Biomedical Engineering Reference
In-Depth Information
Using an FFT, it is a simple matter to divide the resulting power spectrum from
an epoch of EEG into these band segments, then summate all power values for the
individual frequencies within each band to determine the “band power.” Relative
band power is simply band power divided by power over the entire frequency spec-
trum in the epoch of interest.
In the realm of anesthesia-related applications, traditional band power analysis
is of limited utility, because these bands were defined for the activity of the awake or
natural sleep-related EEG without regard for the altered nature of brain activity
during anesthesia. Drug-related EEG oscillations can often be observed to alter their
central frequency and to pass smoothly through the “classic” band boundaries as
the drug dose changes. Familiarity with band analysis is still useful, however,
because of the extensive neurological literature utilizing it.
In an effort to improve the stability of plotted band-related changes, Volgyesi
introduced the augmented delta quotient (ADQ) [48]. This value is approximately
the ratio of power in the 0.5- to 3.0-Hz band to the power in the 0.5- to 30.0-Hz
range. This definition is an approximation because the author used analog
bandpass filters with unspecified, but gentle roll-off characteristics that allowed
them to pass frequencies outside the specified band limits with relatively little atten-
uation. The ADQ was used in a single case series that was looking for cerebral
ischemia in children [49], but was never tested against other EEG parameters or
formally validated.
Jonkman et al. [50] applied a normalizing transformation [31] to render the
probability distribution of power estimates of the delta frequency range close to a
normal distribution in the CIMON EEG analysis system (Cadwell Laboratories,
Kennewick, Washington). After recording a baseline “self-norm” period of EEGs,
increases in delta-band power that are larger than three standard deviations from
the self norm were considered to represent an ischemic EEG change [51]. Other
investigators have concluded this indicator may be nonspecific [52] because it
yielded many false-positive results in control (nonischemic) patients.
Another approach to simplifying the results of a power spectral analysis is to
find a parameter that describes a particular characteristic of the spectrum distribu-
tion. The first of these descriptors was the peak power frequency (PPF), which is
simply the frequency in a spectrum at which the highest power in that epoch occurs.
The PPF has never been the subject of a clinical report. The median power frequency
(MPF) is that frequency which bisects the spectrum, with half the power above and
the other half below. There are approximately 150 publications regarding the use of
MPF in EEG monitoring. Although the MPF has been used as a feedback variable
for closed-loop control of anesthesia, there is little evidence that specific levels of
MPF correspond to specific behavioral states, that is, recall or the ability to follow
commands.
The spectral edge frequency (SEF) [53] is the highest frequency in the EEG, that
is, the high-frequency edge of the spectral distribution. The original SEF algorithm
utilized a form of pattern detection on the power spectrum in order to emulate
mechanically visual recognition of the “edge.” Beginning at 32 Hz, the power spec-
trum is scanned downward to detect the highest frequency where four sequential
spectral frequencies are above a predetermined threshold of power. This approach
provides more noise immunity than the alternative computation, SEF95 [I. J.
Search WWH ::
Custom Search