A New Approach for Parameterizing the ECG for Sleep Stage Classification

摘要

The analysis of the electrocardiogram is an important tool for getting insight into the physiological state of a person. The computation of the analysis is topic of current research activities. In this work a method for parameterizing the electrocardiogram is introduced. The electrocardiogram is expressed as a weighted sum of Gaussian functions. The parameters describing the ECG are the amplitudes, the time shifts and the variances (widths) of the particular Gaussian functions. Because of the fitting method, the parameters of the Gaussian functions are comparable over time. Further on, the extracted parameters are used to classify the different sleep stages, in particular the stages 1 to 4 and the REM sleep (Rapid Eye Movement). For classification, the clustering method with the k-nearest-neighbors classifier is used. The result of the parameterizing is that the ECG shape is different at the particular sleep stages. The classification shows results that follow the basic characteristics of the reliability of the sleep stage assignment of different raters, e.g. the classification accuracy between sleep and wake state is 93.6% + 2.7%.