211
Electroencephalogram
Other sources of error can be introduced by the leads acting as antennas picking
up ambient alternating signals through induction. Sources include the socket power,
switching of equipment, and motion by the technicians in the earth’s magnetic field.
An often overlooked but significant source of error is the saline drip from the intravenous line. The salt solution may cause spikes in the recordings when dripping.
The muscle contractions of the face caused by blinking, the chest motion due to
respiration, and the ECG are all significant sources of fluctuating electric potential
that cannot be simply ruled out and need to be filtered during the measurement of
the EEG. The muscle signal artifacts are generally characterized as relatively high
frequency variations.
Filtering specific frequency bands from the EEG can be used to reduce the influence of muscle activities as well as other sources of noise listed earlier. This is
explained in more detail next.
10.7.2 FREQUENCY-DOMAIN ANALYSIS
The use of a low-pass filter with a cutoff frequency around 12.5 Hz is necessary to
ensure that the residual muscle activities do not interfere with EEGs beta activity.
However, this is not desirable in most recordings, since the true beta activity and
spike-type activity will also be attenuated or even obscured. In case the interference
from the external source noise is persistent, some form of filtering will have to be
incorporated. An adaptive digital notch filter, which allows all frequencies to pass
through except for the frequencies in the narrow band of the interfering noise, is
often preferred over a simple low-pass filter.
Conversely, when only a specific frequency range of the EEG requires particular
interest, a band-pass filter can be applied to filter out low-frequency muscle activity and high-frequency interference from instrumentation to extract the pure EEG
signal. The high-frequency components to be filter out include the power supply’s
frequency (50 or 60 Hz) and the high-frequency muscle activity.
A main method of EEG feature extraction in the Fourier domain is evaluation
of the power of specific frequencies in the power spectra of the signal. A full frequency spectrum recorded from all over the head with 64 electrode placements
during a 1 s segment is shown in Figure 10.7. As mentioned before, due to the noisy
nature of the EEG, it is often preferred to treat EEG as a stochastic process, and as
a result, the frequency analysis of the signal is performed using the power spectra
of EEG. The frequency spectrum of the EEG signal can be easily computed by
taking the DFT of the EEG correlation function. While an equipment artifact that
operates in a specific frequency range may not reveal itself clearly in unprocessed
EEG traces, a spectral analysis will quickly reveal any irregular pattern of higher
harmonics in the frequency spectrum.
The EEG spectrum is often analyzed only over consecutive short-time segment.
The short-time interval of frequency analysis is called an “epoch.” The longer the
selected epochs targeted for frequency transformation, the better the frequency resolution. However, there is a trade-off in taking longer time segments since they will
result in a lower time resolution. The time resolution can be improved by shifting
epochs forward over the chosen time segment. The best accuracy in time resolution
Electroencephalogram
Other sources of error can be introduced by the leads acting as antennas picking
up ambient alternating signals through induction. Sources include the socket power,
switching of equipment, and motion by the technicians in the earth’s magnetic field.
An often overlooked but significant source of error is the saline drip from the intravenous line. The salt solution may cause spikes in the recordings when dripping.
The muscle contractions of the face caused by blinking, the chest motion due to
respiration, and the ECG are all significant sources of fluctuating electric potential
that cannot be simply ruled out and need to be filtered during the measurement of
the EEG. The muscle signal artifacts are generally characterized as relatively high
frequency variations.
Filtering specific frequency bands from the EEG can be used to reduce the influence of muscle activities as well as other sources of noise listed earlier. This is
explained in more detail next.
10.7.2 FREQUENCY-DOMAIN ANALYSIS
The use of a low-pass filter with a cutoff frequency around 12.5 Hz is necessary to
ensure that the residual muscle activities do not interfere with EEGs beta activity.
However, this is not desirable in most recordings, since the true beta activity and
spike-type activity will also be attenuated or even obscured. In case the interference
from the external source noise is persistent, some form of filtering will have to be
incorporated. An adaptive digital notch filter, which allows all frequencies to pass
through except for the frequencies in the narrow band of the interfering noise, is
often preferred over a simple low-pass filter.
Conversely, when only a specific frequency range of the EEG requires particular
interest, a band-pass filter can be applied to filter out low-frequency muscle activity and high-frequency interference from instrumentation to extract the pure EEG
signal. The high-frequency components to be filter out include the power supply’s
frequency (50 or 60 Hz) and the high-frequency muscle activity.
A main method of EEG feature extraction in the Fourier domain is evaluation
of the power of specific frequencies in the power spectra of the signal. A full frequency spectrum recorded from all over the head with 64 electrode placements
during a 1 s segment is shown in Figure 10.7. As mentioned before, due to the noisy
nature of the EEG, it is often preferred to treat EEG as a stochastic process, and as
a result, the frequency analysis of the signal is performed using the power spectra
of EEG. The frequency spectrum of the EEG signal can be easily computed by
taking the DFT of the EEG correlation function. While an equipment artifact that
operates in a specific frequency range may not reveal itself clearly in unprocessed
EEG traces, a spectral analysis will quickly reveal any irregular pattern of higher
harmonics in the frequency spectrum.
The EEG spectrum is often analyzed only over consecutive short-time segment.
The short-time interval of frequency analysis is called an “epoch.” The longer the
selected epochs targeted for frequency transformation, the better the frequency resolution. However, there is a trade-off in taking longer time segments since they will
result in a lower time resolution. The time resolution can be improved by shifting
epochs forward over the chosen time segment. The best accuracy in time resolution