def parcellate_voronoi_vol(mask, nb_parcels, seeds=None):
"""
Produce a parcellation from a Voronoi diagram built on random seeds.
The number of seeds is equal to the nb of parcels.
Seed are randomly placed within the mask, expect on edge positions
Args:
- mask (numpy.ndarray): binary 3D array of valid position to parcellate
- nb_parcels (int): the required number of parcels
- seeds: TODO
Return:
- the parcellation (numpy.ndarray): a 3D array of integers
-
"""
parcellation = np.zeros(mask.shape, dtype=int)
nvox = (mask !=0).sum()
for cc_mask in mg.split_mask_into_cc_iter(mask!=0):
# compute the required number of parcels within the current CC:
size_cc = cc_mask.sum()
cc_np = max(int(np.round(nb_parcels * size_cc / (nvox*1.))), 1)
pyhrf.verbose(2, 'Treating a connected component (CC) of %d positions' \
%cc_mask.sum())
if cc_mask.sum() < 6:
continue
if seeds is None:
# perform voronoi on random seeds
eroded_mask = peelVolume3D(cc_mask)
eroded_mask_size = eroded_mask.sum()
if eroded_mask_size < nb_parcels: #do no erode, mask too small
eroded_mask_size = nvox
eroded_mask = mask.copy()
cc_seeds = np.random.randint(0,eroded_mask_size, cc_np)
mask_for_seed = np.zeros(eroded_mask_size, dtype=int)
mask_for_seed[cc_seeds] = 1
mask_for_seed = expand_array_in_mask(mask_for_seed, eroded_mask)
else:
mask_for_seed = seeds * cc_mask
pyhrf.verbose(2, 'Nb of seeds in current CC: %d' \
%mask_for_seed.sum())
cc_parcellation = voronoi(np.vstack(np.where(cc_mask)).T,
np.vstack(np.where(mask_for_seed)).T) + 1
pyhrf.verbose(3, 'CC parcellation labels: %s' \
%str(np.unique(cc_parcellation)))
maxp = parcellation.max()
parcellation += expand_array_in_mask(cc_parcellation + maxp, cc_mask)
pyhrf.verbose(3, 'Current parcellation labels: %s' \
%str(np.unique(parcellation)))
pyhrf.verbose(1, 'voronoi parcellation: %s, %s' \
%(str(parcellation.shape), str(parcellation.dtype)))
return parcellation
def parcellate_voronoi_vol(mask, nb_parcels, seeds=None):
"""
Produce a parcellation from a Voronoi diagram built on random seeds.
The number of seeds is equal to the nb of parcels.
Seed are randomly placed within the mask, expect on edge positions
Args:
- mask (numpy.ndarray): binary 3D array of valid position to parcellate
- nb_parcels (int): the required number of parcels
- seeds: TODO
Return:
- the parcellation (numpy.ndarray): a 3D array of integers
-
"""
parcellation = np.zeros(mask.shape, dtype=int)
nvox = (mask != 0).sum()
for cc_mask in mg.split_mask_into_cc_iter(mask != 0):
# compute the required number of parcels within the current CC:
size_cc = cc_mask.sum()
cc_np = max(int(np.round(nb_parcels * size_cc / (nvox * 1.))), 1)
logger.info('Treating a connected component (CC) of %d positions',
cc_mask.sum())
if cc_mask.sum() < 6:
continue
if seeds is None:
# perform voronoi on random seeds
eroded_mask = peelVolume3D(cc_mask)
eroded_mask_size = eroded_mask.sum()
if eroded_mask_size < nb_parcels: # do no erode, mask too small
eroded_mask_size = nvox
eroded_mask = mask.copy()
cc_seeds = np.random.randint(0, eroded_mask_size, cc_np)
mask_for_seed = np.zeros(eroded_mask_size, dtype=int)
mask_for_seed[cc_seeds] = 1
mask_for_seed = expand_array_in_mask(mask_for_seed, eroded_mask)
else:
mask_for_seed = seeds * cc_mask
logger.info('Nb of seeds in current CC: %d', mask_for_seed.sum())
cc_parcellation = voronoi(
np.vstack(np.where(cc_mask)).T,
np.vstack(np.where(mask_for_seed)).T) + 1
logger.info('CC parcellation labels: %s',
str(np.unique(cc_parcellation)))
maxp = parcellation.max()
parcellation += expand_array_in_mask(cc_parcellation + maxp, cc_mask)
logger.info('Current parcellation labels: %s',
str(np.unique(parcellation)))
logger.info('voronoi parcellation: %s, %s', str(parcellation.shape),
str(parcellation.dtype))
return parcellation
def parcellate_balanced_vol(mask, nb_parcels):
"""
Performs a balanced partitioning on the input mask using a balloon patroling
algorithm [Eurol 2009]. Values with 0 are discarded position in the mask.
Args:
- mask (numpy.ndarray): binary 3D array of valid position to parcellate
- nb_parcels (int): the required number of parcels
Return:
- the parcellation (numpy.ndarray): a 3D array of integers
"""
parcellation = np.zeros(mask.shape, dtype=int)
nvox = (mask != 0).sum()
# Iterate over connected components in the input mask
for cc_mask in mg.split_mask_into_cc_iter(mask != 0):
logger.info('Treating a connected component (CC) of %d positions',
cc_mask.sum())
g = mg.graph_from_lattice(cc_mask)
size_cc = cc_mask.sum()
# compute the required number of parcels within the current CC:
cc_np = max(int(np.round(nb_parcels * size_cc / (nvox * 1.))), 1)
logger.info('Split (CC) into %d parcels', cc_np)
cc_labels = np.ones(cc_mask.sum(), dtype=int)
if cc_np > 1:
split_parcel(cc_labels, {1: g},
1,
cc_np,
inplace=True,
verbosity=2,
balance_tolerance='draft')
else: # only one parcel expected, CC must be too small to be splited
cc_labels[:] = 1
logger.info('Split done!')
# accumulate parcellation result
maxp = parcellation.max()
parcellation += expand_array_in_mask(cc_labels + maxp, cc_mask > 0)
return parcellation
def parcellate_balanced_vol(mask, nb_parcels):
"""
Performs a balanced partitioning on the input mask using a balloon patroling
algorithm [Eurol 2009]. Values with 0 are discarded position in the mask.
Args:
- mask (numpy.ndarray): binary 3D array of valid position to parcellate
- nb_parcels (int): the required number of parcels
Return:
- the parcellation (numpy.ndarray): a 3D array of integers
"""
parcellation = np.zeros(mask.shape, dtype=int)
nvox = (mask != 0).sum()
# Iterate over connected components in the input mask
for cc_mask in mg.split_mask_into_cc_iter(mask != 0):
pyhrf.verbose(2, 'Treating a connected component (CC) of %d positions' \
%cc_mask.sum())
g = mg.graph_from_lattice(cc_mask)
size_cc = cc_mask.sum()
# compute the required number of parcels within the current CC:
cc_np = max(int(np.round(nb_parcels * size_cc / (nvox*1.))), 1)
pyhrf.verbose(2, 'Split (CC) into %d parcels' %cc_np)
cc_labels = np.ones(cc_mask.sum(), dtype=int)
if cc_np > 1:
split_parcel(cc_labels, {1:g}, 1, cc_np, inplace=True,
verbosity=2, balance_tolerance='draft')
else: #only one parcel expected, CC must be too small to be splited
cc_labels[:] = 1
pyhrf.verbose(2, 'Split done!')
# accumulate parcellation result
maxp = parcellation.max()
parcellation += expand_array_in_mask(cc_labels + maxp, cc_mask>0)
return parcellation
