def test_3D_segments():
points = np.array([[[1, 0, 0],
[1, 1, 0]],
[[3, 1, 0],
[3, 0, 0]],
[[2, 0, 0],
[2, 1, 0]],
[[5, 1, 0],
[5, 0, 0]],
[[5.5, 0, 0],
[5.5, 1, 0]]], dtype="f4")
thresholds = [4, 2, 1]
feature = ResampleFeature(nb_points=20)
metric = AveragePointwiseEuclideanMetric(feature)
qbx = QuickBundlesX(thresholds, metric=metric)
tree = qbx.cluster(points)
clusters_0 = tree.get_clusters(0)
clusters_1 = tree.get_clusters(1)
clusters_2 = tree.get_clusters(2)
assert_equal(len(clusters_0.centroids), len(clusters_1.centroids))
assert_equal(len(clusters_2.centroids) > len(clusters_1.centroids), True)
assert_array_equal(clusters_2[1].indices, np.array([3, 4], dtype=np.int32))
def gen_sagittal_views(show=False):
from dipy.viz import window
from dipy.viz.clustering import show_clusters
streamlines = get_data()
thresholds = [40, 30, 25]#, 20, 15]
qbx_class = QuickBundlesX(thresholds)
print "Clustering {} streamlines ({})...".format(len(streamlines), thresholds)
qbx = qbx_class.cluster(streamlines)
clusters = qbx.get_clusters(len(thresholds))
clusters.refdata = streamlines
print "Displaying {} clusters...".format(len(clusters))
tree = qbx.get_tree_cluster_map()
tree.refdata = streamlines
color_tree(tree)
# #TMP
# clusters = tree.get_clusters(len(thresholds))
# clusters.refdata = streamlines
# ren = show_clusters(clusters, show=True)
# #window.snapshot(ren, fname="sagittal_{}".format(thresholds[-1]), size=(1200, 1200))
# return
for level in range(1, len(thresholds) + 1):
print level, thresholds[level-1]
clusters = tree.get_clusters(level)
clusters.refdata = streamlines
ren = show_clusters(clusters, show=show)
ren.reset_camera_tight()
window.snapshot(ren, fname="sagittal_{}".format(thresholds[level-1]), size=(1200, 1200))
def test_circle_parallel_fornix():
circle = streamlines_in_circle(100, step_size=2)
parallel = streamlines_parallel(100)
thresholds = [1, 0.1]
qbx_class = QuickBundlesX(thresholds)
tree = qbx_class.cluster(circle)
clusters = tree.get_clusters(0)
assert_equal(len(clusters), 1)
clusters = tree.get_clusters(1)
assert_equal(len(clusters), 3)
clusters = tree.get_clusters(2)
assert_equal(len(clusters), 34)
thresholds = [.5]
qbx_class = QuickBundlesX(thresholds)
tree = qbx_class.cluster(parallel)
clusters = tree.get_clusters(0)
assert_equal(len(clusters), 1)
clusters = tree.get_clusters(1)
assert_equal(len(clusters), 100)
def test_3D_segments():
points = np.array([[[1, 0, 0],
[1, 1, 0]],
[[3, 1, 0],
[3, 0, 0]],
[[2, 0, 0],
[2, 1, 0]],
[[5, 1, 0],
[5, 0, 0]],
[[5.5, 0, 0],
[5.5, 1, 0]]], dtype="f4")
thresholds = [4, 2, 1]
feature = ResampleFeature(nb_points=20)
metric = AveragePointwiseEuclideanMetric(feature)
qbx = QuickBundlesX(thresholds, metric=metric)
tree = qbx.cluster(points)
clusters_0 = tree.get_clusters(0)
clusters_1 = tree.get_clusters(1)
clusters_2 = tree.get_clusters(2)
assert_equal(len(clusters_0.centroids), len(clusters_1.centroids))
assert_equal(len(clusters_2.centroids) > len(clusters_1.centroids), True)
assert_array_equal(clusters_2[1].indices, np.array([3, 4], dtype=np.int32))
def test_with_simulated_bundles2():
# Generate synthetic streamlines
bundles = bearing_bundles(4, 2)
bundles.append(straight_bundle(1))
streamlines = list(itertools.chain(*bundles))
thresholds = [10, 2, 1]
qbx_class = QuickBundlesX(thresholds)
tree = qbx_class.cluster(streamlines)
