Utility of complexity analysis in electroencephalography and electromyography for automated classification of sleep-wake states in mice.

We explore an innovative approach to sleep stage analysis by incorporating complexity features into sleep scoring methods for mice. Traditional sleep scoring relies on the power spectral features of electroencephalogram (EEG) and the electromyogram (EMG) amplitude. We introduced a novel methodology for sleep stage classification based on two types of complexity analysis, namely multiscale entropy and detrended fluctuation analysis. Our analysis revealed significant variances in these complexities, not only within the specific theta and delta bands but across a wide frequency spectrum. Based on these findings, we developed a sleep stage scoring model, termed Sleep Analyzer Complex (SAC), a convolutional neural network model that integrates these complexity features with conventional EEG spectrum and EMG amplitude analysis. This integrated model significantly enhances the accuracy of sleep stage identification, achieving an accuracy of 97.4-98.1% for novel wild-type mice, on par with the agreement level among human scorers (97.3-97.8%). The efficacy of SAC was validated through tests conducted on wild-type mice, and it demonstrated remarkable success in identifying sleep architecture abnormalities in narcoleptic mice as well. This approach not only facilitates automated scoring of sleep/wakefulness states but also holds the potential to uncover detailed physiological insights, thereby advancing EEG-based sleep research.