# Authors: Robert Luke <mail@robertluke.net>
#
# License: BSD (3-clause)
from copy import deepcopy
from pathlib import PosixPath
import warnings
import pandas as pd
import numpy as np
from numpy import array_equal, where
from h5io import read_hdf5, write_hdf5
with warnings.catch_warnings(record=True):
warnings.simplefilter('ignore')
import nilearn.glm
from nilearn.glm.first_level import run_glm as nilearn_glm
try: # remove once MNE 1.6 is required
from mne._fiff.meas_info import ContainsMixin
except ImportError:
from mne.io.meas_info import ContainsMixin
from mne.utils import fill_doc, warn, verbose, check_fname, _validate_type
from mne.io.pick import _picks_to_idx
from mne.io.constants import FIFF
from mne import Info, pick_info
from ..visualisation._plot_GLM_topo import _plot_glm_topo,\
_plot_glm_contrast_topo
from ..visualisation import plot_glm_surface_projection
from ..statistics._roi import _glm_region_of_interest
@fill_doc
class _BaseGLM(ContainsMixin):
"""Base GLM results class."""
@property
def ch_names(self):
"""Return the channel names.
Returns
-------
names : array
The channel names.
"""
return self.info['ch_names']
def __str__(self):
return (f"GLM Results for {len(self.ch_names)} channels")
def __repr__(self):
return (f"GLM Results for {len(self.ch_names)} channels")
def __len__(self):
return len(self.info.ch_names)
def _get_state(self): # noqa: D105
"""Return the state of the GLM results as a dictionary.
See _state_to_glm for inverse operation.
Returns
-------
state : dict
State of the object.
"""
state = deepcopy(dict(
data=self._data,
design=self.design,
info=self.info,
preload=self.preload,
classname=str(self.__class__)
))
if isinstance(state['data'], dict):
for channel in state['data']:
state['data'][channel] = state['data'][channel].__dict__
if isinstance(state['data'][channel]['model'],
nilearn.glm.regression.OLSModel):
state['data'][channel]['modelname'] = \
str(state['data'][channel]['model'].__class__)
state['data'][channel]['model'] = \
state['data'][channel]['model'].__dict__
if isinstance(state['data'], nilearn.glm.contrasts.Contrast):
state['data'] = state['data'].__dict__
return state
def copy(self):
"""Return a copy of the GLM results.
Returns
-------
inst : instance of ResultsGLM
A copy of the object.
"""
return deepcopy(self)
@fill_doc
def save(self, fname, overwrite=False):
"""Save GLM results to disk.
Parameters
----------
fname : str
The filename to use to write the HDF5 data.
Should end in ``'glm.h5'``.
%(overwrite)s
"""
_validate_type(fname, 'path-like', 'fname')
if isinstance(fname, PosixPath):
fname = str(fname)
if not fname.endswith('glm.h5'):
raise IOError('The filename must end with glm.h5, '
f'instead received {fname}')
write_hdf5(fname, self._get_state(),
overwrite=overwrite, title='mnepython')
def to_dataframe(self, order=None):
"""Return a tidy dataframe representing the GLM results.
Parameters
----------
order : list
Order in which the rows should be returned by channel name.
Returns
-------
tidy : pandas.DataFrame
Dataframe containing GLM results.
"""
from ..utils import glm_to_tidy
if order is None:
order = self.ch_names
return glm_to_tidy(self.info, self._data, self.design, order=order)
def scatter(self, conditions=[], exclude_no_interest=True, axes=None,
no_interest=None):
"""Scatter plot of the GLM results.
Parameters
----------
conditions : list
List of condition names to plot. By default plots all regressors
of interest.
exclude_no_interest : bool
Exclude regressors of no interest from the figure.
axes : Axes
Optional axes on which to plot the data.
no_interest : list
List of regressors that are of no interest. If none are specified
then conditions starting with
["drift", "constant", "short", "Short"] will be excluded.
Returns
-------
plt : matplotlib.Figure
Scatter plot.
