def yeo_to_91k(dlabel, medial_wall, reference, out):
"""Convert Yeo-style dlabels (Yeo and Schaefer parcellations) to 91k
grayordinate space
The Yeo lab generates dlabel's inclusive of medial wall vertices and only
for the cortical surfaces. This is different from how typical dlabels are
formatted, which exclude medial wall vertices and include voxels from all
subcortical and cerebellar structures (i.e. the full 91k grayordinate
space). This function corrects Yeo dlabels to proper 91k grayordinates.
Parameters
----------
dlabel : str
A Yeo-style .dlabel.nii atlas
medial_wall : str
HCP medial wall mask (.dlabel.nii)
reference : str
A reference .dlabel.nii file with 91k grayordinates and all brain
models included
out : str
Output 91k grayordinate .dlabel.nii file
"""
dlabel = nib.load(dlabel)
medial_wall = nib.load(medial_wall)
ref = nib.load(reference)
# remove medial wall vertices
array = dlabel.get_fdata()
corrected_array = array[np.logical_not(medial_wall.get_fdata())]
# expand to 91k
grayordinates = np.zeros(ref.shape)
grayordinates[0, :corrected_array.shape[0]] = corrected_array
# make header
labels = dlabel.header.get_axis(index=0).label[0]
label_table = ci.Cifti2LabelTable()
for key, (tag, rgba) in labels.items():
label_table[key] = ci.Cifti2Label(key, tag, *rgba)
maps = [ci.Cifti2NamedMap('labels', ci.Cifti2MetaData({}), label_table)]
label_map = ci.Cifti2MatrixIndicesMap(
applies_to_matrix_dimension=(0, ),
indices_map_to_data_type='CIFTI_INDEX_TYPE_LABELS',
maps=maps)
model_map = ci.Cifti2MatrixIndicesMap(
applies_to_matrix_dimension=(1, ),
indices_map_to_data_type='CIFTI_INDEX_TYPE_BRAIN_MODELS',
maps=list(ref.header.get_index_map(1).brain_models))
model_map.volume = ref.header.get_index_map(1).volume
matrix = ci.Cifti2Matrix()
matrix.append(label_map)
matrix.append(model_map)
hdr = ci.Cifti2Header(matrix)
out_dtseries = ci.Cifti2Image(grayordinates, hdr)
out_dtseries.to_filename(out)
return out
def create_scalar_map(applies_to_matrix_dimension):
maps = [
ci.Cifti2NamedMap(name, ci.Cifti2MetaData(meta))
for name, meta in scalars
]
return ci.Cifti2MatrixIndicesMap(applies_to_matrix_dimension,
'CIFTI_INDEX_TYPE_SCALARS',
maps=maps)
def test_cifti2_metadata():
md = ci.Cifti2MetaData(metadata={'a': 'aval'})
assert len(md) == 1
assert list(iter(md)) == ['a']
assert md['a'] == 'aval'
assert md.data == dict([('a', 'aval')])
md = ci.Cifti2MetaData()
assert len(md) == 0
assert list(iter(md)) == []
assert md.data == {}
with pytest.raises(ValueError):
md.difference_update(None)
md['a'] = 'aval'
assert md['a'] == 'aval'
assert len(md) == 1
assert md.data == dict([('a', 'aval')])
del md['a']
assert len(md) == 0
metadata_test = [('a', 'aval'), ('b', 'bval')]
md.update(metadata_test)
assert md.data == dict(metadata_test)
assert list(iter(md)) == list(iter(collections.OrderedDict(metadata_test)))
md.update({'a': 'aval', 'b': 'bval'})
assert md.data == dict(metadata_test)
md.update({'a': 'aval', 'd': 'dval'})
assert md.data == dict(metadata_test + [('d', 'dval')])
md.difference_update({'a': 'aval', 'd': 'dval'})
assert md.data == dict(metadata_test[1:])
with pytest.raises(KeyError):
md.difference_update({'a': 'aval', 'd': 'dval'})
