Automated sleep stage classification using heart rate variability (HRV) may provide an ergonomic and low-cost alternative to gold standard polysomnography, creating possibilities for unobtrusive home-based sleep monitoring. Current methods however are limited in their ability to take into account long-term sleep architectural patterns. A long short-term memory (LSTM) network is proposed as a solution to model long-term cardiac sleep architecture information and validated on a comprehensive data set (292 participants, 584 nights, 541.214 annotated 30 s sleep segments) comprising a wide range of ages and pathological profiles, annotated according to the Rechtschaffen and Kales (R&K) annotation standard. It is shown that the model outperforms state-of-the-art approaches which were often limited to non-temporal or short-term recurrent classifiers. The model achieves a Cohen’s k of 0.61 ± 0.15 and accuracy of 77.00 ± 8.90% across the entire database. Further analysis revealed that the performance for individuals aged 50 years and older may decline. These results demonstrate the merit of deep temporal modelling using a diverse data set and advance the state-of-the-art for HRV-based sleep stage classification. Further research is warranted into individuals over the age of 50 as performance tends to worsen in this sub-population.

Automatic sleep stage classification with cardiorespiratory signals has attracted increasing attention. In contrast to the traditional manual scoring based on polysomnography, these signals can be measured using advanced unobtrusive techniques that are currently available, promising the application for personal and continuous home sleep monitoring. This paper describes a methodology for classifying wake, rapid-eye-movement (REM) sleep, and non-REM (NREM) light and deep sleep on a 30 s epoch basis. A total of 142 features were extracted from electrocardiogram and thoracic respiratory effort measured with respiratory inductance plethysmography. To improve the quality of these features, subject-specific Z-score normalization and spline smoothing were used to reduce between-subject and within-subject variability. A modified sequential forward selection feature selector procedure was applied, yielding 80 features while preventing the introduction of bias in the estimation of cross-validation performance. PSG data from 48 healthy adults were used to validate our methods. Using a linear discriminant classifier and a ten-fold cross-validation, we achieved a Cohen's kappa coefficient of 0.49 and an accuracy of 69% in the classification of wake, REM, light, and deep sleep. These values increased to kappa = 0.56 and accuracy = 80% when the classification problem was reduced to three classes, wake, REM sleep, and NREM sleep.

To compare the accuracy of automatic sleep staging based on heart rate variability measured from photoplethysmography (PPG) combined with body movements measured with an accelerometer, with polysomnography (PSG) and actigraphy. Using wrist-worn PPG to analyze heart rate variability and an accelerometer to measure body movements, sleep stages and sleep statistics were automatically computed from overnight recordings. Sleep–wake, 4-class (wake/N1 + N2/N3/REM) and 3-class (wake/NREM/REM) classifiers were trained on 135 simultaneously recorded PSG and PPG recordings of 101 healthy participants and validated on 80 recordings of 51 healthy middle-aged adults. Epoch-by-epoch agreement and sleep statistics were compared with actigraphy for a subset of the validation set. The sleep–wake classifier obtained an epoch-by-epoch Cohen’s κ between PPG and PSG sleep stages of 0.55 ± 0.14, sensitivity to wake of 58.2 ± 17.3%, and accuracy of 91.5 ± 5.1%. κ and sensitivity were significantly higher than with actigraphy (0.40 ± 0.15 and 45.5 ± 19.3%, respectively). The 3-class classifier achieved a κ of 0.46 ± 0.15 and accuracy of 72.9 ± 8.3%, and the 4-class classifier, a κ of 0.42 ± 0.12 and accuracy of 59.3 ± 8.5%. The moderate epoch-by-epoch agreement and, in particular, the good agreement in terms of sleep statistics suggest that this technique is promising for long-term sleep monitoring, although more evidence is needed to understand whether it can complement PSG in clinical practice. It also offers an improvement in sleep/wake detection over actigraphy for healthy individuals, although this must be confirmed on a larger, clinical population.

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.

This paper proposes the use of dynamic warping (DW) methods for improving automatic sleep and wake classification using actigraphy and respiratory effort. DW is an algorithm that finds an optimal nonlinear alignment between two series allowing scaling and shifting. It is widely used to quantify (dis)similarity between two series. To compare the respiratory effort between sleep and wake states by means of (dis)similarity, we constructed two novel features based on DW. For a given epoch of a respiratory effort recording, the features search for the optimally aligned epoch within the same recording in time and frequency domain. This is expected to yield a high (or low) similarity score when this epoch is sleep (or wake). Since the comparison occurs throughout the entire-night recording of a subject, it may reduce the effects of within- and between-subject variations of the respiratory effort, and thus help discriminate between sleep and wake states. The DW-based features were evaluated using a linear discriminant classifier on a dataset of 15 healthy subjects. Results show that the DW-based features can provide a Cohen's Kappa coefficient of agreement κ = 0.59 which is significantly higher than the existing respiratory-based features and is comparable to actigraphy. After combining the actigraphy and the DW-based features, the classifier achieved a κ of 0.66 and an overall accuracy of 95.7%, outperforming an earlier actigraphy- and respiratory-based feature set ( κ = 0.62). The results are also comparable with those obtained using an actigraphy- and cardiorespiratory-based feature set but have the important advantage that they do not require an ECG signal to be recorded.