Assessment of a Novel Single-Channel EEG Method for Automated Recognition of Sleep Phases

Assessing sleep stages is crucial for identifying and managing sleep disorders such as REM sleep disorders and narcolepsy. Automating this process not only expedites detection but also enhances diagnostic accuracy. Thus, this research used the discrete wavelet transform technique and a convolutional neural network to automate sleep stage identification. In this proposed method, the single-channel EEG undergoes decomposition into four levels via discrete wavelet transform, followed by extraction of statistical features from these levels. Subsequently, employing CNN, relevant features are selected and utilized as input to the model. The CNN model achieved high accuracies of 95.81% five-class (five sleep stages). This proposed approach for automated sleep stage detection has the potential to expedite the identification of sleep stages and even sleep disorders, and it can accommodate large-scale EEG datasets. However, it's important to note that this approach was solely evaluated on one dataset, underscoring the need for validation across other databases in future investigations.