask) - 0.5
if batch_flair is not None:
batch_flair = ants.rank_intensity(flair_preprocessed,
batch_brain_mask) - 0.5
else:
if batch_t1 is not None:
batch_t1 = (batch_t1 -
batch_t1[batch_brain_mask == 1].mean()
) / batch_t1[batch_brain_mask == 1].std()
if batch_flair is not None:
batch_flair = (
batch_flair -
batch_flair[batch_brain_mask == 1].mean()
) / batch_flair[batch_brain_mask == 1].std()
slice_count = 0
for d in range(len(dimensions_to_predict)):
number_of_slices = None
if batch_t1 is not None:
number_of_slices = batch_t1.shape[dimensions_to_predict[d]]
else:
number_of_slices = batch_flair.shape[
dimensions_to_predict[d]]
if verbose == True:
print("Extracting slices for dimension ",
dimensions_to_predict[d])
for i in range(number_of_slices):
brain_mask_slice = pad_or_crop_image_to_size(
ants.slice_image(batch_brain_mask,
dimensions_to_predict[d], i),
template_size)
channel_count = 0
if batch_flair is not None:
flair_slice = pad_or_crop_image_to_size(
ants.slice_image(batch_flair,
dimensions_to_predict[d], i),
template_size)
flair_slice[brain_mask_slice == 0] = 0
batchX[slice_count, :, :,
channel_count] = flair_slice.numpy()
channel_count += 1
if batch_t1 is not None:
t1_slice = pad_or_crop_image_to_size(
ants.slice_image(batch_t1,
dimensions_to_predict[d], i),
template_size)
t1_slice[brain_mask_slice == 0] = 0
batchX[slice_count, :, :,
channel_count] = t1_slice.numpy()
channel_count += 1
if t1_segmentation is not None:
t1_segmentation_slice = pad_or_crop_image_to_size(
ants.slice_image(batch_t1_segmentation,
dimensions_to_predict[d], i),
template_size)
t1_segmentation_slice[brain_mask_slice == 0] = 0
batchX[slice_count, :, :,
channel_count] = t1_segmentation_slice.numpy(
) / 6 - 0.5
slice_count += 1
################################
#
# Do prediction and then restack into the image
#
################################
if verbose == True:
if n == 0:
print("Prediction")
else:
print("Prediction (simulation " + str(n) + ")")
prediction = unet_model.predict(batchX, verbose=verbose)
permutations = list()
permutations.append((0, 1, 2))
permutations.append((1, 0, 2))
permutations.append((1, 2, 0))
prediction_image_average = ants.image_clone(
wmh_probability_image) * 0
current_start_slice = 0
for d in range(len(dimensions_to_predict)):
current_end_slice = current_start_slice + wmh_probability_image.shape[
dimensions_to_predict[d]]
which_batch_slices = range(current_start_slice,
current_end_slice)
if isinstance(prediction, list):
prediction_per_dimension = prediction[0][
which_batch_slices, :, :, 0]
else:
prediction_per_dimension = prediction[
which_batch_slices, :, :, 0]
prediction_array = np.transpose(
np.squeeze(prediction_per_dimension),
permutations[dimensions_to_predict[d]])
prediction_image = ants.copy_image_info(
wmh_probability_image,
pad_or_crop_image_to_size(
ants.from_numpy(prediction_array),
wmh_probability_image.shape))
prediction_image_average = prediction_image_average + (
prediction_image - prediction_image_average) / (d + 1)
current_start_slice = current_end_slice
wmh_probability_image = wmh_probability_image + (
prediction_image_average - wmh_probability_image) / (n + 1)
if isinstance(prediction, list):
wmh_site = wmh_site + (np.mean(prediction[1], axis=0) -
wmh_site) / (n + 1)
if do_resampling:
if t1 is not None:
wmh_probability_image = ants.resample_image_to_target(
wmh_probability_image, t1)
if flair is not None:
wmh_probability_image = ants.resample_image_to_target(
wmh_probability_image, flair)
if isinstance(prediction, list):
return ([wmh_probability_image, wmh_site])
else:
return (wmh_probability_image)
def brain_extraction(image,
modality="t1",
antsxnet_cache_directory=None,
verbose=False):
"""
Perform brain extraction using U-net and ANTs-based training data. "NoBrainer"
is also possible where brain extraction uses U-net and FreeSurfer training data
ported from the
https://github.com/neuronets/nobrainer-models
Arguments
---------
image : ANTsImage
input image (or list of images for multi-modal scenarios).
modality : string
Modality image type. Options include:
* "t1": T1-weighted MRI---ANTs-trained. Update from "t1v0".
