Compute source level time-frequency timecourses using a DICS beamformer

In this example, a Dynamic Imaging of Coherent Sources (DICS) [1] beamformer is used to transform sensor-level time-frequency objects to the source level. We will look at the event-related synchronization (ERS) of beta band activity in the somato dataset.

# Authors: Marijn van Vliet <w.m.vanvliet@gmail.com>
#          Alex Rockhill <aprockhill@mailbox.org>
#
# License: BSD-3-Clause
# Copyright the MNE-Python contributors.

import numpy as np

import mne
from mne.beamformer import apply_dics_tfr_epochs, make_dics
from mne.datasets import somato
from mne.time_frequency import csd_tfr

print(__doc__)

Organize the data that we will use for this example.

data_path = somato.data_path()
subject = "01"
task = "somato"
raw_fname = data_path / f"sub-{subject}" / "meg" / f"sub-{subject}_task-{task}_meg.fif"
fname_fwd = (
    data_path / "derivatives" / f"sub-{subject}" / f"sub-{subject}_task-{task}-fwd.fif"
)
subjects_dir = data_path / "derivatives" / "freesurfer" / "subjects"

First, we load the data and compute for each epoch the time-frequency decomposition in sensor space.

# Load raw data and make epochs.
raw = mne.io.read_raw_fif(raw_fname)
events = mne.find_events(raw)
epochs = mne.Epochs(
    raw,
    events[:22],  # just for execution speed of the tutorial
    event_id=1,
    tmin=-1,
    tmax=2.5,
    reject=dict(
        grad=5000e-13,  # unit: T / m (gradiometers)
        mag=5e-12,  # unit: T (magnetometers)
        eog=250e-6,  # unit: V (EOG channels)
    ),
    preload=True,
)

# We are mostly interested in the beta band since it has been shown to be
# active for somatosensory stimulation
freqs = np.linspace(13, 31, 5)

# Use Morlet wavelets to compute sensor-level time-frequency (TFR)
# decomposition for each epoch. We must pass ``output='complex'`` if we wish to
# use this TFR later with a DICS beamformer. We also pass ``average=False`` to
# compute the TFR for each individual epoch.
epochs_tfr = epochs.compute_tfr(
    "morlet", freqs, n_cycles=5, return_itc=False, output="complex", average=False
)

# crop either side to use a buffer to remove edge artifact
epochs_tfr.crop(tmin=-0.5, tmax=2)

Now, we build a DICS beamformer and project the sensor-level TFR to the source level.

# Compute the Cross-Spectral Density (CSD) matrix for the sensor-level TFRs.
# We are interested in increases in power relative to the baseline period, so
# we will make a separate CSD for just that period as well.
csd = csd_tfr(epochs_tfr, tmin=-0.5, tmax=2)
baseline_csd = csd_tfr(epochs_tfr, tmin=-0.5, tmax=-0.1)

# use the CSDs and the forward model to build the DICS beamformer
fwd = mne.read_forward_solution(fname_fwd)

# compute scalar DICS beamfomer
filters = make_dics(
    epochs.info,
    fwd,
    csd,
    noise_csd=baseline_csd,
    pick_ori="max-power",
    reduce_rank=True,
    real_filter=True,
)

# project the TFR for each epoch to source space
epochs_stcs = apply_dics_tfr_epochs(epochs_tfr, filters, return_generator=True)

# average across frequencies and epochs
data = np.zeros((fwd["nsource"], epochs_tfr.times.size))
for epoch_stcs in epochs_stcs:
    for stc in epoch_stcs:
        data += (stc.data * np.conj(stc.data)).real

stc.data = data / len(epochs) / len(freqs)

# apply a baseline correction
stc.apply_baseline((-0.5, -0.1))

Let’s visualize the source time course estimate. We can see the expected activation of the two gyri bordering the central sulcus, the primary somatosensory and motor cortices (S1 and M1).

fmax = 4500
brain = stc.plot(
    subjects_dir=subjects_dir,
    hemi="both",
    views="dorsal",
    initial_time=1.2,
    brain_kwargs=dict(show=False),
    add_data_kwargs=dict(
        fmin=fmax / 10,
        fmid=fmax / 2,
        fmax=fmax,
        scale_factor=0.0001,
        colorbar_kwargs=dict(label_font_size=10),
    ),
)

Estimated memory usage: 0 MB