Extracting and visualizing subject head movement

Continuous head movement can be encoded during MEG recordings by use of HPI coils that continuously emit sinusoidal signals. These signals can then be extracted from the recording and used to estimate head position as a function of time. Here we show an example of how to do this, and how to visualize the result.

HPI frequencies

First let’s load a short bit of raw data where the subject intentionally moved their head during the recording. Its power spectral density shows five peaks (most clearly visible in the gradiometers) corresponding to the HPI coil frequencies, plus other peaks related to power line interference (60 Hz and harmonics).

# Authors: Eric Larson <larson.eric.d@gmail.com>
#          Richard Höchenberger <richard.hoechenberger@gmail.com>
#          Daniel McCloy <dan@mccloy.info>
#
# License: BSD-3-Clause
# Copyright the MNE-Python contributors.

from os import path as op

import mne

data_path = op.join(mne.datasets.testing.data_path(verbose=True), "SSS")
fname_raw = op.join(data_path, "test_move_anon_raw.fif")
raw = mne.io.read_raw_fif(fname_raw, allow_maxshield="yes").load_data()
raw.compute_psd().plot(picks="data", exclude="bads", amplitude=False)

We can use mne.chpi.get_chpi_info to retrieve the coil frequencies, the index of the channel indicating when which coil was switched on, and the respective “event codes” associated with each coil’s activity.

chpi_freqs, ch_idx, chpi_codes = mne.chpi.get_chpi_info(info=raw.info)
print(f"cHPI coil frequencies extracted from raw: {chpi_freqs} Hz")

Estimating continuous head position

First, let’s extract the HPI coil amplitudes as a function of time:

chpi_amplitudes = mne.chpi.compute_chpi_amplitudes(raw)

Second, let’s compute time-varying HPI coil locations from these:

chpi_locs = mne.chpi.compute_chpi_locs(raw.info, chpi_amplitudes)

Lastly, compute head positions from the coil locations:

head_pos = mne.chpi.compute_head_pos(raw.info, chpi_locs, verbose=True)

Note that these can then be written to disk or read from disk with mne.chpi.write_head_pos() and mne.chpi.read_head_pos(), respectively.

Visualizing continuous head position

We can plot as traces, which is especially useful for long recordings:

mne.viz.plot_head_positions(head_pos, mode="traces", totals=True)

Or we can visualize them as a continuous field (with the vectors pointing in the head-upward direction):

mne.viz.plot_head_positions(head_pos, mode="field")

These head positions can then be used with mne.preprocessing.maxwell_filter() to compensate for movement, or with mne.preprocessing.annotate_movement() to mark segments as bad that deviate too much from the average head position.

Computing SNR of the HPI signal

It is also possible to compute the SNR of the continuous HPI measurements. This can be a useful proxy for head position along the vertical dimension, i.e., it can indicate the distance between the HPI coils and the MEG sensors. Using compute_chpi_snr, the HPI power and SNR are computed separately for each MEG sensor type and each HPI coil (frequency), along with the residual power for each sensor type. The results can then be visualized with plot_chpi_snr. Here we’ll just show a few seconds, for speed:

raw.crop(tmin=5, tmax=10)
snr_dict = mne.chpi.compute_chpi_snr(raw)
fig = mne.viz.plot_chpi_snr(snr_dict)

Estimated memory usage: 0 MB