Computer analysis of EEG signals with parametric models

Fifty years ago Berger made the first registrations of the electrical activity of the brain with electrodes placed on the intact skull. It immediately became clear that the frequency content of recorded signals plays an important role in describing these signals and also the state of the brain. This paper briefly surveys the main properties the electroencephalogram (EEG), and points out several influential factors. A number of methods have been developed to quantify the EEG in order to complement visual screening; these are conveniently classified as being parametric or nonparametric. The paper emphasizes parametric methods, in which signal analysis is based on a mathematical model of the observed process. The scalar or multivariate model is typically linear, with parameters being either time invariant or time variable. Algorithms to fit the model to observed data are surveyed. Results from the analysis my be used to describe the spectral properties of the EEG, including the way in which characteristic variables change with time. Parametric models have successfully been applied to detect the occurrence of transients with epiliptic origin, so-called spikes and sharp waves. Interesting results have also been obtained by combining parameter estimation with classification algorithms in order to recognize significant functional states of the brain. The paper emphasizes methodology but includes also brief accounts of applications for research and clinical use. These mainly serve to illustrate the progress being made and to indicate the need for further work. The rapid advance of computer technology makes the processed EEG an increasingly viable tool in research and clinical practice.