David E. McCarty1, Paul Y. Kim1, Clifton Frilot II2, Andrew A. Marino1
1Division of Sleep Medicine, Department of Neurology, LSU Health , Shreveport, LA
2School of Allied Health Professions, LSU Health Sciences Center, Shreveport, LA
Presented at SLEEP 2014, the 28th Annual Meeting of the Associated Professional Sleep Societies, Minneapolis, MN, June, 2014
REM sleep is a prime example of the connectivity model of cognition because REM is mediated by a dynamic array of interconnected neuronal networks. Analysis of brain recurrence (ABR) is a method for quantifying brain connectivity that is particularly useful for studying sleep. We previously found that ABR depth variables were correlated with NREM stages but were unable to disambiguate REM from N1 and N2 sleep. To overcome this problem, we defined an index of variability in sleep depth (GA index) that captured the short-term temporal changes occurring within the depth variables, and tested the hypothesis that REM could be disambiguated using a statistically determined combination of ABR markers for sleep depth and variability. Because of the mathematical structure of ABR, we expected that the disambiguation could be achieved using any one of the recorded EEGs.
When the principal components of the ABR markers were computed and used in 3-class support vector machines, the results matched ground truth as determined by expert staging, indicating that, as hypothesized, REM sleep could be reliably identified using a single EEG channel, irrespective of the presence or absence of OSA. Within the limitations of a computational approach, we conclude that REM sleep can be disambiguated from the other stages of sleep by means of ABR.