Deep learning enables sleep staging from photoplethysmogram for patients with suspected sleep apnea

Abstract

Study Objectives: Accurate identification of sleep stages is essential in the diagnosis of sleep disorders (e.g. obstructive sleep apnea [OSA]) but relies on labor-intensive electroencephalogram (EEG)-based manual scoring. Furthermore, long-term assessment of sleep relies on actigraphy differentiating only between wake and sleep periods without identifying specific sleep stages and having low reliability in identifying wake periods after sleep onset. To address these issues, we aimed to develop an automatic method for identifying the sleep stages from the photoplethysmogram (PPG) signal obtained with a simple finger pulse oximeter. Methods: PPG signals from the diagnostic polysomnographies of susptected OSA patients (n = 894) were utilized to develop a combined convolutional and recurrent neural network. The deep learning model was trained individually for three-stage (wake/NREM/REM), four-stage (wake/N1+N2/N3/REM), and five-stage (wake/N1/N2/N3/REM) classification of sleep. Results: The three-stage model achieved an epoch-by-epoch accuracy of 80.1% with Cohen's κ of 0.65. The four- and five-stage models achieved 68.5% (κ = 0.54), and 64.1% (κ = 0.51) accuracies, respectively. With the five-stage model, the total sleep time was underestimated with a mean bias error (SD) of of 7.5 (55.2) minutes. Conclusion: The PPG-based deep learning model enabled accurate estimation of sleep time and differentiation between sleep stages with a moderate agreement to manual EEG-based scoring. As PPG is already included in ambulatory polygraphic recordings, applying the PPG-based sleep staging could improve their diagnostic value by enabling simple, low-cost, and reliable monitoring of sleep and help assess otherwise overlooked conditions such as REM-related OSA. Statement of Significance Sleep staging is the cornerstone of diagnosing sleep disorders. However, the diagnosis of obstructive sleep apnea is increasingly reliant on home-based recordings without the ability for sleep staging due to the lack of EEG recording. This hinders the ability to assess sleep architecture, with total sleep time having to be manually estimated from other signals. This leads to large errors in diagnostic parameters that rely on the accurate determination of sleep time. We developed a novel, deep learning-based sleep staging method relying only on photoplethysmogram measured with a finger pulse oximeter. The deep learning approach enables differentiation of sleep stages and accurate estimation of total sleep time. This could easily enhance the diagnostic yield of home-based recordings and enable cost-efficient, long-term monitoring of sleep.