* "t1v0": T1-weighted MRI---ANTs-trained.
* "t1nobrainer": T1-weighted MRI---FreeSurfer-trained: h/t Satra Ghosh and Jakub Kaczmarzyk.
* "t1combined": Brian's combination of "t1" and "t1nobrainer". One can also specify
"t1combined[X]" where X is the morphological radius. X = 12 by default.
* "flair": FLAIR MRI.
* "t2": T2 MRI.
* "bold": 3-D BOLD MRI.
* "fa": Fractional anisotropy.
* "t1t2infant": Combined T1-w/T2-w infant MRI h/t Martin Styner.
* "t1infant": T1-w infant MRI h/t Martin Styner.
* "t2infant": T2-w infant MRI h/t Martin Styner.
antsxnet_cache_directory : string
Destination directory for storing the downloaded template and model weights.
Since these can be resused, if is None, these data will be downloaded to a
~/.keras/ANTsXNet/.
verbose : boolean
Print progress to the screen.
Returns
-------
ANTs probability brain mask image.
Example
-------
>>> probability_brain_mask = brain_extraction(brain_image, modality="t1")
"""
from ..architectures import create_unet_model_3d
from ..utilities import get_pretrained_network
from ..utilities import get_antsxnet_data
from ..architectures import create_nobrainer_unet_model_3d
classes = ("background", "brain")
number_of_classification_labels = len(classes)
channel_size = 1
if isinstance(image, list):
channel_size = len(image)
if antsxnet_cache_directory == None:
antsxnet_cache_directory = "ANTsXNet"
input_images = list()
if channel_size == 1:
input_images.append(image)
else:
input_images = image
if input_images[0].dimension != 3:
raise ValueError("Image dimension must be 3.")
if "t1combined" in modality:
brain_extraction_t1 = brain_extraction(
image,
modality="t1",
antsxnet_cache_directory=antsxnet_cache_directory,
verbose=verbose)
brain_mask = ants.iMath_get_largest_component(
ants.threshold_image(brain_extraction_t1, 0.5, 10000))
# Need to change with voxel resolution
morphological_radius = 12
if '[' in modality and ']' in modality:
morphological_radius = int(modality.split("[")[1].split("]")[0])
brain_extraction_t1nobrainer = brain_extraction(
image * ants.iMath_MD(brain_mask, radius=morphological_radius),
modality="t1nobrainer",
antsxnet_cache_directory=antsxnet_cache_directory,
verbose=verbose)
brain_extraction_combined = ants.iMath_fill_holes(
ants.iMath_get_largest_component(brain_extraction_t1nobrainer *
brain_mask))
brain_extraction_combined = brain_extraction_combined + ants.iMath_ME(
brain_mask, morphological_radius) + brain_mask
return (brain_extraction_combined)
if modality != "t1nobrainer":
#####################
#
# ANTs-based
#
#####################
weights_file_name_prefix = None
if modality == "t1v0":
weights_file_name_prefix = "brainExtraction"
elif modality == "t1":
weights_file_name_prefix = "brainExtractionT1"
elif modality == "t2":
weights_file_name_prefix = "brainExtractionT2"
elif modality == "flair":
weights_file_name_prefix = "brainExtractionFLAIR"
elif modality == "bold":
weights_file_name_prefix = "brainExtractionBOLD"
elif modality == "fa":
weights_file_name_prefix = "brainExtractionFA"
elif modality == "t1t2infant":
weights_file_name_prefix = "brainExtractionInfantT1T2"
elif modality == "t1infant":
weights_file_name_prefix = "brainExtractionInfantT1"
elif modality == "t2infant":
weights_file_name_prefix = "brainExtractionInfantT2"
else:
raise ValueError("Unknown modality type.")
weights_file_name = get_pretrained_network(
weights_file_name_prefix,
antsxnet_cache_directory=antsxnet_cache_directory)
if verbose == True:
print("Brain extraction: retrieving template.")