"""
if no_interest is None:
no_interest = ["drift", "constant", "short", "Short"]
import matplotlib.pyplot as plt
df = self.to_dataframe()
x_column = "Condition"
y_column = "theta"
if "ContrastType" in df.columns:
x_column = "ch_name"
y_column = "effect"
if len(conditions) == 0:
conditions = ["t", "f"]
df = df.query('ContrastType in @conditions')
else:
if len(conditions) == 0:
conditions = self.design.columns
df = df.query('Condition in @conditions')
if exclude_no_interest:
for no_i in no_interest:
df = df[~df[x_column].astype(str).str.startswith(no_i)]
df_hbo = df.query('Chroma == "hbo"')
df_hbr = df.query('Chroma == "hbr"')
if axes is None:
fig, axes = plt.subplots(nrows=1, ncols=1, figsize=(10, 5))
axes.scatter(df_hbo[x_column], df_hbo[y_column] * 1e6, c="r")
axes.scatter(df_hbr[x_column], df_hbr[y_column] * 1e6, c="b")
axes.set_xlabel(x_column)
axes.set_ylabel(y_column.capitalize())
axes.legend(["Oxyhaemoglobin", "Deoxyhaemoglobin"])
axes.hlines([0.0], 0, len(np.unique(df[x_column])) - 1)
if len(np.unique(df[x_column])) > 8:
plt.xticks(rotation=45, ha='right')
return axes
[docs]
@fill_doc
class RegressionResults(_BaseGLM):
"""
Class containing GLM regression results.
Parameters
----------
info : mne.Info
Info.
data : dict
Dictionary.
design : dataframe
Design matrix.
Returns
-------
glm_est : ResultsGLM,
Result class.
"""
@property
def data(self):
return self._data
@data.setter
def data(self, data):
if type(data) is not dict:
raise TypeError("Data must be a dictionary type")
if not array_equal(list(data.keys()), self.info.ch_names):
raise TypeError("Dictionary keys must match ch_names")
for idx in range(len(self.info.ch_names)):
if list(data.keys())[idx] is not self.info.ch_names[idx]:
raise TypeError("Data names and channel names do not match")
for d in data:
if type(data[d]) is not nilearn.glm.regression.RegressionResults:
raise TypeError("Dictionary items must be"
" nilearn RegressionResults")
self._data = data
def __init__(self, info, data, design): # noqa: D102
self.info = info
self.data = data
self.design = design
self.preload = True
def __eq__(self, res):
same_keys = self.data.keys() == res.data.keys()
same_design = (self.design == res.design).all().all()
same_ch = self.info.ch_names == res.info.ch_names
same_theta = np.sum([(res.theta()[idx] == val).all()
for idx, val in
enumerate(self.theta())]) == len(self.ch_names)
return int(same_ch and same_design and same_keys and same_theta)
[docs]
@fill_doc
def pick(self, picks, exclude=()):
"""Pick a subset of channels.
Parameters
----------
%(picks_all)s
exclude : list | str
Set of channels to exclude, only used when picking based on
types (e.g., exclude="bads" when picks="meg").
Returns
-------
inst : instance of ResultsGLM
The modified instance.
"""
picks = _picks_to_idx(self.info, picks, 'all', exclude,
allow_empty=False)
pick_info(self.info, picks, copy=False)
self._data = {key: self._data[key] for key in self.info.ch_names}
return self
[docs]
def theta(self):
"""Return the GLM theta results.
Returns
-------
thetas : array
Array of thetas. An array is provided per channel.
"""
return [self._data[key].theta for key in self.info.ch_names]
[docs]
def MSE(self):
"""Return the GLM MSE.
Returns
-------
thetas : array
Array of MSEs. A value is provided per channel.
"""
return [self._data[key].MSE[0] for key in self.info.ch_names]
[docs]
def model(self):
"""Return the GLM model.
Returns
-------
models : array
Array of models. An model is provided per channel.
