The Glossary provides short definitions of vocabulary specific to MNE-Python and general neuroimaging concepts. If you think a term is missing, please consider creating a new issue or opening a pull request to add it.
An annotation is defined by an onset, a duration, and a textual
description. It can contain information about the experiment, but
also details on signals marked by a human such as bad data segments,
sleep stages, sleep events (spindles, K-complex), and so on.
An Annotations
object is a container for multiple annotations,
which is available as the annotations
attribute of Raw
objects. See Annotations
for the class definition and
Parsing events from raw data for a short tutorial.
See also events.
Something that acts like – or can be converted to – a
NumPy array
.
This includes (but is not limited to)
arrays
, lists
, and
tuples
.
A beamformer is a popular source estimation approach that uses a set of
spatial filters (beamformer weights) to compute time courses of sources
at predefined locations. See beamformer.Beamformer
for the class
definition. See also LCMV.
BEM is the acronym for boundary element method or boundary element
model. Both are related to the definion of the conductor model in the
forward model computation. The boundary element model consists of surfaces
such as the inner skull, outer skull, and outer skin (scalp) that define
compartments of tissues of the head. You can compute the BEM surfaces with
bem.make_watershed_bem()
or bem.make_flash_bem()
.
See Head model and forward computation for a usage demo.
Channels refer to MEG sensors, EEG electrodes or other sensors such as EOG, ECG, sEEG, ECoG, etc. Channels usually have a type (such as gradiometer), and a unit (such as T/m) used e.g. for plotting. See also data channels.
Many functions in MNE-Python operate on “data channels” by default. These are channels that contain electrophysiological data from the brain, as opposed to other channel types such as EOG, ECG, stimulus/trigger, or acquisition system status data. The set of channels considered “data channels” in MNE contains the following types (together with scale factors for plotting):
'mag'
: Magnetometers (scaled by 1e+15 to plot in fT)
'grad'
: Gradiometers (scaled by 1e+13 to plot in fT/cm)
'eeg'
: EEG (scaled by 1e+06 to plot in µV)
'csd'
: Current source density (scaled by 1000 to plot in mV/m²)
'seeg'
: sEEG (scaled by 1000 to plot in mV)
'ecog'
: ECoG (scaled by 1e+06 to plot in µV)
'dbs'
: DBS (scaled by 1e+06 to plot in µV)
'hbo'
: Oxyhemoglobin (scaled by 1e+06 to plot in µM)
'hbr'
: Deoxyhemoglobin (scaled by 1e+06 to plot in µM)
'fnirs_cw_amplitude'
: fNIRS (CW amplitude) (scaled by 1 to plot in V)
'fnirs_fd_ac_amplitude'
: fNIRS (FD AC amplitude) (scaled by 1 to plot in V)
'fnirs_fd_phase'
: fNIRS (FD phase) (scaled by 1 to plot in rad)
'fnirs_od'
: fNIRS (OD) (scaled by 1 to plot in V)
Dynamic Imaging of Coherent Sources is a method for computing source
power in different frequency bands. See Compute source power using DICS beamformer
and beamformer.make_dics()
for more details.
Digitization is a procedure of recording the head shape and locations of fiducial coils (or HPI) and/or EEG electrodes on the head. They are represented as a set of points in 3D space. See Reading sensor digitization files and Supported formats for digitized 3D locations.
An equivalent current dipole (ECD) is an approximate representation of post-synaptic activity in a small cortical region. The intracellular currents that give rise to measurable EEG/MEG signals are thought to originate in populations of cortical pyramidal neurons aligned perpendicularly to the cortical surface. Because the length of such current sources is very small relative to the distance between the cortex and the EEG/MEG sensors, the fields measured by these techniques are well approximated by (i.e., equivalent to) fields generated by idealized point sources (dipoles) located on the cortical surface.
Dynamic statistical parametric mapping (dSPM) gives a noise-normalized minimum-norm estimate at a given source location. It is calculated by dividing the activity estimate at each source location by the baseline standard deviation of the noise.
eLORETA and sLORETA (exact and standardized low resolution brain
electromagnetic tomography) are linear source estimation techniques
like dSPM and MNE. sLORETA outputs
standardized values (like dSPM), while eLORETA generates normalized
current estimates. See minimum_norm.apply_inverse()
,
Source localization with MNE, dSPM, sLORETA, and eLORETA, and Compute sLORETA inverse solution on raw data.
Epochs (sometimes called “trials” in other software packages) are
equal-length segments of data extracted from continuous data. Usually,
epochs are extracted around stimulus events or responses,
though sometimes sequential or overlapping epochs are used (e.g.,
for analysis of resting-state activity). See Epochs
for the
class definition and The Epochs data structure: discontinuous data for a narrative overview.
