Robust sleep stage classification with single-channel EEG signals using multimodal decomposition and HMM-based refinement

Abstract Sleep stage classification is a most important process in sleep scoring which is used to evaluate sleep quality and diagnose sleep-related diseases. Compared to complex sleep analysis devices, automatic sleep stage classification methods using single-channel electroencephalography (EEG) records benefit from the convenience of wearing and less interference in the sleep, thus are appropriate for home-based sleep analysis. In these methods, the design of representative features for classification plays the most important role in determining the performance. Previous works have not achieved satisfactory outcomes for ignoring several kinds of effective features. In this work, a novel multimodal signal decomposition and feature extraction strategy is presented to obtain effective features for sub-band signals. Meanwhile, a rule-free refinement process based on hidden Markov model (HMM) is proposed to optimize the classification results automatically. Experimental results show the superior classification performance of the proposed method compared to state-of-the-art works, wherein the rule-free refinement also outperforms previous rule-based correction algorithms. This sleep stage classification method is expected to contribute to the design of home-based sleep monitoring and analyzing system.