# By default `refdata` refers to data being clustered.
assert_equal(tree.refdata, streamlines)
def test_with_simulated_bundles2():
# Generate synthetic streamlines
bundles = bearing_bundles(4, 2)
bundles.append(straight_bundle(1))
streamlines = list(itertools.chain(*bundles))
thresholds = [10, 2, 1]
qbx_class = QuickBundlesX(thresholds)
tree = qbx_class.cluster(streamlines)
# By default `refdata` refers to data being clustered.
assert_equal(tree.refdata, streamlines)
def test_3D_points():
points = np.array(
[[[1, 0, 0]], [[3, 0, 0]], [[2, 0, 0]], [[5, 0, 0]], [[5.5, 0, 0]]],
dtype="f4")
thresholds = [4, 2, 1]
metric = AveragePointwiseEuclideanMetric()
qbx = QuickBundlesX(thresholds, metric=metric)
tree = qbx.cluster(points)
clusters_2 = tree.get_clusters(2)
assert_array_equal(clusters_2.clusters_sizes(), [3, 2])
clusters_0 = tree.get_clusters(0)
assert_array_equal(clusters_0.clusters_sizes(), [5])
def test_with_simulated_bundles():
streamlines = simulated_bundle(3, False, 2)
thresholds = [10, 3, 1]
qbx_class = QuickBundlesX(thresholds)
tree = qbx_class.cluster(streamlines)
for level in range(len(thresholds) + 1):
clusters = tree.get_clusters(level)
assert_equal(tree.leaves[0].indices[0], 0)
assert_equal(tree.leaves[2][0], 2)
clusters.refdata = streamlines
assert_array_equal(clusters[0][0],
np.array([[0., -10., -5.], [0., 10., -5.]]))
def test_3D_points():
points = np.array([[[1, 0, 0]],
[[3, 0, 0]],
[[2, 0, 0]],
[[5, 0, 0]],
[[5.5, 0, 0]]], dtype="f4")
thresholds = [4, 2, 1]
metric = AveragePointwiseEuclideanMetric()
qbx = QuickBundlesX(thresholds, metric=metric)
tree = qbx.cluster(points)
clusters_2 = tree.get_clusters(2)
assert_array_equal(clusters_2.clusters_sizes(), [3, 2])
clusters_0 = tree.get_clusters(0)
assert_array_equal(clusters_0.clusters_sizes(), [5])
def test_with_simulated_bundles():
streamlines = simulated_bundle(3, False, 2)
thresholds = [10, 3, 1]
qbx_class = QuickBundlesX(thresholds)
tree = qbx_class.cluster(streamlines)
for level in range(len(thresholds) + 1):
clusters = tree.get_clusters(level)
assert_equal(tree.leaves[0].indices[0], 0)
assert_equal(tree.leaves[2][0], 2)
clusters.refdata = streamlines
assert_array_equal(clusters[0][0],
np.array([[0., -10., -5.],
[0., 10., -5.]]))
def gen_qbx_tree(show=False):
from dipy.viz import window
from dipy.viz.clustering import show_clusters_graph
streamlines = get_data()
thresholds = [40, 30, 25]#, 20, 15]
#thresholds = [30, 25, 15]
qbx_class = QuickBundlesX(thresholds)
print "Clustering {} streamlines ({})...".format(len(streamlines), thresholds)
qbx = qbx_class.cluster(streamlines)
print "Displaying clusters graph..."
tree = qbx.get_tree_cluster_map()
tree.refdata = streamlines
color_tree2(tree, min_level=0)
#color_tree(tree, min_level=0)
ren = show_clusters_graph(tree, show=show)
ren.reset_camera_tight()
def test_qbx_and_merge():
# Generate synthetic streamlines
bundles = bearing_bundles(4, 2)
bundles.append(straight_bundle(1))
streamlines = Streamlines(list(itertools.chain(*bundles)))
thresholds = [10, 2, 1]
rng = np.random.RandomState(seed=42)
qbxm_centroids = qbx_and_merge(streamlines, thresholds, rng=rng).centroids
qbx = QuickBundlesX(thresholds)
tree = qbx.cluster(streamlines)
qbx_centroids = tree.get_clusters(3).centroids
assert_equal(len(qbx_centroids) > len(qbxm_centroids), True)
def gen_qbx_tree_progress():
from dipy.viz import window
from dipy.viz.clustering import show_clusters_graph_progress
streamlines = get_data()
thresholds = [40, 30, 25]#, 20, 15]
qbx_class = QuickBundlesX(thresholds)
print "Clustering {} streamlines ({})...".format(len(streamlines), thresholds)
qbx = qbx_class.cluster(streamlines)
print "Displaying clusters graph..."