assert md.to_xml().decode(
'utf-8'
) == '<MetaData><MD><Name>b</Name><Value>bval</Value></MD></MetaData>'
def create_label_map(applies_to_matrix_dimension):
maps = []
for name, meta, label in labels:
label_table = ci.Cifti2LabelTable()
for key, (tag, rgba) in label.items():
label_table[key] = ci.Cifti2Label(key, tag, *rgba)
maps.append(
ci.Cifti2NamedMap(name, ci.Cifti2MetaData(meta), label_table))
return ci.Cifti2MatrixIndicesMap(applies_to_matrix_dimension,
'CIFTI_INDEX_TYPE_LABELS',
maps=maps)
def test_cifti2_metadata():
md = ci.Cifti2MetaData(metadata={'a': 'aval'})
assert_equal(len(md), 1)
assert_equal(list(iter(md)), ['a'])
assert_equal(md['a'], 'aval')
assert_equal(md.data, dict([('a', 'aval')]))
md = ci.Cifti2MetaData()
assert_equal(len(md), 0)
assert_equal(list(iter(md)), [])
assert_equal(md.data, {})
assert_raises(ValueError, md.difference_update, None)
md['a'] = 'aval'
assert_equal(md['a'], 'aval')
assert_equal(len(md), 1)
assert_equal(md.data, dict([('a', 'aval')]))
del md['a']
assert_equal(len(md), 0)
metadata_test = [('a', 'aval'), ('b', 'bval')]
md.update(metadata_test)
assert_equal(md.data, dict(metadata_test))
assert_equal(list(iter(md)),
list(iter(collections.OrderedDict(metadata_test))))
md.update({'a': 'aval', 'b': 'bval'})
assert_equal(md.data, dict(metadata_test))
md.update({'a': 'aval', 'd': 'dval'})
assert_equal(md.data, dict(metadata_test + [('d', 'dval')]))
md.difference_update({'a': 'aval', 'd': 'dval'})
assert_equal(md.data, dict(metadata_test[1:]))
assert_raises(KeyError, md.difference_update, {'a': 'aval', 'd': 'dval'})
assert_equal(
md.to_xml().decode('utf-8'),
'<MetaData><MD><Name>b</Name><Value>bval</Value></MD></MetaData>')
def to_mapping(self, dim):
"""
Converts the hcp_labels to a MatrixIndicesMap for storage in CIFTI format
Parameters
----------
dim : int
which dimension of the CIFTI vector/matrix is described by this dataset (zero-based)
Returns
-------
cifti2.Cifti2MatrixIndicesMap
"""
mim = cifti2.Cifti2MatrixIndicesMap([dim], 'CIFTI_INDEX_TYPE_SCALARS')
for elem in self.arr:
meta = None if len(elem['meta']) == 0 else elem['meta']
named_map = cifti2.Cifti2NamedMap(elem['name'], cifti2.Cifti2MetaData(meta))
mim.append(named_map)
return mim
def to_mapping(self, dim):
"""
Converts the hcp_labels to a MatrixIndicesMap for storage in CIFTI format
Parameters
----------
dim : int
which dimension of the CIFTI vector/matrix is described by this dataset (zero-based)
Returns
-------
cifti2.Cifti2MatrixIndicesMap
"""
mim = cifti2.Cifti2MatrixIndicesMap([dim], 'CIFTI_INDEX_TYPE_LABELS')
for elem in self.arr:
label_table = cifti2.Cifti2LabelTable()
for key, value in elem['label'].items():
label_table[key] = (value[0],) + tuple(value[1])
meta = None if len(elem['meta']) == 0 else elem['meta']
named_map = cifti2.Cifti2NamedMap(elem['name'], cifti2.Cifti2MetaData(meta),
label_table)
mim.append(named_map)
return mim
def _create_cifti_image(bold_file, label_file, annotation_files, gii_files,
volume_target, surface_target, tr):
"""
Generate CIFTI image in target space
Parameters
bold_file : 4D BOLD timeseries
label_file : label atlas
annotation_files : FreeSurfer annotations