"""
return [self._data[key].model for key in self.info.ch_names]
[docs]
def compute_contrast(self, contrast, contrast_type=None):
"""
Compute contrasts on regression results.
This is a wrapper function for nilearn.stats.contrasts.
Parameters
----------
contrast : numpy.ndarray of shape (p) or (q, p),
Where q = number of contrast vectors and p = number of regressors.
contrast_type : {None, ‘t’, ‘F’}, optional
Type of the contrast. If None, then defaults to ‘t’ for 1D con_val
and ‘F’ for 2D con_val.
Returns
-------
contrast : Contrast instance,
Yields the statistics of the contrast
(effects, variance, p-values).
"""
cont = _compute_contrast(self._data, contrast,
contrast_type=contrast_type)
return ContrastResults(self.info, cont, self.design)
[docs]
def plot_topo(self, conditions=None, axes=None, *, vlim=(None, None),
vmin=None, vmax=None, colorbar=True, figsize=(12, 7),
sphere=None):
"""Plot 2D topography of GLM data.
Parameters
----------
conditions : array
Which conditions should be displayed.
axes : instance of Axes | None
The axes to plot to. If None, a new figure is used.
vlim : tuple of length 2
Colormap limits to use. If a :class:`tuple` of floats, specifies
the lower and upper bounds of the colormap (in that order);
providing ``None`` for either entry will set the corresponding
boundary at the positive or negative max of the absolute value of
the data.
vmin : float | None
Deprecated, use vlim instead.
vmax : float | None
Deprecated, use vlim instead.
colorbar : Bool
Should a colorbar be plotted.
figsize : two values
Figure size.
sphere : As specified in MNE
Sphere parameter from mne.viz.topomap.plot_topomap.
Returns
-------
fig : figure
Figure of each design matrix component for hbo (top row)
and hbr (bottom row).
"""
return _plot_glm_topo(self.info, self._data, self.design,
requested_conditions=conditions,
axes=axes, vlim=vlim, vmin=vmin, vmax=vmax,
colorbar=colorbar,
figsize=figsize, sphere=sphere)
[docs]
def to_dataframe_region_of_interest(self, group_by, condition,
weighted=True, demographic_info=False):
"""Region of interest results as a dataframe.
Parameters
----------
group_by : dict
Specifies which channels are aggregated into a single ROI.
The dict key will be used as the ROI label and the dict
values must be lists of picks (either channel names or
integer indices of ``epochs.ch_names``). For example::
group_by=dict(Left_ROI=[1, 2, 3, 4], Right_ROI=[5, 6, 7, 8])
condition : str | list
Name to be used for condition.
weighted : Bool | dict
Weighting to be applied to each channel in the ROI computation.
If False, then all channels will be weighted equally.
If True, channels will be weighted by the inverse of
the standard error of the GLM fit.
For manual specification of the channel weighting a dictionary
can be provided.
If a dictionary is provided, the keys and length of lists must
match the ``group_by`` parameters.
The weights will be scaled internally to sum to 1.
demographic_info : Bool
Add an extra column with demographic information from
info["subject_info"].
Returns
-------
tidy : pandas.DataFrame
Dataframe containing GLM results.