Events correspond to specific time points in raw data, such as triggers,
experimental condition events, etc. MNE-Python represents events with
integers stored in NumPy arrays of shape (n_events, 3)
. The first
column contains the event onset (in samples) with first_samp
included. The last column contains the event code. The second
column contains the signal value of the immediately preceding sample,
and reflects the fact that event arrays sometimes originate from
analog voltage channels (“trigger channels” or “stim channels”). In
most cases, the second column is all zeros and can be ignored.
Event arrays can be created with mne.make_fixed_length_events()
,
mne.read_events()
, and mne.find_events()
.
See Parsing events from raw data for a short tutorial.
See also annotations.
Evoked data are obtained by averaging epochs. Typically, an evoked object
is constructed for each subject and each condition, but it can also be
obtained by averaging a list of evoked objects over different subjects.
See EvokedArray
for the class definition and
The Evoked data structure: evoked/averaged data for a narrative overview.
Fiducials are objects placed in the field of view of an imaging system to act as known spatial references that are easy to localize. In neuroimaging, fiducials are often placed on anatomical landmarks such as the nasion (NAS) or left/right preauricular points (LPA and RPA).
These known reference locations are used to define a coordinate system for localizing sensors (hence NAS, LPA and RPA are often called “cardinal points” because they define the cardinal directions of the head coordinate system). The cardinal points are also useful when co-registering measurements in different coordinate systems (such as aligning EEG sensor locations to an MRI of the head).
Due to the common neuroimaging practice of placing fiducial objects on anatomical landmarks, the terms “fiducial”, “anatomical landmark”, and “cardinal point” are often (erroneously) used interchangeably.
The first_samp
attribute of Raw
objects is an integer representing the number of time samples that
passed between the onset of the hardware acquisition system and the
time when data recording started. This approach to sample
numbering is a peculiarity of VectorView MEG systems, but for
consistency it is present in all Raw
objects
regardless of the source of the data. In other words,
first_samp
will be 0
in Raw
objects loaded from non-VectorView data files. See also
last_samp.
The forward solution is a linear operator capturing the
relationship between each dipole location in the source space
and the corresponding field distribution measured by the sensors
(the “lead field matrix”). Calculating a forward solution requires a
conductivity model of the head, which encapsulates the geometries and
electrical conductivities of the different tissue compartments (see
boundary element model and bem.ConductorModel
).
For information about the Forward object and the data it stores, see
mne.Forward
.
A FreeSurfer lookup table (LUT) provides a mapping between a given
volumetric atlas or surface label name, its integer value
(e.g., in aparc+aseg.mgz
), and its standard color (see the
FreeSurfer wiki
for more information). Custom LUTs can be also be created from different
surface parcellations, see for example this comment about HCPMMP.
Global Field Power (GFP) is a measure of the (non-)uniformity of the electromagnetic field at the sensors. It is typically calculated as the standard deviation of the sensor values at each time point. Thus, it is a one-dimensional time series capturing the spatial variability of the signal across sensor locations.
Hierarchical event descriptors (HED) are tags that use
keywords separated by slashes (/) to describe different types of
experimental events (for example, stimulus/circle/red/left
and
stimulus/circle/blue/left
). These tags can be used to group
experimental events and select event types for analysis.
Head position indicators (HPI, sometimes cHPI for continuous head position indicators) are small coils attached to a subject’s head during MEG acquisition. Each coil emits a sinusoidal signal of a different frequency, which is picked up by the MEG sensors and can be used to infer the head position. With cHPI, the sinusoidal signals are typically set at frequencies above any neural signal of interest, and thus can be removed after head position correction via low-pass filtering. See Extracting and visualizing subject head movement.
A “measurement info” (or short “info”) object is a collection of metadata
related to Raw
, Epochs
, or Evoked
objects. It contains channel locations and types, sampling frequency,
preprocessing history such as filters, etc.
See The Info data structure for a narrative overview.
The inverse operator is an \(M \times N\) matrix (\(M\) source
locations by \(N\) sensors) that, when applied to the sensor
signals, yields estimates of the brain activity that gave rise to the
observed sensor signals. Inverse operators are available for the linear
inverse methods MNE, dSPM, sLORETA, and
eLORETA. See minimum_norm.apply_inverse()
.
A Label
refers to a defined region in the cortex, often called
a region of interest (ROI) in the literature. Labels can be defined
anatomically (based on the physical structure of the cortex) or functionally
(based on cortical responses to specific stimuli). See also ROI.
The last_samp
attribute of Raw
objects is an integer representing the number of time samples that
passed between the start and end of data recording. This approach to sample
numbering is a peculiarity of VectorView MEG systems, but for
consistency it is present in all Raw
objects
regardless of the source of the data. See also first_samp.
A Layout
gives sensor positions in two
dimensions (defined by x
, y
, width
, and height
values for
each sensor). It is primarily used for illustrative purposes (i.e., making
diagrams of approximate sensor positions in cartoons of the head,
so-called topographies or topomaps). See also montage.