gii_files : 4D BOLD surface timeseries in GIFTI format
volume_target : label atlas space
surface_target : gii_files space
tr : repetition timeseries
Returns
out_file : BOLD data as CIFTI dtseries
"""
label_img = nb.load(label_file)
bold_img = resample_to_img(bold_file, label_img)
bold_data = bold_img.get_data()
timepoints = bold_img.shape[3]
label_data = label_img.get_data()
# set up CIFTI information
series_map = ci.Cifti2MatrixIndicesMap(
(0, ),
'CIFTI_INDEX_TYPE_SERIES',
number_of_series_points=timepoints,
series_exponent=0,
series_start=0.0,
series_step=tr,
series_unit='SECOND')
# Create CIFTI brain models
idx_offset = 0
brainmodels = []
bm_ts = np.empty((timepoints, 0))
for structure, labels in CIFTI_STRUCT_WITH_LABELS.items():
if labels is None: # surface model
model_type = "CIFTI_MODEL_TYPE_SURFACE"
# use the corresponding annotation
hemi = structure.split('_')[-1]
annot = nb.freesurfer.read_annot(
annotation_files[hemi == "RIGHT"])
# currently only supports L/R cortex
gii = nb.load(gii_files[hemi == "RIGHT"])
# calculate total number of vertices
surf_verts = len(annot[0])
# remove medial wall for CIFTI format
vert_idx = np.nonzero(
annot[0] != annot[2].index(b'unknown'))[0]
# extract values across volumes
ts = np.array([tsarr.data[vert_idx] for tsarr in gii.darrays])
vert_idx = ci.Cifti2VertexIndices(vert_idx)
bm = ci.Cifti2BrainModel(index_offset=idx_offset,
index_count=len(vert_idx),
model_type=model_type,
brain_structure=structure,
vertex_indices=vert_idx,
n_surface_vertices=surf_verts)
bm_ts = np.column_stack((bm_ts, ts))
idx_offset += len(vert_idx)
brainmodels.append(bm)
else:
model_type = "CIFTI_MODEL_TYPE_VOXELS"
vox = []
ts = None
for label in labels:
ijk = np.nonzero(label_data == label)
ts = (bold_data[ijk] if ts is None else np.concatenate(
(ts, bold_data[ijk])))
vox += [[ijk[0][ix], ijk[1][ix], ijk[2][ix]]
for ix, row in enumerate(ts)]
bm_ts = np.column_stack((bm_ts, ts.T))
vox = ci.Cifti2VoxelIndicesIJK(vox)
bm = ci.Cifti2BrainModel(index_offset=idx_offset,
index_count=len(vox),
model_type=model_type,
brain_structure=structure,
voxel_indices_ijk=vox)
idx_offset += len(vox)
brainmodels.append(bm)
volume = ci.Cifti2Volume(
bold_img.shape[:3],
ci.Cifti2TransformationMatrixVoxelIndicesIJKtoXYZ(
-3, bold_img.affine))
brainmodels.append(volume)
# create CIFTI geometry based on brainmodels
geometry_map = ci.Cifti2MatrixIndicesMap(
(1, ), 'CIFTI_INDEX_TYPE_BRAIN_MODELS', maps=brainmodels)
# provide some metadata to CIFTI matrix
meta = {
"target_surface": surface_target,
"target_volume": volume_target,
}
# generate and save CIFTI image
matrix = ci.Cifti2Matrix()
matrix.append(series_map)
matrix.append(geometry_map)
matrix.metadata = ci.Cifti2MetaData(meta)
hdr = ci.Cifti2Header(matrix)
img = ci.Cifti2Image(bm_ts, hdr)
img.nifti_header.set_intent('NIFTI_INTENT_CONNECTIVITY_DENSE_SERIES')
_, out_base, _ = split_filename(bold_file)
out_file = "{}.dtseries.nii".format(out_base)
ci.save(img, out_file)
return os.path.join(os.getcwd(), out_file)
def _create_cifti_image(bold_file, label_file, bold_surfs, annotation_files,
tr, targets):
"""
Generate CIFTI image in target space.