tree = qbx.get_tree_cluster_map()
tree.refdata = streamlines
color_tree(tree, min_level=0)
#max_indices = [100, 500, 1000, 3000, len(streamlines)]
max_indices = [100, 250, 500, 750, 1000, 2000, 3000, 5000, len(streamlines)]
#max_indices = np.arange(10, len(streamlines), 100)
for i, ren in enumerate(show_clusters_graph_progress(tree, max_indices, show=False)):
ren.reset_camera_tight()
window.snapshot(ren, fname="tree_{}_part_{}".format("-".join(map(str, thresholds)), i), size=(1200, 1200))
def test_circle_parallel_fornix():
circle = streamlines_in_circle(100, step_size=2)
parallel = streamlines_parallel(100)
thresholds = [1, 0.1]
qbx_class = QuickBundlesX(thresholds)
tree = qbx_class.cluster(circle)
clusters = tree.get_clusters(0)
assert_equal(len(clusters), 1)
clusters = tree.get_clusters(1)
assert_equal(len(clusters), 3)
clusters = tree.get_clusters(2)
assert_equal(len(clusters), 34)
thresholds = [.5]
qbx_class = QuickBundlesX(thresholds)
tree = qbx_class.cluster(parallel)
clusters = tree.get_clusters(0)
assert_equal(len(clusters), 1)
clusters = tree.get_clusters(1)
assert_equal(len(clusters), 100)
def score_auto_extract_auto_IBs(streamlines, bundles_masks, ref_bundles, ROIs, wm,
save_segmented=False, save_IBs=False,
save_VBs=False, save_VCWPs=False,
out_segmented_strl_dir='',
base_out_segmented_strl='',
ref_anat_fname=''):
"""
TODO document
Parameters
------------
streamlines : sequence
sequence of T streamlines. One streamline is an ndarray of shape (N, 3),
where N is the number of points in that streamline, and
``streamlines[t][n]`` is the n-th point in the t-th streamline. Points
are of form x, y, z in *voxel* coordinates.
bundles_masks : sequence
list of nibabel objects corresponding to mask of bundles
ROIs : sequence
list of nibabel objects corresponding to mask of ROIs
wm : nibabel object
mask of the white matter
save_segmented : bool
if true, returns indices of streamlines composing VC, IC, VCWP and NC
Returns
---------
scores : dict
dictionnary containing a score for each metric
indices : dict
dictionnary containing the indices of streamlines composing VC, IC,
VCWP and NC
"""
# Load all streamlines, since streamlines is a generator.
# full_strl = [s for s in streamlines]
VC_indices, found_vbs_info = _auto_extract_VCs(streamlines, ref_bundles)
VC = len(VC_indices)
logging.debug('Found {} candidate VC'.format(VC))
if save_VBs:
_save_extracted_VBs(found_vbs_info, streamlines, out_segmented_strl_dir,
base_out_segmented_strl, ref_anat_fname)
# TODO might be readded
# To keep track of streamlines that have been classified
# classified_streamlines_indices = VC_indices
# New algorithm
# Step 1: remove streamlines shorter than threshold (currently 35)
# Step 2: apply Quickbundle with a distance threshold of 20
# Step 3: remove singletons
# Step 4: assign to closest ROIs pair
logging.debug("Starting IC, IB scoring")
total_strl_count = len(streamlines)
candidate_ic_strl_indices = sorted(set(range(total_strl_count)) - VC_indices)
length_thres = 35.
