def test_reduce(values, labels, kwds, out, expects):
if expects:
with pytest.raises(expects):
parc.reduce_by_labels(values, labels, **kwds)
else:
res = parc.reduce_by_labels(values, labels, **kwds)
assert np.allclose(res, out)
assert res.dtype == out.dtype
assert res.shape == out.shape
def compute_mpc(profile, labels):
"""Computes MPC for given labels on a surface template.
Parameters
----------
profile : numpy.ndarray
Histological profiles of size surface-by-vertex.
labels : numpy.ndarray
Labels of regions of interest. Use 0 to denote regions that will not be included.
Returns
-------
numpy.ndarray
Microstructural profile covariance.
"""
roi_profile = reduce_by_labels(profile, labels)
if np.any(labels == 0):
# Remove 0's in the labels.
roi_profile = roi_profile[:, 1:]
p_corr = partial_correlation(roi_profile, np.mean(roi_profile, axis=1))
mpc = 0.5 * np.log((1 + p_corr) / (1 - p_corr))
mpc[p_corr > 0.99999] = 0 # Deals with floating point issues where p_corr==1
mpc[mpc == np.inf] = 0
mpc[mpc == np.nan] = 0
return mpc
def _get_prob_icn(schaefer1000_lab, yeo7_lab):
prob_icn = reduce_by_labels(pd.get_dummies(yeo7_lab),
schaefer1000_lab,
red_op='mean',
dtype=np.float32,
axis=1)
prob_icn /= prob_icn.sum(axis=1, keepdims=True)
return prob_icn
def _compute_net_idio(df_phen, dd, sd, lab_ref):
"""Network-wise idiosyncrasy. """
dd_net = reduce_by_labels(dd, lab_ref, axis=0)
dd_net = pd.DataFrame(dd_net, columns=ICN_NAMES)
dd_net['Average'] = dd.mean(1)
sd_net = reduce_by_labels(sd, lab_ref, axis=0)
sd_net = pd.DataFrame(sd_net, columns=ICN_NAMES)
sd_net['Average'] = sd.mean(1)
df_ados = pd.concat([sd_net, dd_net], keys=['SD', 'DD'], axis=1)
for k in ['Age', 'Group', 'Site', 'Sex', 'CSS']:
df_ados[k] = df_phen[k].to_numpy()
# Remove subject with no CSS
mask_na = np.logical_or(~pd.isna(df_ados['CSS']), df_ados.Group != 'ASD')
df_ados = df_ados[mask_na].copy()
return df_ados
def yeo_networks_associations(
data: ArrayLike,
template: str = "fsaverage5",
seven_networks: bool = True,
data_dir: Optional[Union[str, Path]] = None,
reduction_operation: Union[str, Callable] = np.nanmean,
) -> np.ndarray:
"""Computes association
Parameters
----------
data : ArrayLike
Data to be summarized in the Yeo networks in a sample-by-feature format.
template : str, optional
Surface template. Valid values are "fsaverage5", "fsaverage", and
"fslr32k", "civet41k", and "civet164k", by default "fsaverage5".
seven_networks : bool, optional
If true, uses the 7 network parcellation, otherwise uses the 17 network
parcellation, by default True.
data_dir : str, Path, optional
Data directory to store the Yeo network files, by default $HOME_DIR/brainstat_data/parcellation_data.
reduction_operation : str, callable, optional
How to summarize data. If str, options are: {‘min’, ‘max’, ‘sum’,
‘mean’, ‘median’, ‘mode’, ‘average’}. If callable, it should receive a
1D array of values, array of weights (or None) and return a scalar
value. Default is ‘mean’.
Returns
-------
np.ndarray
Summary statistic in the yeo networks.
