Akaike's model versus conventional spectral analysis as tools for analyzing multivariate clinical time series

Abstract

Akaike's method of multivariate autoregressive (AR) modeling is applied to time-series analysis of clinical data. The present approach successfully demonstrated the peculiar power spectrum in various time-series data, which failed to be detected by FFT analysis because of abundant noise. Once AR coefficients are computed from the observed time-series of the relevant variables they can be used to describe the peculiar behavior of the system under study in two different ways: impulse response (IR) curves and Akaike's relative power contribution. The original program of Akaike is modified for exclusive uses in the analysis of clinical data.<>