Recent advances in EEG data processing

Abstract

It is argued that the most interesting advances in EEG signal processing are with methods based on descriptive mathematical models of the process. Formulation of auto-regressive (AR) and mixed autoregressive and moving average (ARMA) models is reviewed for the scalar and the multidimensional cases and extensions to allow time-varying coefficients are pointed out. Data processing with parametric models, DPPM, involves parameter estimation and a large number of algorithms are available. Emphasis is put on those that are simple to apply and require a modest amount of computation. A recursive algorithm by Levinson, Robinson and Durbin is well suited for estimation of the coefficients in the AR model and for tests of model order. It is applicable to both the scalar and multidimensional cases. The ARMA model can be handled by approximation of an AR model or by nonlinear optimization. Recursive estimation with AR and ARMA models is reviewed and the connection with the Kalman filter pointed out. In this way processes with time-varying properties may be handled and a stationarity index is defined. The recursive algorithms can deal with AR or ARMA models in the same way. A reformulation of the algorithm to include sparsely updated parameter estimates significantly speeds up the calculations. It will allow several EEG channels to be handled simultaneously in real time on a modern minicomputer installation. DPPM has been particularly successful in the areas of spectral analysis and detection of short transients such as spikes and sharp waves. Recently some interesting attempts have been made to apply classification algorithms to estimated parameters. A brief review is made of the main results in these areas.