50
D.-W. Kim and C.-H. Im
each frequency band by dividing the absolute power of each band by the total power
(usually, 1–50/60 Hz) or by the sum of powers in the frequency bands of interest. The
relative power represents the proportion of each band to the given signal. According
to Klimesch [24], the relative power of the alpha band highlighted the increase of
alpha power in children with respect to age, although it was hard to observe such a
trend in absolute power.
In comparison with absolute power, relative power could be an intuitive indicator
to track changes in overall dominance of each band over time or under different
conditions. However, some researchers insist that relative power may lead to misinterpretation of the data. Indeed, an increase of relative power in one frequency
band could be interpreted as a decrease of relative power in another band, even when
the absolute power of the latter frequency band did not change [8]. In addition, the
definition of frequency bands is not standardized, making it difficult to compare
the analysis results among studies. Nevertheless, both absolute and relative powers
contributed to many important findings in the field of neuroscience; therefore, they
should be treated equally.
3.3.6 Other Considerations in Spectral Analysis
There is no restriction in the number of electrodes used for spectral analysis; however,
at least 19 (according to the international 10–20 system) or more electrodes are
recommended to observe the overall spatial distribution of spectral power for each
frequency band [2]. It is common to use reference electrodes placed at electrically
neutral places, such as nose tip reference and linked-ear reference; however, some
researchers prefer to use reference-free methods, such as common average reference
(CAR). Physical reference electrodes often suffer from local contamination but have
an advantage that each electrode can be treated independently. On the other hand,
CAR generally requires a number of evenly distributed electrodes on the scalp surface
to fulfill its mathematical assumption [11]. Moreover, CAR is sensitive to an artifact
(e.g., eye blink artifact and electromyogram artifact), because the artifact is evenly
distributed to other electrodes. Therefore, if one decides to use CAR, it is important to
reject epochs including artifacts or remove/reduce the artifacts before the application
of CAR using signal processing procedures, e.g., independent component analysis
(ICA).
In practical EEG analysis, it is a widely accepted process to group a few adjacent electrodes within a specific region of interest and calculate the average spectral
power of the region. Division of the regions is generally made according to hemispheres (left/right) or lobes (frontal/central/temporal/occipital) or by combining both.
Grouping the electrodes by regions or hemispheres can better highlight regional differences—for instance, alpha asymmetry [16] of left and right hemispheres. Averaging the spectral power in small areas might be reasonable, considering the low
spatial resolution of EEGs. Moreover, grouping nearby electrodes is advantageous
not only to increase the reliability of the power spectrum by averaging, but also
D.-W. Kim and C.-H. Im
each frequency band by dividing the absolute power of each band by the total power
(usually, 1–50/60 Hz) or by the sum of powers in the frequency bands of interest. The
relative power represents the proportion of each band to the given signal. According
to Klimesch [24], the relative power of the alpha band highlighted the increase of
alpha power in children with respect to age, although it was hard to observe such a
trend in absolute power.
In comparison with absolute power, relative power could be an intuitive indicator
to track changes in overall dominance of each band over time or under different
conditions. However, some researchers insist that relative power may lead to misinterpretation of the data. Indeed, an increase of relative power in one frequency
band could be interpreted as a decrease of relative power in another band, even when
the absolute power of the latter frequency band did not change [8]. In addition, the
definition of frequency bands is not standardized, making it difficult to compare
the analysis results among studies. Nevertheless, both absolute and relative powers
contributed to many important findings in the field of neuroscience; therefore, they
should be treated equally.
3.3.6 Other Considerations in Spectral Analysis
There is no restriction in the number of electrodes used for spectral analysis; however,
at least 19 (according to the international 10–20 system) or more electrodes are
recommended to observe the overall spatial distribution of spectral power for each
frequency band [2]. It is common to use reference electrodes placed at electrically
neutral places, such as nose tip reference and linked-ear reference; however, some
researchers prefer to use reference-free methods, such as common average reference
(CAR). Physical reference electrodes often suffer from local contamination but have
an advantage that each electrode can be treated independently. On the other hand,
CAR generally requires a number of evenly distributed electrodes on the scalp surface
to fulfill its mathematical assumption [11]. Moreover, CAR is sensitive to an artifact
(e.g., eye blink artifact and electromyogram artifact), because the artifact is evenly
distributed to other electrodes. Therefore, if one decides to use CAR, it is important to
reject epochs including artifacts or remove/reduce the artifacts before the application
of CAR using signal processing procedures, e.g., independent component analysis
(ICA).
In practical EEG analysis, it is a widely accepted process to group a few adjacent electrodes within a specific region of interest and calculate the average spectral
power of the region. Division of the regions is generally made according to hemispheres (left/right) or lobes (frontal/central/temporal/occipital) or by combining both.
Grouping the electrodes by regions or hemispheres can better highlight regional differences—for instance, alpha asymmetry [16] of left and right hemispheres. Averaging the spectral power in small areas might be reasonable, considering the low
spatial resolution of EEGs. Moreover, grouping nearby electrodes is advantageous
not only to increase the reliability of the power spectrum by averaging, but also