ABSTRACT
In this chapter, electroencephalogram (EEG) signals obtained from polysomnography (PSG) recordings are analyzed using the discrete wavelet transform (DWT) and dispersion entropy (DEn). The PSG recordings are taken from the PhysioNet Sleep European Data Format (EDF) Database. We investigate the performance of DEn and one of its variant fluctuation based dispersion entropies (FDEn) computed from the the wavelet sub-bands of the EEG sleep recordings. The random forest classifier is employed for classification with computed entropies as features. The performance of the algorithm is further compared using the DEn and FDEn separately as well as in combination. Measures, like sensitivity, specificity, and accuracy, are used to compare the performance of the multi-class sleep stage classification problem. The results show the suitability of the DEn and FDEn for automatic scoring of the sleep stages. It is found that FDEn is more suitable for classification when compared to DEn. We also observed that classification using both DEn and FDEn combined as a feature resulted in even better sleep stage classification accuracy.
