Towards Automated Sleep State Estimation using a Holter-Oximeter

Single-channel ECG and linger photoplethysmogram were used to estimate sleep states in obstructive sleep apnea patients. Overnight Holter-oximeter recordings from 14 subjects with suspected sleep apnea were analysed. Parameters including the RR interval, photoplethysmogram amplitude and rise time, and pulse arrival time were characterised and a hidden Markov model classifier used to identify wake, REM and sleep states. An overall accuracy of 77% and Cohen´s kappa of 0.54 was achieved, establishing a baseline level of usefulness for sleep state estimation using a Holter-oximeter. Autonomic arousals detected using pulse arrival time also appeared to be strongly correlated with Epworth Sleepiness Scale scores.