214
Biomedical Signal and Image Processing
not display any significant amount of coherence, as there is no common denominator
driving this activity. One possible explanation of the fact that the coherence analysis
of neither delta nor theta spectra reveal any useful task-related information is the fact
that the wavelength of these waves is too long to be recognized with a high enough
accuracy to produce the benefits of coincidental detection.
10.7.4 WAVELET-DOMAIN ANALYSIS
As shown in Chapter 5, wavelet techniques are typically used for the detection of
known waveform patterns against a noisy background signal.
A major application of wavelet analysis is processing of EEG by detecting spikelike epileptic patterns. The detection of epileptic pattern is of particular concern
because, in most suspected epileptic cases, these patterns appear at random and only
for brief periods of time. Due to infrequent occurrence of these patterns, Fourier
analysis can miss these patterns. In addition, since the patterns of these spikes are
often known beforehand, one can design a wavelet method to not only detect the
existence of such pattern but also identify the exact time of the occurrence of these
abnormal patterns.
An important issue in wavelet analysis of EEG is the choice of epoch length.
When analyzing a single epoch, the total number of sequences available is of direct
influence on the standardized time period testing. This influence is due to the unreliable intercorrelation between adjacent sequences. Epoch lengths of 1–2 s duration are
recommended for EEG processing. This duration guarantees a widespread stability
in the data features. This is even more important when using low sample frequencies.
The wavelet analysis is particularly important in analysis of EPs. Wavelet analysis of EP finds the typical shape of the EP pattern by finding the largest coefficients
identifying the highest correlation of decomposed signals with the pattern of applied
stimuli. In other words, when a decomposed version of the signal at a certain level
shows a high correlation with the stimuli pattern, the resulting waves in the decomposed signal identify a good estimation of the EP pattern.
Wavelet is also helpful in the determination of synchrony at the scalp. Specifically,
the determination of the exact delay between two patterns can quantitatively identify
the level of synchrony of the two waves. This decomposition is used in a number of
measures of harmonization, where the measures are based on both amplitude and
phase information.
Wavelet analysis can also be used for denoising and filtering of the EEG signal.
Often the very low scales, i.e., high frequencies, identify the additive noise, and, by
removing the low-scale components, the signal can be filtered. This was demonstrated in the chapter dedicated to wavelet analysis.
10.8 SUMMARY
In this chapter, we described the recording of nerve cell electric activity in the brain
by means of a signal called EEG. EEG can be characterized by four specific frequency ranges that roughly differentiate different mental activities. Alpha waves are
associated with sensory input and data processing. Higher frequencies in the beta
Biomedical Signal and Image Processing
not display any significant amount of coherence, as there is no common denominator
driving this activity. One possible explanation of the fact that the coherence analysis
of neither delta nor theta spectra reveal any useful task-related information is the fact
that the wavelength of these waves is too long to be recognized with a high enough
accuracy to produce the benefits of coincidental detection.
10.7.4 WAVELET-DOMAIN ANALYSIS
As shown in Chapter 5, wavelet techniques are typically used for the detection of
known waveform patterns against a noisy background signal.
A major application of wavelet analysis is processing of EEG by detecting spikelike epileptic patterns. The detection of epileptic pattern is of particular concern
because, in most suspected epileptic cases, these patterns appear at random and only
for brief periods of time. Due to infrequent occurrence of these patterns, Fourier
analysis can miss these patterns. In addition, since the patterns of these spikes are
often known beforehand, one can design a wavelet method to not only detect the
existence of such pattern but also identify the exact time of the occurrence of these
abnormal patterns.
An important issue in wavelet analysis of EEG is the choice of epoch length.
When analyzing a single epoch, the total number of sequences available is of direct
influence on the standardized time period testing. This influence is due to the unreliable intercorrelation between adjacent sequences. Epoch lengths of 1–2 s duration are
recommended for EEG processing. This duration guarantees a widespread stability
in the data features. This is even more important when using low sample frequencies.
The wavelet analysis is particularly important in analysis of EPs. Wavelet analysis of EP finds the typical shape of the EP pattern by finding the largest coefficients
identifying the highest correlation of decomposed signals with the pattern of applied
stimuli. In other words, when a decomposed version of the signal at a certain level
shows a high correlation with the stimuli pattern, the resulting waves in the decomposed signal identify a good estimation of the EP pattern.
Wavelet is also helpful in the determination of synchrony at the scalp. Specifically,
the determination of the exact delay between two patterns can quantitatively identify
the level of synchrony of the two waves. This decomposition is used in a number of
measures of harmonization, where the measures are based on both amplitude and
phase information.
Wavelet analysis can also be used for denoising and filtering of the EEG signal.
Often the very low scales, i.e., high frequencies, identify the additive noise, and, by
removing the low-scale components, the signal can be filtered. This was demonstrated in the chapter dedicated to wavelet analysis.
10.8 SUMMARY
In this chapter, we described the recording of nerve cell electric activity in the brain
by means of a signal called EEG. EEG can be characterized by four specific frequency ranges that roughly differentiate different mental activities. Alpha waves are
associated with sensory input and data processing. Higher frequencies in the beta