Abstract

Continuous long-term electroencephalogram (cEEG) has been shown to be extremely valuable in monitoring patients whose brain function may be in jeopardy, particularly in the neurological intensive care unit (NICU). Since cEEG monitoring can last from days to weeks, the amount of data generated can become unwieldy and methods that can help in the review process are necessary. Recently, a technique that compresses several hours of cEEG has been presented. The compressed summary includes a graph showing the temporal evolution of the different patterns in the cEEG along with their representative samples presented in the traditional EEG display format. This technique is based on segmentation and classification using a set of features called the generic features consisting of the amplitude, frequency, and frequency-weighted energy. Because the generic features used in this method are not always optimal, our aim in this thesis is to incorporate an alternative feature set in the above cEEG compression method. Specifically, in this thesis we propose a set of features based on power spectral density, referred to as the spectral features.