Analysis of EEG Time Series Recorded from Alzheimer Patients Based on Their Spectral Content

摘要

In this paper, EEG recordings on each channel are seen as non stationary time series. We assume a piecewise stationary model for the signal where the changes affect the power spectrum. Such a model is particularly relevant for analyzing Alzheimer disease (AD) dataset as the disease leads to variation of the frequency rhythms. Our method is devoted to the estimation of the partition. We define a criterion based on a regularized periodogram to estimate the spectrum and on the power spectrum on predefined frequency bands to estimate the change points. The method produces new markers designed to be used for the discrimination between control and AD patients.