A deep transfer learning approach for wearable sleep stage classification with photoplethysmography

Mustafa Radha,Arnaud Moreau,Pedro Fonseca,Andreas Cerny,Ronald M Aarts,Xi Long,Peter Anderer,Marco Ross

doi:10.1038/s41746-021-00510-8

Mustafa Radha, Arnaud Moreau + Show 6 more

Open Access

https://doi.org/10.1038/s41746-021-00510-8

Copy DOI

Abstract

Unobtrusive home sleep monitoring using wrist-worn wearable photoplethysmography (PPG) could open the way for better sleep disorder screening and health monitoring. However, PPG is rarely included in large sleep studies with gold-standard sleep annotation from polysomnography. Therefore, training data-intensive state-of-the-art deep neural networks is challenging. In this work a deep recurrent neural network is first trained using a large sleep data set with electrocardiogram (ECG) data (292 participants, 584 recordings) to perform 4-class sleep stage classification (wake, rapid-eye-movement, N1/N2, and N3). A small part of its weights is adapted to a smaller, newer PPG data set (60 healthy participants, 101 recordings) through three variations of transfer learning. Best results (Cohen’s kappa of 0.65 ± 0.11, accuracy of 76.36 ± 7.57%) were achieved with the domain and decision combined transfer learning strategy, significantly outperforming the PPG-trained and ECG-trained baselines. This performance for PPG-based 4-class sleep stage classification is unprecedented in literature, bringing home sleep stage monitoring closer to clinical use. The work demonstrates the merit of transfer learning in developing reliable methods for new sensor technologies by reusing similar, older non-wearable data sets. Further study should evaluate our approach in patients with sleep disorders such as insomnia and sleep apnoea.

Highlights

The objective measurement of sleep at home in an unobtrusive manner has become an increasingly important topic of study as systematic sleep deprivation is increasingly linked to health adversities such as weight gain[1], systemic inflammation[2], weakened glucose regulation[3] and poor fitness to drive[4]
We show that a transfer learning approach, in which a model is first trained on a large ECG data set and adapted using a small PPG data set, leads to better performance for PPG-based sleep stage classification than when using only ECG data or only PPG data to train the model
The first data set is Siesta[34], including 292 participants (584 overnight recordings), a large data set with ECG signals and sleep scoring according to older Rechtschaffen & Kales (R&K) guidelines from the PSG signals

Summary

Introduction

The objective measurement of sleep at home in an unobtrusive manner has become an increasingly important topic of study as systematic sleep deprivation is increasingly linked to health adversities such as weight gain[1], systemic inflammation[2], weakened glucose regulation[3] and poor fitness to drive[4]. Sleep assessments in sleep medicine could benefit from longitudinal home monitoring[5], which could provide a complementary role to gold-standard polysomnography (PSG) measurements. Sleep stage scoring is normally done through manual visual annotation of PSG data, which include electro-graphic measurements of cortical brain activity as well as eye and chin muscle activity. Before 2007 the Rechtschaffen & Kales (R&K)[9] guidelines were the most commonly used, since their publication in 1968 While these standards are very comparable, structural differences have been found that lead to different results when comparing AASM to R&K annotation: increase in N1 and N3 scoring; a decrease in N2 scoring; and a decrease in REM scoring in younger people[10]. An example of a night concurrently scored using both guidelines is illustrated in

Objectives

Methods

Results

Conclusion