36
D.-W. Kim and C.-H. Im
analysis is still used routinely in neurology and sleep studies, mainly to find abnormal
signal patterns or roughly categorize well-known states of the brain that can be easily
recognized by visual inspection [5, 18]. For instance, clinicians determine the origin
of epileptic activity from ictal EEG recordings acquired during epileptic seizure [19,
38]. Another good example of qualitative EEG analysis is EEG-based sleep stage
scoring, using visually evident patterns in the signal (e.g., slowing rhythm, existence
of K-complex or spindle) [15, 37]. Naturally, the experience and expertise of the rater
may influence the outcome, and this motivated researchers to develop less subjective
measures. Quantitative analysis uses mathematical and statistical methods to find
evident features to characterize the given EEG signal, even ones that are difficult
to detect by visual inspection. Each approach classifies the EEG signal in terms
of frequency or period, amplitude, phase relations, and morphology (waveform,
topology, abundance, reactivity, and variability of these parameters) [7].
Among the various quantitative features of EEGs, one of the basic and common features is the frequency power of the EEG signal. From the earliest stages
of EEG research, probably from the first report of the existence of alpha waves by
Hans Berger, different velocities of the EEG recording were believed to reflect different states of the brain. Based on extensive research, now it is widely accepted
that we can define distinct frequency bands with different roles and characteristics,
namely, the delta (1–3 Hz), theta (4–7 Hz), alpha (8–12 Hz), beta (15–30 Hz), and
gamma (30–100 Hz) frequency bands. The frequency ranges of each band slightly
vary among different studies and researchers; however, it is generally accepted that a
subtle difference (less than 1 Hz) in defining the frequency range does not make
a significant difference [29]. Normally, the lower frequency band is believed to
reflect subconscious states, mostly showing dominant activation during deep sleep
or drowsiness, while relatively higher frequency bands reflect more alerted, active
states or are associated with higher cognitive functions [20, 35].
Two important factors are mainly considered in traditional EEG spectral analysis:
the amount (usually reported as power) of a specific frequency band and its spatial
distribution—see Fig. 3.1a. The power represents the amount of the frequency band
included in the signal, where both increase and decrease of EEG power are meaningful information to understand the underlying brain function. The spatial distribution
of the power is also considered to be crucial, because the power changes in different brain areas may represent different processes of the brain. EEG spectral powers
are generally represented as topological distributions on the scalp surface, and this
is usually referred to as quantitative EEG (qEEG) analysis, allowing for intuitive
comparison among groups or conditions in clinical applications. For instance, traditional qEEG analysis was done by comparing the individual data with a normative
database [22, 23]. The normative database basically contains frequency band power
data carefully collected by several hundreds of healthy subjects with diverse ages.
The spectral power of the individual EEG is usually converted to z-score, which
highlights whether the spectral powers of any EEG channel are enhanced or reduced
compared with the normative database.
In the initial stage of development of spectral analysis methods, it was not easy
to track the temporal variations of the frequency spectrum, because the temporo-
D.-W. Kim and C.-H. Im
analysis is still used routinely in neurology and sleep studies, mainly to find abnormal
signal patterns or roughly categorize well-known states of the brain that can be easily
recognized by visual inspection [5, 18]. For instance, clinicians determine the origin
of epileptic activity from ictal EEG recordings acquired during epileptic seizure [19,
38]. Another good example of qualitative EEG analysis is EEG-based sleep stage
scoring, using visually evident patterns in the signal (e.g., slowing rhythm, existence
of K-complex or spindle) [15, 37]. Naturally, the experience and expertise of the rater
may influence the outcome, and this motivated researchers to develop less subjective
measures. Quantitative analysis uses mathematical and statistical methods to find
evident features to characterize the given EEG signal, even ones that are difficult
to detect by visual inspection. Each approach classifies the EEG signal in terms
of frequency or period, amplitude, phase relations, and morphology (waveform,
topology, abundance, reactivity, and variability of these parameters) [7].
Among the various quantitative features of EEGs, one of the basic and common features is the frequency power of the EEG signal. From the earliest stages
of EEG research, probably from the first report of the existence of alpha waves by
Hans Berger, different velocities of the EEG recording were believed to reflect different states of the brain. Based on extensive research, now it is widely accepted
that we can define distinct frequency bands with different roles and characteristics,
namely, the delta (1–3 Hz), theta (4–7 Hz), alpha (8–12 Hz), beta (15–30 Hz), and
gamma (30–100 Hz) frequency bands. The frequency ranges of each band slightly
vary among different studies and researchers; however, it is generally accepted that a
subtle difference (less than 1 Hz) in defining the frequency range does not make
a significant difference [29]. Normally, the lower frequency band is believed to
reflect subconscious states, mostly showing dominant activation during deep sleep
or drowsiness, while relatively higher frequency bands reflect more alerted, active
states or are associated with higher cognitive functions [20, 35].
Two important factors are mainly considered in traditional EEG spectral analysis:
the amount (usually reported as power) of a specific frequency band and its spatial
distribution—see Fig. 3.1a. The power represents the amount of the frequency band
included in the signal, where both increase and decrease of EEG power are meaningful information to understand the underlying brain function. The spatial distribution
of the power is also considered to be crucial, because the power changes in different brain areas may represent different processes of the brain. EEG spectral powers
are generally represented as topological distributions on the scalp surface, and this
is usually referred to as quantitative EEG (qEEG) analysis, allowing for intuitive
comparison among groups or conditions in clinical applications. For instance, traditional qEEG analysis was done by comparing the individual data with a normative
database [22, 23]. The normative database basically contains frequency band power
data carefully collected by several hundreds of healthy subjects with diverse ages.
The spectral power of the individual EEG is usually converted to z-score, which
highlights whether the spectral powers of any EEG channel are enhanced or reduced
compared with the normative database.
In the initial stage of development of spectral analysis methods, it was not easy
to track the temporal variations of the frequency spectrum, because the temporo-