Detecting slow wave sleep and rapid eye movement stage using cortical effective connectivity

Abstract

In recent neuroimaging research, there has been considerable interest in identifying neuromarkers of sleep. Automatic slow wave sleep (SWS) and rapid eye movement (REM) are two known phases of sleep. However, the level by which those changes contribute to brain interactions has not been well characterized. In recent years, it has been shown that brain connectivity measuring can be helpful in investigation of behavioral states of the brain. By considering the fact that brains have different states in different stages of sleep, the present work employs effective connectivity and machine-learning analysis to quantify and classify SWS and REM stages of sleep. We examine low-density 12-channel EEG data from 8 healthy participants during a full night of sleep. Data were epoched into 30-s windows and SWS and REM stages were labeled by a sleep consultant. Effective connectivity was quantified using a directed metric, generalized partial directed coherence, and measures were used as input features for a machine-learning system. A support vector machine classifier was used to solve 2 binary problems of REM vs. nREM and SWS vs. nSWS. Findings revealed an excellent balanced accuracy of 89.80 % in REM detection and 87.32 % in SWS detection. Overall, our work demonstrates a successful application of effective connectivity analysis and machine learning for sleep neuromarkers in EEG.