"""
if isinstance(condition, str):
condition = [condition]
if isinstance(weighted, dict):
if weighted.keys() != group_by.keys():
raise KeyError("Keys of group_by and weighted "
"must be the same")
for key in weighted.keys():
if len(weighted[key]) != len(group_by[key]):
raise ValueError("The length of the keys for group_by "
"and weighted must match")
if (np.array(weighted[key]) < 0).any():
raise ValueError("Weights must be positive values")
tidy = pd.DataFrame()
for cond in condition:
cond_idx = where([c == cond for c in self.design.columns])[0]
if not len(cond_idx):
raise KeyError(f'condition {repr(cond)} not found in '
f'self.design.columns: {self.design.columns}')
roi = _glm_region_of_interest(self._data, group_by,
cond_idx, cond, weighted)
tidy = pd.concat([tidy, roi])
if weighted is True:
tidy["Weighted"] = "Inverse standard error"
elif weighted is False:
tidy["Weighted"] = "Equal"
elif isinstance(weighted, dict):
tidy["Weighted"] = "Custom"
if demographic_info:
if 'age' in self.info['subject_info'].keys():
tidy['Age'] = float(self.info["subject_info"]['age'])
if 'sex' in self.info['subject_info'].keys():
if self.info["subject_info"]['sex'] == \
FIFF.FIFFV_SUBJ_SEX_MALE:
sex = "male"
elif self.info["subject_info"]['sex'] == \
FIFF.FIFFV_SUBJ_SEX_FEMALE:
sex = "female"
else:
sex = "unknown"
tidy['Sex'] = sex
if 'Hand' in self.info['subject_info'].keys():
tidy['Hand'] = self.info["subject_info"]['hand']
return tidy
[docs]
@verbose
def surface_projection(self, chroma="hbo", condition=None,
background='w', figure=None, clim='auto',
mode='weighted', colormap='RdBu_r',
surface='pial', hemi='both', size=800,
view=None, colorbar=True, distance=0.03,
subjects_dir=None, src=None, verbose=False):
"""
Project GLM results on to the surface of the brain.
Note: This function provides a convenient wrapper around low level
MNE-Python functions. It is convenient if you wish to use a generic
head model.
If you have acquired fMRI images you may wish to use the underlying
lower level functions.
Note: This function does not conduct a forward model analysis with
photon migration etc. It simply projects the values from each channel
to the local cortical surface.
It is useful for visualisation, but users should
take this in to consideration when drawing conclusions from the
visualisation.
Parameters
----------
chroma : str
Chromophore to plot`.
condition : str
Condition to plot`.
background : matplotlib color
Color of the background of the display window.
figure : mayavi.core.api.Scene, matplotlib.figure.Figure, list, None
If None, a new figure will be created. If multiple views or a
split view is requested, this must be a list of the appropriate
length. If int is provided it will be used to identify the Mayavi
figure by it's id or create a new figure with the given id. If an
instance of matplotlib figure, mpl backend is used for plotting.
%(clim)s
mode : str
Can be "sum" to do a linear sum of weights, "weighted" to make this
a weighted sum, "nearest" to
use only the weight of the nearest sensor, or "single" to
do a distance-weight of the nearest sensor. Default is "sum".
colormap : str
Colormap to use.
surface : str
The type of surface (inflated, white etc.).
hemi : str
Hemisphere id (ie 'lh', 'rh', 'both', or 'split'). In the case
of 'both', both hemispheres are shown in the same window.
In the case of 'split' hemispheres are displayed side-by-side
in different viewing panes.
size : float or tuple of float
The size of the window, in pixels. can be one number to specify
a square window, or the (width, height) of a rectangular window.
Has no effect with mpl backend.
view : str
View to set brain to.
colorbar : bool
If True, display colorbar on scene.
distance : float
Distance (m) defining the activation "ball" of the sensor.
%(subjects_dir)s
src : instance of SourceSpaces
The source space.
%(verbose)s
Returns
-------
figure : instance of mne.viz.Brain | matplotlib.figure.Figure
An instance of :class:`mne.viz.Brain` or matplotlib figure.
"""
df = self.to_dataframe(order=self.ch_names)
if condition is None:
warn("You must provide a condition to plot", ValueError)
df_use = df.query("Condition in @condition")
if len(df_use) == 0:
raise KeyError(
f'condition={repr(condition)} not found in conditions: '
f'{sorted(set(df["Condition"]))}')
df = df_use
df = df.query("Chroma in @chroma").copy()
df["theta"] = df["theta"] * 1e6
info = self.copy().pick(chroma).info
return plot_glm_surface_projection(info, df, value="theta",
picks=chroma, background=background,
figure=figure, clim=clim,
mode=mode, colormap=colormap,
surface=surface, hemi=hemi,
size=size,
view=view, colorbar=colorbar,
distance=distance,
subjects_dir=subjects_dir, src=src,
verbose=verbose
)
[docs]
@fill_doc
class ContrastResults(_BaseGLM):
"""
Class containing GLM contrast results.
