Application of frequency domain state-space analysis to sleep
Abstract
A frequency-domain based state-space analysis of electroencephalograms (EEGs) from sleeping individuals is shown to be effective for identifying sleep stages. The approach is based on the time-domain method of delay coordinate embedding applied to chaotic time series analysis. Using the isomorphic and reversible nature of the Fourier transformation, a frequency-domain analogue of the chaotic state space is created. An innovative, fast, and robust feed-forward neural network (FFNN) method is applied to this state-space representation to predict EEG chaotic dimension and to identify sleep stages. This analytical method, evaluated using data collected by the Sleep Heart Health Study, is shown to be quite successful. It is intended that this approach may be able to reveal more detail about functional sleep states, mixed-stage information, and differences between various arousal occurrences. Clinically, these new revelations are likely to be valuable in the study and evaluation of sleep disorders. This dissertation is a first step toward a quantitative sleep staging method to improve the understanding and evaluation of sleep disorders.
Degree
Ph.D.
Advisors
Talavage, Purdue University.
Subject Area
Biomedical engineering|Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.