New trends in electrophysiological signal parametrization and feature extraction.

Abstract

New trends in physiological time series analysis are described. The discussed topics concern: parametric methods including their multi-channel version, non-linear formalism, artificial neural networks and wavelet analysis. The parametric methods which have been used mainly in EEG analysis are making their way into other fields. In their multi-channel version autoregressive models furnish information about the intrinsic relationship between signals. The chaotic formalism have brought new insight into the mechanisms of physiological processes. Nevertheless, in the description of physiological time series parameters estimated by means of non-linear formalism should be used with care. Artificial neural networks are becoming a popular method of time series classification. They are especially suitable in case of noisy and incomplete data. Interesting application of neural networks can be foreseen in the problems where not only classification but also control are involved. Wavelet analysis is a new method which makes possible the simultaneous estimation of signal frequency and localization in time with the accuracy superior to any previously known method. The method is especially suitable for the analysis of fast-varying non-stationary signals.