Parameters
----------
info : mne.Info
Info.
data : dict
Dictionary.
design : dataframe
Design matrix.
Returns
-------
glm_est : ResultsGLM,
Result class.
"""
def __init__(self, info, data, design): # noqa: D102
self.info = info
self.data = data
self.design = design
self.preload = True
def __eq__(self, res):
same_design = (self.design == res.design).all().all()
same_ch = self.info.ch_names == res.info.ch_names
return int(same_ch and same_design)
@property
def data(self):
return self._data
@data.setter
def data(self, data):
if not isinstance(data, nilearn.glm.contrasts.Contrast):
raise TypeError("Data must be a nilearn glm contrast type")
if data.effect.size != len(self.info.ch_names):
raise TypeError("Data results must be the same length "
"as the number of channels")
self._data = data
[docs]
def plot_topo(self, figsize=(12, 7), sphere=None):
"""
Plot topomap GLM contrast data.
Parameters
----------
figsize : numbers
TODO: Remove this, how does MNE usually deal with this.
sphere : numbers
As specified in MNE.
Returns
-------
fig : figure
Figure of each design matrix component for hbo (top row)
and hbr (bottom row).
"""
return _plot_glm_contrast_topo(self.info, self._data,
figsize=figsize, sphere=sphere)
def run_GLM(raw, design_matrix, noise_model='ar1', bins=0,
n_jobs=1, verbose=0):
"""
Run GLM on data using supplied design matrix.
This is a wrapper function for nilearn.stats.first_level_model.run_glm.
Parameters
----------
raw : instance of Raw
The haemoglobin data.
design_matrix : as specified in Nilearn
The design matrix.
noise_model : {'ar1', 'ols', 'arN', 'auto'}, optional
The temporal variance model. Defaults to first order
auto regressive model 'ar1'.
The AR model can be set to any integer value by modifying the value
of N. E.g. use `ar5` for a fifth order model.
If the string `auto` is provided a model with order 4 times the sample
rate will be used.
bins : int, optional
Maximum number of discrete bins for the AR coef histogram/clustering.
By default the value is 0, which will set the number of bins to the
number of channels, effectively estimating the AR model for each
channel.
n_jobs : int, optional
The number of CPUs to use to do the computation. -1 means
'all CPUs'.
verbose : int, optional
The verbosity level. Default is 0.
Returns
-------
glm_estimates : dict
Keys correspond to the different labels values values are
RegressionResults instances corresponding to the voxels.
"""
warn('"run_GLM" has been deprecated in favor of the more '
'comprehensive run_glm function, and will be removed in v1.0.0. '
'See the changelog for further details.',
DeprecationWarning)
res = run_glm(raw, design_matrix, noise_model=noise_model, bins=bins,
n_jobs=n_jobs, verbose=verbose)
return res.data
[docs]
def run_glm(raw, design_matrix, noise_model='ar1', bins=0,
n_jobs=1, verbose=0):
"""
GLM fit for an MNE structure containing fNIRS data.
This is a wrapper function for nilearn.stats.first_level_model.run_glm.
Parameters
----------
raw : instance of Raw
The haemoglobin data.
design_matrix : as specified in Nilearn
The design matrix as generated by
`mne_nirs.make_first_level_design_matrix`.
See example ``9.5.5. Examples of design matrices`` at
https://nilearn.github.io/auto_examples/index.html
for details on how to specify design matrices.
noise_model : {'ar1', 'ols', 'arN', 'auto'}, optional
The temporal variance model. Defaults to first order
auto regressive model 'ar1'.
The AR model can be set to any integer value by modifying the value
of N. E.g. use `ar5` for a fifth order model.
