Single-channel EEG based sleep staging is of interest to researchers due to its broad application prospect in daily sleep monitoring recently. We proposed using contextual scalograms as input and developed a convolutional neural network with attention modules named Co-ScaleNet for sleep staging. The contextual scalograms were obtained by combining the same color channels of three original RGB scalograms from consecutive epochs, and a simple and efficient data augmentation was designed according to their various forms. The Co-ScaleNet consists of two main parts. Firstly, three parallel convolutional branches with attention modules correspondingly extract and fuse features from contextual scalograms at the top layers. The remaining part is a stack of lightweight blocks. We achieved an overall accuracy of 87.0% for healthy individuals, 84.7% for depressed patients. And we obtained comparable performance on the public Sleep-EDFx (82.8%), ISRUC (84.6%) and SHHS datasets (87.7%), including a high recall of N1. The contextual scalograms of R channel as input achieved the best performance, which conform to the features of interest in visual scoring. The attention modules improved the recall of N1 and N3. Overall, the contextual scalograms provided a novel scheme for both contextual information extraction and data augmentation. Our study successfully expanded its application to depression datasets, as well as patients with sleep apnea, demonstrating its wide applicability.

Objective : Sleep-stage scoring is important for sleep-quality evaluation and the diagnosis of related diseases. In this study, an automatic sleep-stage scoring method using photoplethysmographic (PPG) signals was proposed. Approach : To construct the classification model, we extracted 14 time-domain features, 17 frequency-domain features, and 20 pulse rate variability (PRV) features along with four SpO ₂ features from PPG signals. An artificial neural network classifier was used to integrate the results of ten binary support vector machine classifiers and realise sleep-stage classification. Leave-one-subject-out validation was applied to evaluate our proposed model. Main results : Thirty-one subjects were enrolled in the study, in which 21 subjects were with high sleep quality (sleep efficiencies ⩾85%). Our model achieved accuracies of 57% (κ = 0.39), 62% (κ = 0.41), and 78% (κ = 0.54) for the classification of five sleep stages (wake, N1, N2, N3, and rapid eye movement (REM) sleeps), four sleep stages (wake, light, deep, and REM sleeps) and three sleep stages (wake, non-rapid eye movement (NREM), and REM sleeps), respectively. For the remaining ten subjects with poor sleep quality, the results came to 55% (κ = 0.39), 62% (κ = 0.43), and 75% (κ = 0.52). Significance : The satisfactory performance of our proposed model reveals the potential of PPG signals for sleep-stage scoring, which may contribute to automatic sleep monitoring in the home environment.

Sleep apnea–hypopnea syndrome is a common breathing disorder that can lead to organic brain injury, prevent memory consolidation, and cause other adverse mental-related complications. Brain activity while sleeping during respiratory events is related to these dysfunctions. In this study, we analyzed variations in electroencephalography (EEG) signals before, during, and after such events. Absolute and relative powers, as well as symbolic transfer entropy (STE) of scalp EEG signals, were calculated to unveil the activity of brain regions and information interactions between them, respectively. During the respiratory events, only low-frequency power increased during rapid eye movement (REM) stage (δ-band absolute and relative power) and N1 (δ- and θ-band absolute power, δ-band relative power) sleep. But absolute power increased in low- and medium-frequency bands (δ, θ, α, and σ bands), and relative power increased mainly in the medium-frequency band (α and σ bands) during stage N2 sleep. After the respiratory events, absolute power increased in all frequency bands and sleep stages, but relative power increased in medium and high frequencies. Regarding information interactions, the β-band STE decreased during and after events. In the γ band, the intrahemispheric STE increased during events and decreased afterward. Moreover, the interhemisphere STE increased after events during REM and stage N1 sleep. The EEG changes throughout respiratory events are supporting evidence for previous EEG knowledge of the impact of sleep apnea on the brain. These findings may provide insights into the influence of the sleep apnea–hypopnea syndrome on cognitive function and neuropsychiatric defects.