Segmentation of Sleep EEG Signal by Optimal Thresholds

Daria G. Migotina and Agostinho C. da Rosa


Medical Imaging, Image Processing, and Signal Processing, Biomedical Signal Processing


This paper proposes a new methodology for sleep electroencephalogram (EEG) signals segmentation, based on maximum segments thresholding. The proposed methodology extracts segments that contain information about frequency characteristics of the sleep EEG signal. The objectives of the present work are to propose a new algorithm for selection of optimal thresholds, and to apply those same thresholds for the segmentation of sleep EEG signals. Maximum segments thresholding is a method of selecting optimal thresholds by choosing the value that allows detecting the maximum possible number of segments in EEG signals. From empirical results we found a specific threshold criterion, which is hypothesized as a universal relationship since it holds for different signals bandwidths, for all known physiological rhythms recorded, for different EEG leads or montage, for EEG from different age group and gender, and for recordings with different signal to noise ratios. At first, sleep EEG signals are described by their spectral characteristics, and then, applying the segmentation procedure on the data, optimal thresholds are obtained. The results of this research will be used in future works for a novel macro- and micro-structural sleep representation and clinical analysis of different populations.

Important Links:

Go Back