Repeated automatic sleep scoring based on ear-EEG is a valuable alternative to manually scored polysomnography

Abstract

While polysomnography (PSG) is the gold standard to quantify sleep, modern technology allows for new alternatives. PSG is obtrusive, affects the sleep it is set out to measure and requires technical assistance for mounting. A number of less obtrusive solutions based on alternative methods have been introduced, but few have been clinically validated. Here we validate one of these solutions, the ear-EEG method, against concurrently recorded PSG in twenty healthy subjects each measured for four nights. Two trained technicians scored the 80 nights of PSG independently, while an automatic algorithm scored the ear-EEG. The sleep stages and eight sleep metrics (Total Sleep Time (TST), Sleep Onset Latency, Sleep Efficiency, Wake After Sleep Onset, REM latency, REM fraction of TST, N2 fraction of TST, and N3 fraction of TST) were used in the further analysis. We found the sleep metrics: Total Sleep Time, Sleep Onset Latency, Sleep Efficiency, Wake After Sleep Onset were estimated with high accuracy and precision between automatic sleep scoring and manual sleep scoring. However, the REM latency and REM fraction of sleep showed high accuracy but low precision. Further, the automatic sleep scoring systematically overestimated the N2 fraction of sleep and slightly underestimated the N3 fraction of sleep. We demonstrate that sleep metrics estimated from automatic sleep scoring based on repeated ear-EEG in some cases are more reliably estimated with repeated nights of automatically scored ear-EEG than with a single night of manually scored PSG. Thus, given the obtrusiveness and cost of PSG, ear-EEG seems to be a useful alternative for sleep staging for the single night recording and an advantageous choice for several nights of sleep monitoring. Author summary Sleep is important to our overall health, and as such, it is valuable in a clinical context to monitor sleep, for instance in terms of quality and duration. However, the established method for performing sleep measurements, the polysomnography, is uncomfortable to wear and expensive to use. In this study we have looked at the tradeoffs involved in using a smaller, less intrusive recording method, the ear-EEG, which has slightly worse data quality for a single night, but which is more feasible to use for multiple nights on the same patient. We find that while the single-night sleep information is slightly less reliable for the ear-EEG than for the polysomnogram, this can be counter acted by simply recording multiple nights; and ear-EEG used for two nights is generally more reliable than PSG used for 1 night. Considering that multiple nights of ear-EEG may in the future be much cheaper and easier to record than a single night of polysomnography, we conclude that ear-EEG sleep monitoring in the patients own home is promising method for clinical sleep monitoring.