"""
n_regions = 7 if seven_networks else 17
yeo_networks = fetch_parcellation(
template=template,
atlas="yeo",
n_regions=n_regions,
join=True,
data_dir=data_dir,
)
if np.array(data).ndim == 1:
data_2d = np.array(data)[:, None]
else:
data_2d = np.array(data)
n_features = data_2d.shape[1]
yeo_mean = np.zeros((n_regions + 1, n_features))
for i in range(n_features):
yeo_mean[:, i] = reduce_by_labels(
data_2d[:, i], yeo_networks, red_op=reduction_operation
)
return yeo_mean[1:, :]
def load_hcp_timeseries(file, parcellations=None, cortex_only=True):
"""Fetches (parcellated) timeseries of a subject
Parameters
----------
subject_dir : str
Path to an HCP subject directory
parcellations : str, list, optional
Files describing the parcellation of the timeseries. If multiple files
are provided (e.g. left and right hemisphere), then the parcellations
are concatenated in the order that they're provided in. Defaults to
None.
cortex_only : bool, optional
If true, returns only cortical data, defaults to True.
Returns
-------
numpy.array
Array containing the timeseries
"""
if isinstance(parcellations, str):
parcellations = [parcellations]
cii = nib.load(file)
timeseries = cii.get_fdata()
if cortex_only:
timeseries = timeseries[:, :64984]
if parcellations is not None:
parcel_list = [nib.load(x).darrays[0].data for x in parcellations]
parcellation = np.concatenate(parcel_list, axis=0)
timeseries = [
reduce_by_labels(x, parcellation)[1:] for x in timeseries
]
return timeseries
masked_regions = [b'Medial_wall', b'Unknown']
masked_labels = [regions.index(r) for r in masked_regions]
for r in masked_regions:
regions.remove(r)
# Build Destrieux parcellation and mask
labeling = np.concatenate([atlas['map_left'], atlas['map_right']])
mask = ~np.isin(labeling, masked_labels)
# Distinct labels for left and right hemispheres
lab_lh = atlas['map_left']
labeling[lab_lh.size:] += lab_lh.max() + 1
# extract mean timeseries for each label
seed_ts = reduce_by_labels(clean_ts[mask],
labeling[mask],
axis=1,
red_op='mean')
###############################################################################
# Calculate functional connectivity matrix
# ++++++++++++++++++++++++++++++++++++++++
# The following example uses
# `nilearn <https://nilearn.github.io/auto_examples/03_connectivity/plot_
# signal_extraction.html#compute-and-display-a-correlation-matrix/>`_:
from nilearn.connectome import ConnectivityMeasure
correlation_measure = ConnectivityMeasure(kind='correlation')
correlation_matrix = correlation_measure.fit_transform([seed_ts.T])[0]
###############################################################################
# Expression is a pandas DataFrame which shows the genetic expression of genes
# within each region of the atlas. By default, the values will fall in the range
# [0, 1] where higher values represent higher expression. However, if you change
# the normalization function then this may change. Some regions may return NaN
# values for all genes. This occurs when there are no samples within this
# region across all donors. We've denoted this region with the black color in the
# matrix. By default, BrainStat uses all the default abagen parameters. If you wish to
# customize these parameters then the keyword arguments can be passed directly
# to `surface_genetic_expression`. For a full list of these arguments and their
# function please consult the abagen documentation.
#
# Next, lets have a look at the correlation between one gene (WFDC1) and our
# t-statistic map. Lets also plot the expression of this gene to the surface.
# Plot correlation with WFDC1 gene
t_stat_schaefer_100 = reduce_by_labels(slm.t.flatten(), schaefer_100_fs5)[1:]
df = pd.DataFrame({"x": t_stat_schaefer_100, "y": expression["WFDC1"]})
df.dropna(inplace=True)
plt.scatter(df.x, df.y, s=20, c="k")
plt.xticks(fontsize=14)
plt.yticks(fontsize=14)
plt.xlabel("t-statistic", fontdict={"fontsize": 16})
plt.ylabel("WFDC1 expression", fontdict={"fontsize": 16})
plt.plot(np.unique(df.x),
np.poly1d(np.polyfit(df.x, df.y, 1))(np.unique(df.x)), "k")
plt.text(-1.0, 0.75, f"r={df.x.corr(df.y):.2f}", fontdict={"size": 14})
plt.show()
########################################################################