If the string `auto` is provided a model with order 4 times the sample
rate will be used.
bins : int, optional
Maximum number of discrete bins for the AR coef histogram/clustering.
By default the value is 0, which will set the number of bins to the
number of channels, effectively estimating the AR model for each
channel.
n_jobs : int, optional
The number of CPUs to use to do the computation. -1 means
'all CPUs'.
verbose : int, optional
The verbosity level. Default is 0.
Returns
-------
glm_estimates : RegressionResults
RegressionResults class which stores the GLM results.
"""
picks = _picks_to_idx(raw.info, 'fnirs', exclude=[], allow_empty=True)
ch_names = raw.ch_names
if noise_model == 'auto':
noise_model = f"ar{int(np.round(raw.info['sfreq'] * 4))}"
if bins == 0:
bins = len(raw.ch_names)
results = dict()
for pick in picks:
labels, glm_estimates = nilearn_glm(raw.get_data(pick).T,
design_matrix.values,
noise_model=noise_model, bins=bins,
n_jobs=n_jobs, verbose=verbose)
results[ch_names[pick]] = glm_estimates[labels[0]]
return RegressionResults(raw.info, results, design_matrix)
def _compute_contrast(glm_est, contrast, contrast_type=None):
from nilearn.glm.contrasts import compute_contrast as _cc
return _cc(np.array(list(glm_est.keys())), glm_est, contrast,
contrast_type=contrast_type)
[docs]
def read_glm(fname):
"""
Read GLM results from disk.
Parameters
----------
fname : str
The file name, which should end with ``glm.h5``.
Returns
-------
glm : RegressionResults or ContrastResults
RegressionResults or ContrastResults class
which stores the GLM results.
"""
check_fname(fname, 'path-like', 'glm.h5')
glm = read_hdf5(fname, title='mnepython')
return _state_to_glm(glm)
def _state_to_glm(glm):
"""
Convert state of GLM stored as dictionary to a MNE-NIRS GLM type.
Parameters
----------
glm : dict
State of GLM type.
Returns
-------
glm : _BaseGLM
RegressionResults or ContrastResults class
which stores the GLM results.
"""
if glm['classname'] == "<class 'mne_nirs.statistics._glm_level_first" \
".RegressionResults'>":
for channel in glm['data']:
# Recreate model type
if glm['data'][channel]['modelname'] == \
"<class 'nilearn.glm.regression.ARModel'>":
model = nilearn.glm.regression.ARModel(
glm['data'][channel]['model']['design'],
glm['data'][channel]['model']['rho'],
)
elif glm['data'][channel]['modelname'] == \
"<class 'nilearn.glm.regression.OLSModel'>":
model = nilearn.glm.regression.OLSModel(
glm['data'][channel]['model']['design'],
)
else:
raise IOError("Unknown model type "
f"{glm['data'][channel]['modelname']}")
for key in glm['data'][channel]['model']:
model.__setattr__(key, glm['data'][channel]['model'][key])
glm['data'][channel]['model'] = model
# Then recreate result type
res = nilearn.glm.regression.RegressionResults(
glm['data'][channel]['theta'],
glm['data'][channel]['Y'],
glm['data'][channel]['model'],
glm['data'][channel]['whitened_Y'],
glm['data'][channel]['whitened_residuals'],
cov=glm['data'][channel]['cov']
)
for key in glm['data'][channel]:
res.__setattr__(key, glm['data'][channel][key])
glm['data'][channel] = res
# Ensure order of dictionary matches info
data = {k: glm['data'][k] for k in glm['info']['ch_names']}
return RegressionResults(Info(glm['info']), data, glm['design'])
elif glm['classname'] == "<class 'mne_nirs.statistics._glm_level_first" \
".ContrastResults'>":
data = nilearn.glm.contrasts.Contrast(glm['data']['effect'],
glm['data']['variance'])
for key in glm['data']:
data.__setattr__(key, glm['data'][key])
return ContrastResults(Info(glm['info']), data, glm['design'])
else:
raise IOError('Unable to read data')