candidate_ic_streamlines = []
rejected_streamlines = []
for idx in candidate_ic_strl_indices:
if slength(streamlines[idx]) >= length_thres:
candidate_ic_streamlines.append(streamlines[idx].astype('f4'))
else:
rejected_streamlines.append(streamlines[idx].astype('f4'))
logging.debug('Found {} candidate IC'.format(len(candidate_ic_streamlines)))
logging.debug('Found {} streamlines that were too short'.format(len(rejected_streamlines)))
ic_counts = 0
ib_pairs = {}
if len(candidate_ic_streamlines):
# Fix seed to always generate the same output
# Shuffle to try to reduce the ordering dependency for QB
random.seed(0.2)
random.shuffle(candidate_ic_streamlines)
# TODO threshold on distance as arg
qb = QuickBundlesX([30, 25, 20, 15])
clusters_obj = qb.cluster(candidate_ic_streamlines)
clusters = clusters_obj.get_clusters(-1) # Retrieves clusters obtained with the smallest threshold.
# clusters = qb.cluster(candidate_ic_streamlines)
logging.debug("Found {} potential IB clusters".format(len(clusters)))
# TODO this should be better handled
rois_info = []
for roi in ROIs:
rois_info.append((get_root_image_name(os.path.basename(roi.get_filename())),
np.array(np.where(roi.get_data())).T))
centroids = nib.streamlines.Tractogram(clusters.centroids)
centroids.apply_affine(np.linalg.inv(ROIs[0].affine))
all_centroids_closest_pairs = get_closest_roi_pairs_for_all_streamlines(centroids.streamlines, rois_info)
for c_idx, c in enumerate(clusters):
closest_for_cluster = all_centroids_closest_pairs[c_idx]
if closest_for_cluster not in ib_pairs:
ib_pairs[closest_for_cluster] = []
ic_counts += len(c)
ib_pairs[closest_for_cluster].extend(c.indices)
# all_ics_closest_pairs = get_closest_roi_pairs_for_all_streamlines(candidate_ic_streamlines, rois_info)
# for c_idx, c in enumerate(clusters):
# closest_for_cluster = [all_ics_closest_pairs[i] for i in clusters[c]['indices']]
# if len(clusters[c]['indices']) > 1:
# ic_counts += len(clusters[c]['indices'])
# occurences = Counter(closest_for_cluster)
# # TODO handle either an equality or maybe a range
# most_frequent = occurences.most_common(1)[0][0]
# val = ib_pairs.get(most_frequent)
# if val is None:
# # Check if flipped pair exists
# val1 = ib_pairs.get((most_frequent[1], most_frequent[0]))
# if val1 is not None:
# val1.append(c_idx)
# else:
# ib_pairs[most_frequent] = [c_idx]
# else:
# val.append(c_idx)
# else:
# rejected_streamlines.append(candidate_ic_streamlines[clusters[c]['indices'][0]])
if save_segmented and save_IBs:
for k, v in ib_pairs.iteritems():
out_strl = []
# for c_idx in v:
# out_strl.extend([s for s in np.array(candidate_ic_streamlines)[clusters[c_idx]['indices']]])
out_strl = np.array(candidate_ic_streamlines)[v]
out_fname = os.path.join(out_segmented_strl_dir,
base_out_segmented_strl + \
'_IB_{0}_{1}.tck'.format(k[0], k[1]))
nib.streamlines.save(nib.streamlines.Tractogram(out_strl, affine_to_rasmm=np.eye(4)), out_fname)
# ib_f = TCK.create(out_fname)
# save_tracts_tck_from_dipy_voxel_space(ib_f, ref_anat_fname,
# out_strl)
if len(rejected_streamlines) > 0 and save_segmented:
out_nc_fname = os.path.join(out_segmented_strl_dir,
'{}_NC.tck'.format(base_out_segmented_strl))
nib.streamlines.save(nib.streamlines.Tractogram(rejected_streamlines, affine_to_rasmm=np.eye(4)), out_nc_fname)
# out_file = TCK.create(out_nc_fname)
# save_tracts_tck_from_dipy_voxel_space(out_file, ref_anat_fname,
# rejected_streamlines)
# TODO readd classifed_steamlines_indices to validate
if ic_counts != len(candidate_ic_strl_indices) - len(rejected_streamlines):
raise ValueError("Some streamlines were not correctly assigned to NC")
