Biomedical Engineering Reference
In-Depth Information
14.3.3 Analysis of Nonphase-Locked Responses in iEEG
Compared to both scalp EEG and iEEG studies of phase-locked responses (ERPs),
studies of nonphase-locked responses have been more recent. Early studies of
nonphase-locked EEG activity focused on event-related power changes in relatively
narrow frequency bands [27-29]. These studies primarily addressed well-defined
narrowband oscillations such as the occipital alpha rhythm and the sensorimotor
mu rhythm. Such prominent oscillations were assumed to require the synchronized
activity of a large population of cortical neurons. Indeed, any potentials visible at
the scalp likely require the summation of dendritic membrane potentials in a large
population of neurons, and temporal synchronization at some spatial scale must be
present for substantial summation to occur. Thus, event-related power changes in
narrow frequency bands were thought to reflect the degree of synchronization
among oscillating elements in cortical networks, and the terms
event-related
desynchronization
(ERD) and
event-related synchronization
(ERS) were coined to
refer to suppression or augmentation, respectively, of power in a particular fre-
quency range [27]. More recent studies have sometimes avoided this terminology
because it may not accurately reflect the neurophysiological mechanisms underlying
all phenomena observed with this approach. Furthermore, significant event-related
power changes are not limited to narrowband oscillations, as in alpha ERD. Never-
theless, the term
ERD/ERS
is still often used as a convenient, easily articulated, and
widely recognized “nickname” for changes in signal energy that are time locked,
but not necessarily phase locked, to an event and that are derived by the same
general approach to signal analysis.
Most analyses of ERD/ERS in scalp EEG and iEEG no longer rely on quantifica-
tion of signal energy in narrow frequency bands. Bandpass filtering requires a priori
knowledge of the frequency bands where event-related signal changes will occur.
There is not only considerable variability in these reactive frequency bands across
subjects [30, 31], but there may also be significant variability across recording sites
and functional tasks [32]. Although reactive frequency bands can be determined
empirically by comparing power spectra in activated versus baseline EEG epochs,
this requires a priori knowledge of the timing of functional brain activation. For
these reasons, time-frequency analysis
is the new standard for analyses
of
event-related EEG energy changes.
Time-frequency analysis consists of two basic steps: (1) time-frequency decom-
position of the EEG signal, and (2) statistical analysis of event-related energy
changes in time-frequency space. Both steps can be accomplished by a variety of
methods.
14.3.3.1 Time-Frequency Decomposition
The methods that have been used for time-frequency decomposition of EEG signals
include Fourier transformation, wavelet transformation, complex demodulation,
and matching pursuit decomposition, among others. For reasons discussed later,
our laboratory has chosen to use matching pursuits with a dictionary of Gabor
functions.
The time-frequency representation of the signal can be interpreted either as the
integral transformation of the time series to time-frequency space or as a decompo-
Search WWH ::
Custom Search