Evaluation of a deep learning sleep staging algorithm utilizing a single frontal EEG channel on a clinical population with suspected or known obstructive sleep apnea.

Abstract

Home Sleep Apnea Testing (HSAT) for the diagnosis of Obstructive Sleep Apnea (OSA) has emerged as a simpler and cheaper diagnostic option compared with attended in-lab Polysomnography (PSG). The identification of sleep stages forms an essential part of the OSA diagnosis as it allows for proper phenotyping of OSA, specifically the REM phenotype. The manual staging of sleep is arduous and costly, so the development of accurate Deep Learning (DL) algorithms that automatically classify sleep stages forms a crucial role in the diagnosis of OSA with HSAT. The purpose of this study is to investigate the accuracy of a DL sleep staging algorithm in a new miniaturized sleep monitoring device – Compumedics Ltd Somfit®. Agreement between Somfit and PSG hypnograms is close to that between manual PSG hypnograms thus confirming acceptability of the single frontal EEG electrode placement for accurate automatic staging.