reorient_template_file_name_path = get_antsxnet_data(
"S_template3", antsxnet_cache_directory=antsxnet_cache_directory)
reorient_template = ants.image_read(reorient_template_file_name_path)
resampled_image_size = reorient_template.shape
if modality == "t1":
classes = ("background", "head", "brain")
number_of_classification_labels = len(classes)
unet_model = create_unet_model_3d(
(*resampled_image_size, channel_size),
number_of_outputs=number_of_classification_labels,
number_of_layers=4,
number_of_filters_at_base_layer=8,
dropout_rate=0.0,
convolution_kernel_size=(3, 3, 3),
deconvolution_kernel_size=(2, 2, 2),
weight_decay=1e-5)
unet_model.load_weights(weights_file_name)
if verbose == True:
print("Brain extraction: normalizing image to the template.")
center_of_mass_template = ants.get_center_of_mass(reorient_template)
center_of_mass_image = ants.get_center_of_mass(input_images[0])
translation = np.asarray(center_of_mass_image) - np.asarray(
center_of_mass_template)
xfrm = ants.create_ants_transform(
transform_type="Euler3DTransform",
center=np.asarray(center_of_mass_template),
translation=translation)
batchX = np.zeros((1, *resampled_image_size, channel_size))
for i in range(len(input_images)):
warped_image = ants.apply_ants_transform_to_image(
xfrm, input_images[i], reorient_template)
warped_array = warped_image.numpy()
batchX[0, :, :, :, i] = (warped_array -
warped_array.mean()) / warped_array.std()
if verbose == True:
print("Brain extraction: prediction and decoding.")
predicted_data = unet_model.predict(batchX, verbose=verbose)
origin = reorient_template.origin
spacing = reorient_template.spacing
direction = reorient_template.direction
probability_images_array = list()
probability_images_array.append(
ants.from_numpy(np.squeeze(predicted_data[0, :, :, :, 0]),
origin=origin,
spacing=spacing,
direction=direction))
probability_images_array.append(
ants.from_numpy(np.squeeze(predicted_data[0, :, :, :, 1]),
origin=origin,
spacing=spacing,
direction=direction))
if modality == "t1":
probability_images_array.append(
ants.from_numpy(np.squeeze(predicted_data[0, :, :, :, 2]),
origin=origin,
spacing=spacing,
direction=direction))
if verbose == True:
print(
"Brain extraction: renormalize probability mask to native space."
)
probability_image = ants.apply_ants_transform_to_image(
ants.invert_ants_transform(xfrm),
probability_images_array[number_of_classification_labels - 1],
input_images[0])
return (probability_image)
else:
#####################
#
# NoBrainer
#
#####################
if verbose == True:
print("NoBrainer: generating network.")
model = create_nobrainer_unet_model_3d((None, None, None, 1))
weights_file_name = get_pretrained_network(
"brainExtractionNoBrainer",
antsxnet_cache_directory=antsxnet_cache_directory)
model.load_weights(weights_file_name)
if verbose == True:
print(
"NoBrainer: preprocessing (intensity truncation and resampling)."
)
image_array = image.numpy()
image_robust_range = np.quantile(
image_array[np.where(image_array != 0)], (0.02, 0.98))
threshold_value = 0.10 * (image_robust_range[1] - image_robust_range[0]
) + image_robust_range[0]
thresholded_mask = ants.threshold_image(image, -10000, threshold_value,
0, 1)
thresholded_image = image * thresholded_mask
image_resampled = ants.resample_image(thresholded_image,
(256, 256, 256),
use_voxels=True)
image_array = np.expand_dims(image_resampled.numpy(), axis=0)
image_array = np.expand_dims(image_array, axis=-1)
if verbose == True:
print("NoBrainer: predicting mask.")
brain_mask_array = np.squeeze(
model.predict(image_array, verbose=verbose))
brain_mask_resampled = ants.copy_image_info(
image_resampled, ants.from_numpy(brain_mask_array))
brain_mask_image = ants.resample_image(brain_mask_resampled,
image.shape,
use_voxels=True,
interp_type=1)
spacing = ants.get_spacing(image)
spacing_product = spacing[0] * spacing[1] * spacing[2]
minimum_brain_volume = round(649933.7 / spacing_product)
brain_mask_labeled = ants.label_clusters(brain_mask_image,
minimum_brain_volume)
return (brain_mask_labeled)