Adaptive segmentation of electroencephalographic data using a nonlinear energy operator

Analysis of long-term EEG requires that it be segmented into piece-wise stationary sections. This is accomplished by drawing boundaries at time instants of change in the amplitude or frequency content of the EEG. In this paper we describe a method of signal characterization that can be used to segment EEGs. This method is based on a nonlinear energy operator that inherently combines the amplitude and frequency content of the EEG. We show how the resulting frequency-weighted energy measure can be used for segmentation. By using synthetic and real data, the proposed method is compared to a popular segmentation method from the EEG literature. Enhanced sensitivity of the proposed method (particularly to the changes in frequency) are highlighted.