Parameters
----------
bold_file : str
BOLD volumetric timeseries
label_file : str
Subcortical label file
bold_surfs : list
BOLD surface timeseries [L,R]
annotation_files : list
Surface label files used to remove medial wall
tr : float
BOLD repetition time
targets : tuple or list
Surface and volumetric output spaces
Returns
-------
out :
BOLD data saved as CIFTI dtseries
"""
bold_img = nb.load(bold_file)
label_img = nb.load(label_file)
if label_img.shape != bold_img.shape[:3]:
warnings.warn("Resampling bold volume to match label dimensions")
bold_img = resample_to_img(bold_img, label_img)
bold_data = bold_img.get_fdata(dtype='float32')
timepoints = bold_img.shape[3]
label_data = np.asanyarray(label_img.dataobj).astype('int16')
# Create brain models
idx_offset = 0
brainmodels = []
bm_ts = np.empty((timepoints, 0))
for structure, labels in CIFTI_STRUCT_WITH_LABELS.items():
if labels is None: # surface model
model_type = "CIFTI_MODEL_TYPE_SURFACE"
# use the corresponding annotation
hemi = structure.split('_')[-1]
# currently only supports L/R cortex
surf = nb.load(bold_surfs[hemi == "RIGHT"])
surf_verts = len(surf.darrays[0].data)
if annotation_files[0].endswith('.annot'):
annot = nb.freesurfer.read_annot(
annotation_files[hemi == "RIGHT"])
# remove medial wall
medial = np.nonzero(annot[0] != annot[2].index(b'unknown'))[0]
else:
annot = nb.load(annotation_files[hemi == "RIGHT"])
medial = np.nonzero(annot.darrays[0].data)[0]
# extract values across volumes
ts = np.array([tsarr.data[medial] for tsarr in surf.darrays])
vert_idx = ci.Cifti2VertexIndices(medial)
bm = ci.Cifti2BrainModel(index_offset=idx_offset,
index_count=len(vert_idx),
model_type=model_type,
brain_structure=structure,
vertex_indices=vert_idx,
n_surface_vertices=surf_verts)
idx_offset += len(vert_idx)
bm_ts = np.column_stack((bm_ts, ts))
else:
model_type = "CIFTI_MODEL_TYPE_VOXELS"
vox = []
ts = None
for label in labels:
ijk = np.nonzero(label_data == label)
if ijk[0].size == 0: # skip label if nothing matches
continue
ts = (bold_data[ijk] if ts is None else np.concatenate(
(ts, bold_data[ijk])))
vox += [[ijk[0][ix], ijk[1][ix], ijk[2][ix]]
for ix, row in enumerate(ts)]
vox = ci.Cifti2VoxelIndicesIJK(vox)
bm = ci.Cifti2BrainModel(index_offset=idx_offset,
index_count=len(vox),
model_type=model_type,
brain_structure=structure,
voxel_indices_ijk=vox)
idx_offset += len(vox)
bm_ts = np.column_stack((bm_ts, ts.T))
# add each brain structure to list
brainmodels.append(bm)
# add volume information
brainmodels.append(
ci.Cifti2Volume(
bold_img.shape[:3],
ci.Cifti2TransformationMatrixVoxelIndicesIJKtoXYZ(
-3, bold_img.affine)))
# generate Matrix information
series_map = ci.Cifti2MatrixIndicesMap((0, ),
'CIFTI_INDEX_TYPE_SERIES',
number_of_series_points=timepoints,
series_exponent=0,
series_start=0.0,
series_step=tr,
series_unit='SECOND')
geometry_map = ci.Cifti2MatrixIndicesMap((1, ),
'CIFTI_INDEX_TYPE_BRAIN_MODELS',
maps=brainmodels)
# provide some metadata to CIFTI matrix
meta = {
"surface": targets[0],
"volume": targets[1],
}
# generate and save CIFTI image
matrix = ci.Cifti2Matrix()
matrix.append(series_map)
matrix.append(geometry_map)
matrix.metadata = ci.Cifti2MetaData(meta)
hdr = ci.Cifti2Header(matrix)
img = ci.Cifti2Image(bm_ts, hdr)
img.nifti_header.set_intent('NIFTI_INTENT_CONNECTIVITY_DENSE_SERIES')
out_file = "{}.dtseries.nii".format(split_filename(bold_file)[1])
ci.save(img, out_file)
return os.path.join(os.getcwd(), out_file)
def create_scalar_map(applies_to_matrix_dimension, info):
"""Creates a scalar map form a list of NamedMapInfo"""
maps = [ci.Cifti2NamedMap(i.name, ci.Cifti2MetaData(i.meta)) for i in info]
return ci.Cifti2MatrixIndicesMap(applies_to_matrix_dimension,
Map.SCALARS,
maps=maps)