Linearly constrained minimum variance beamformer attempt to
estimate activity for a given source while suppressing cross-talk from
other regions (beamformer.make_lcmv()
). See also
beamformer.
A method to display pixel-wise activity within some volume by finding the maximum value along a vector from the viewer to the pixel (i.e., along the vector pependicular to the view plane).
Minimum-norm estimation (MNE) can be used to generate a distributed map of activation on a source space (usually on a cortical surface). MNE uses a linear inverse operator to project sensor measurements into the source space. The inverse operator is computed from the forward solution for a subject and an estimate of the noise covariance of sensor measurements.
EEG channel names and relative positions of sensors on the scalp.
While layouts are 2D locations, montages are 3D locations. A montage
can also contain locations for HPI points, fiducial points, or
extra head shape points.
See DigMontage
for the class definition. See also
layout.
Morphing refers to the operation of transferring source estimates from one anatomy to another. It is known as realignment in the fMRI literature. This operation is necessary for group studies to get the data into a common space for statistical analysis. See Morphing and averaging source estimates for more details.
The noise covariance is a matrix that contains the covariance between data
channels. It is a square matrix with shape n_channels
\(\times\)
n_channels
. It is especially useful when working with multiple sensor
types (e.g. EEG and MEG). In practice, the matrix is estimated from baseline
periods or empty room measurements, and it also provides a noise model
that can be used for subsequent analysis (like source imaging).
Something that acts like a path in a file system. This can be a str
or a pathlib.Path
.
An integer that is the index of a channel in the measurement info.
It allows to obtain the information on a channel in the list of channels
available in info['chs']
.
A projector, also referred to as Signal Space
Projection (SSP), defines a linear operation applied spatially to EEG
or MEG data. A matrix multiplication of an SSP projector with the data
will reduce the rank of the data by projecting it to a
lower-dimensional subspace. Such projections are typically applied to
both the data and the forward operator when performing
source localization. Note that EEG average referencing can be done
using such a projection operator. Projectors are stored alongside data
in the measurement info in the field info['projs']
.
Right-Anterior-Superior, denoting the standard way to define coordinate frames in MNE-Python:
+X is right, -X is left
+Y is anterior (front), -Y is posterior (rear)
+Z is superior (top), -Z is inferior (bottom)
Raw
objects hold continuous data (preprocessed or not), typically
obtained from reading recordings stored in a file.
See RawArray
for the class definition and The Raw data structure: continuous data
for a narrative overview.
A spatial region where an experimental effect is expected to manifest. This can be a collection of sensors or, when performing inverse imaging, a set of vertices on the cortical surface or within the cortical volume. See also label.
A selection is a set of picked channels (for example, all sensors falling within a region of interest).
A source space specifies where in the brain source amplitudes are
estimated. It corresponds to locations of a set of
candidate equivalent current dipoles. MNE-Python mostly
works with source spaces defined on the cortical surfaces estimated
by FreeSurfer from a T1-weighted MRI image. See Head model and forward computation
to read about how to compute a forward operator in a source space.
See SourceSpaces
for the class definition and information
about the data it contains.
Source estimates, commonly referred to as STC (Source Time Courses),
are obtained from source localization methods such as dSPM,
sLORETA, LCMV, or MxNE.
STCs contain the amplitudes of the neural sources over time.
In MNE-Python, SourceEstimate
objects only store the
amplitudes of activation but not the locations of the sources. The
locations are stored separately in the SourceSpaces
object
that was used to compute the forward operator.
See SourceEstimate
, VolSourceEstimate
,
VectorSourceEstimate
, and MixedSourceEstimate
.
A stim channel or trigger channel is a channel that encodes
events during the recording. It is typically a channel that is always
zero and takes positive values when something happens (such as the
onset of a stimulus or a subject response). Stim channels are often
prefixed with STI
to distinguish them from other channel types. See
What is a STIM channel? for more details.
An idealized EEG montage, often provided by the manufacturer of the EEG system or cap. The electrode positions were not actually measured on the participants’ heads, but rather were calculated assuming optimal theoretical placement on a sphere.
A time-frequency representation (TFR) is often a spectrogram (STFT) or scaleogram (wavelet) showing the frequency content as a function of time.
A coordinate frame affine transformation, usually between the Neuromag head coordinate frame and the MRI Surface RAS coordinate frame used by Freesurfer.
A linear operation that transforms data with a known covariance structure into “whitened data”, which has a covariance structure equal to the identity matrix. In other words, whitening creates virtual channels that are uncorrelated and have unit variance. This is also known as a sphering transformation.
The term “whitening” comes from the fact that light with a flat
frequency spectrum in the visible range is white, whereas
non-uniform frequency spectra lead to perception of different colors
(e.g., “pink noise” has a 1/f
characteristic, which for visible
light would appear pink).