Feedback Control of Intravenous Anesthetics by Quantitative EEG

The use of a feedback system to control intravenous anesthetic drug delivery based on the spontaneous EEG requires at first mapping the anesthesia-induced EEG changes onto a figure which can be easily calculated and used for control. The power spectrum of the EEG signal has been widely investigated as a basis for the derivation of EEG parameters. The power spectrum is nothing but a distribution; thus, the derivation of spectral EEG parameters is equivalent to the derivation of descriptors of a distribution. It should, however, be mentioned here that the estimation of the power spectrum is not a unique procedure. The transformation from the time domain into the frequency domain requires a signal extending from −∞ to +∞ in time. Because an EEG epoch is finite in time, methods for extrapolating the signal to the future and into the past are required. Fourier transformation assumes that the epoch under consideration is repeated indefinitely to ±00, while other methods of power spectrum estimation, such as maximum entropy analysis [28], assume other methods of extrapolation. For the purpose of EEG monitoring during anesthesia, these differences in estimating the power spectrum are of minor importance if the EEG epoch is long enough. Figure 1 gives the scheme of an EEG power spectrum and the three monoparametric descriptors of distribution which were used. The simplest one is mean EEG frequency, defined as the mean of the power spectrum regarded as a distribution. As is known from ordinary statistics, the mean can give inappropriate measures of the center in the case of skew symmetric distributions and is, in addition, rather sensitive to outliers.