Fully automatic REM sleep stage-specific intervention systems using single EEG in mice

Abstract

Sleep stage-specific intervention is widely used to elucidate the functions of sleep and their underlying mechanisms. For this intervention, it is imperative to accurately classify rapid-eye-movement (REM) sleep. However, the proof of fully automatic real-time REM sleep classification in vivo has not been obtained in mice. Here, we report the in vivo implementation of a system that classifies sleep stages in real-time from a single-channel electroencephalogram (EEG). It enabled REM sleep-specific intervention with 90 % sensitivity and 86 % precision without prior configuration to each mouse. We further derived systems capable of classification with higher frequency sampling and time resolution. This attach-and-go sleep staging system provides a fully automatic accurate and scalable tool for investigating the functions of sleep.

Keywords: AI; Deep learning; EEG; Mice; REM sleep; Sleep; Sleep stage classification.