VC /= total_strl_count
IC = (len(candidate_ic_strl_indices) - len(rejected_streamlines)) / total_strl_count
NC = len(rejected_streamlines) / total_strl_count
VCWP = 0
# TODO could have sanity check on global extracted streamlines vs all
# possible indices
nb_VB_found = [v['nb_streamlines'] > 0 for k, v in found_vbs_info.iteritems()].count(True)
streamlines_per_bundle = {k: v['nb_streamlines'] for k, v in found_vbs_info.iteritems() if v['nb_streamlines'] > 0}
scores = {}
scores['version'] = 2
scores['algo_version'] = 5
scores['VC'] = VC
scores['IC'] = IC
scores['VCWP'] = VCWP
scores['NC'] = NC
scores['VB'] = nb_VB_found
scores['IB'] = len(ib_pairs.keys())
scores['streamlines_per_bundle'] = streamlines_per_bundle
scores['total_streamlines_count'] = total_strl_count
return scores
def score_auto_extract_auto_IBs(streamlines,
bundles_masks,
ref_bundles,
ROIs,
wm,
save_segmented=False,
save_IBs=False,
save_VBs=False,
save_VCWPs=False,
out_segmented_strl_dir='',
base_out_segmented_strl='',
ref_anat_fname=''):
"""
TODO document
Parameters
------------
streamlines : sequence
sequence of T streamlines. One streamline is an ndarray of shape (N, 3),
where N is the number of points in that streamline, and
``streamlines[t][n]`` is the n-th point in the t-th streamline. Points
are of form x, y, z in *voxel* coordinates.
bundles_masks : sequence
list of nibabel objects corresponding to mask of bundles
ROIs : sequence
list of nibabel objects corresponding to mask of ROIs
wm : nibabel object
mask of the white matter
save_segmented : bool
if true, returns indices of streamlines composing VC, IC, VCWP and NC
Returns
---------
scores : dict
dictionnary containing a score for each metric
indices : dict
dictionnary containing the indices of streamlines composing VC, IC,
VCWP and NC
"""
# Load all streamlines, since streamlines is a generator.
# full_strl = [s for s in streamlines]
VC_indices, found_vbs_info = _auto_extract_VCs(streamlines, ref_bundles)
VC = len(VC_indices)
logging.debug('Found {} candidate VC'.format(VC))
if save_VBs:
_save_extracted_VBs(found_vbs_info, streamlines,
out_segmented_strl_dir, base_out_segmented_strl,
ref_anat_fname)
# TODO might be readded
# To keep track of streamlines that have been classified
# classified_streamlines_indices = VC_indices
# New algorithm
# Step 1: remove streamlines shorter than threshold (currently 35)
# Step 2: apply Quickbundle with a distance threshold of 20
# Step 3: remove singletons
# Step 4: assign to closest ROIs pair
logging.debug("Starting IC, IB scoring")
total_strl_count = len(streamlines)
candidate_ic_strl_indices = sorted(
set(range(total_strl_count)) - VC_indices)
length_thres = 35.
candidate_ic_streamlines = []
rejected_streamlines = []
for idx in candidate_ic_strl_indices:
if slength(streamlines[idx]) >= length_thres:
candidate_ic_streamlines.append(streamlines[idx].astype('f4'))
else:
rejected_streamlines.append(streamlines[idx].astype('f4'))
logging.debug('Found {} candidate IC'.format(
len(candidate_ic_streamlines)))
logging.debug('Found {} streamlines that were too short'.format(
len(rejected_streamlines)))
ic_counts = 0
ib_pairs = {}
if len(candidate_ic_streamlines):
# Fix seed to always generate the same output
# Shuffle to try to reduce the ordering dependency for QB
random.seed(0.2)
random.shuffle(candidate_ic_streamlines)
# TODO threshold on distance as arg
qb = QuickBundlesX([30, 25, 20, 15])
clusters_obj = qb.cluster(candidate_ic_streamlines)
clusters = clusters_obj.get_clusters(
-1) # Retrieves clusters obtained with the smallest threshold.
