Application to biomedical signals - Practical Biomedical Signal Analysis Using MATLAB - page 116

discriminant analysis, or artificial neural networks (ANN).

4.1.6.1 Single channel analysis

Estimation of power spectra is one of the most frequently used methods of EEG
analysis. It provides information about the basic rhythms present in the signal and
can be easily and rapidly calculated by means of the fast Fourier transform (FFT).
Maximum entropy power spectrum may be obtained by means of the autoregres-
sive model, which can be recommended for the EEG analysis. Notwithstanding the
advantages of AR spectral estimate mentioned already (Sect. 2.3.2.2.3) the transfer
function of the AR model represented as a system of parallel filters is compatible
with the presence of EEG rhythms.
The impulse response function of the AR model is a sum of damped sinusoids
(Sect. 2.3.2.2.4) as was demonstrated by [Franaszczuk and Blinowska, 1985], who
introduced the method of parametrization of EEG rhythms called FAD (frequency,
amplitude, damping). These parameters determined from the AR model give directly
frequency and amplitude of the rhythms without the need of plotting the spectra.
The parameter concerning the damping of rhythms is useful in modeling of EEG
time series [Wright, 1989]. FAD method was used, e.g., for a study of stability of
electrocortical rhythm generators [Mitraszewski et al., 1987]. It was demonstrated
that the transfer function of the AR model is compatible with the transfer function of
the physiological model of rhythms generation. The FAD parameters were connected
with the parameters describing action of neural populations connected in a feedback
loop [Blinowska and Franaszczuk, 1989].
The spectral content of EEG changes during childhood and adolescence, attains
a stable character about 21 years of age, and then may be subject to changes in
old age. In the immature brain there is a preponderance of low frequency rhythms
and progressively the role of higher rhythms increases. The enhancement of the
low frequency activity in the adult EEG may be a symptom of neuropathology and
also learning and performance difficulties. This observation lead [Matousek and Pe-
tersen, 1973] to introduce the so-called age quotient based on the contribution to
the EEG, rhythms appropriate for each age (determined by regression analysis on a
large pool of subjects). The discrepancy above 0.8 in age quotient (a ratio between
the age estimated from EEG and the actual one) was considered as a sign of possible
pathology. In the computer assisted diagnostic system for neurological and psychi-
atric diseases—Neurometric [John and Thatcher, 1977] the ratio of the slow rhythms
delta and theta to the alpha rhythm computed for each derivation by means of spec-
tral analysis was one of the markers of possible disturbances. Another marker in this
system was the degree of asymmetry of spectral power between hemispheres.
The application of the psychoactive or anti-epileptic drugs causes pronounced
changes in EEG spectral characteristics. The influence of drugs and the effects of
medication are evaluated by means of calculation of spectral power in the charac-
teristic frequency bands and their ratios. In psychopharmacology the topographical
distribution of spectral power is also considered. Usually the statistically significant
differences between spectral power in frequency bands before and after medication