# clusters = qb.cluster(candidate_ic_streamlines)
logging.debug("Found {} potential IB clusters".format(len(clusters)))
# TODO this should be better handled
rois_info = []
for roi in ROIs:
rois_info.append(
(get_root_image_name(os.path.basename(roi.get_filename())),
np.array(np.where(roi.get_data())).T))
centroids = nib.streamlines.Tractogram(clusters.centroids)
centroids.apply_affine(np.linalg.inv(ROIs[0].affine))
all_centroids_closest_pairs = get_closest_roi_pairs_for_all_streamlines(
centroids.streamlines, rois_info)
for c_idx, c in enumerate(clusters):
closest_for_cluster = all_centroids_closest_pairs[c_idx]
if closest_for_cluster not in ib_pairs:
ib_pairs[closest_for_cluster] = []
ic_counts += len(c)
ib_pairs[closest_for_cluster].extend(c.indices)
# all_ics_closest_pairs = get_closest_roi_pairs_for_all_streamlines(candidate_ic_streamlines, rois_info)
# for c_idx, c in enumerate(clusters):
# closest_for_cluster = [all_ics_closest_pairs[i] for i in clusters[c]['indices']]
# if len(clusters[c]['indices']) > 1:
# ic_counts += len(clusters[c]['indices'])
# occurences = Counter(closest_for_cluster)
# # TODO handle either an equality or maybe a range
# most_frequent = occurences.most_common(1)[0][0]
# val = ib_pairs.get(most_frequent)
# if val is None:
# # Check if flipped pair exists
# val1 = ib_pairs.get((most_frequent[1], most_frequent[0]))
# if val1 is not None:
# val1.append(c_idx)
# else:
# ib_pairs[most_frequent] = [c_idx]
# else:
# val.append(c_idx)
# else:
# rejected_streamlines.append(candidate_ic_streamlines[clusters[c]['indices'][0]])
if save_segmented and save_IBs:
for k, v in ib_pairs.iteritems():
out_strl = []
# for c_idx in v:
# out_strl.extend([s for s in np.array(candidate_ic_streamlines)[clusters[c_idx]['indices']]])
out_strl = np.array(candidate_ic_streamlines)[v]
out_fname = os.path.join(out_segmented_strl_dir,
base_out_segmented_strl + \
'_IB_{0}_{1}.tck'.format(k[0], k[1]))
nib.streamlines.save(
nib.streamlines.Tractogram(out_strl,
affine_to_rasmm=np.eye(4)),
out_fname)
# ib_f = TCK.create(out_fname)
# save_tracts_tck_from_dipy_voxel_space(ib_f, ref_anat_fname,
# out_strl)
if len(rejected_streamlines) > 0 and save_segmented:
out_nc_fname = os.path.join(
out_segmented_strl_dir,
'{}_NC.tck'.format(base_out_segmented_strl))
nib.streamlines.save(
nib.streamlines.Tractogram(rejected_streamlines,
affine_to_rasmm=np.eye(4)),
out_nc_fname)
# out_file = TCK.create(out_nc_fname)
# save_tracts_tck_from_dipy_voxel_space(out_file, ref_anat_fname,
# rejected_streamlines)
# TODO readd classifed_steamlines_indices to validate
if ic_counts != len(candidate_ic_strl_indices) - len(rejected_streamlines):
raise ValueError("Some streamlines were not correctly assigned to NC")
VC /= total_strl_count
IC = (len(candidate_ic_strl_indices) -
len(rejected_streamlines)) / total_strl_count
NC = len(rejected_streamlines) / total_strl_count
VCWP = 0
# TODO could have sanity check on global extracted streamlines vs all
# possible indices
nb_VB_found = [
v['nb_streamlines'] > 0 for k, v in found_vbs_info.iteritems()
].count(True)
streamlines_per_bundle = {
k: v['nb_streamlines']
for k, v in found_vbs_info.iteritems() if v['nb_streamlines'] > 0
}
scores = {}
scores['version'] = 2
scores['algo_version'] = 5
scores['VC'] = VC
scores['IC'] = IC
scores['VCWP'] = VCWP
scores['NC'] = NC
scores['VB'] = nb_VB_found
scores['IB'] = len(ib_pairs.keys())
scores['streamlines_per_bundle'] = streamlines_per_bundle
scores['total_streamlines_count'] = total_strl_count
return scores
