Abstract:We developed a real-time sleep stage classification system with a convolutional neural network using only a one-channel electro-encephalogram source from mice and universally available features in any time-series data: raw signal, spectrum, and zeitgeber time. To accommodate historical information from each subject, we included a long short-term memory recurrent neural network in combination with the universal features. The resulting system (UTSN-L) achieved 90% overall accuracy and 81% multi-class Matthews Co… Show more
“…S2 ). A similar proportion has been reported in other studies in mice with the same genetic background 13 . Figure 1 While EEG signals of wakefulness and REMS share some visual similarities, their imbalanced frequencies makes REMS detection difficult for machine learning.…”
Section: Resultssupporting
confidence: 90%
“…They reported a macro-averaging F1 score of 69% to 86% for REMS between different cohorts of mice and rats. Tezuka et al developed a real-time sleep stage classification system with a CNN, named the universal time-series network, processing raw EEG, FFT, and zeitgeber time together 13 for 91% sensitivity and 98% specificity for REMS. Using supervised methods, both experiments reported very successful results; however, both assessed EEG frequencies up to 100 Hz and did not look at the possible role of higher frequencies in REMS detection, which would be affected by vigilance levels 14 and could be used as a feature to score sleep stages in an entirely unsupervised manner.…”
Rapid eye movement sleep (REMS) is characterized by the appearance of fast, desynchronized rhythms in the cortical electroencephalogram (EEG), similar to wakefulness. The low electromyogram (EMG) amplitude during REMS distinguishes it from wakefulness; therefore, recording EMG signal seems to be imperative for discriminating between the two states. The present study evaluated the high frequency components of the EEG signal from mice (80–500 Hz) to support REMS detection during sleep scoring without an EMG signal and found a strong positive correlation between waking and the average power of 80–120 Hz, 120–200 Hz, 200–350 Hz and 350–500 Hz. A highly negative correlation was observed with REMS. Furthermore, our machine learning approach demonstrated that simple EEG time-series features are enough to discriminate REMS from wakefulness with sensitivity of roughly 98 percent and specificity of around 92 percent. Interestingly, assessing only the higher frequency bands (200–350 Hz as well as 350–500 Hz) gives significantly greater predictive power than assessing only the lower end of the EEG frequency spectrum. This paper proposes an approach that can detect subtle changes in REMS reliably, and future unsupervised sleep-scoring approaches could greatly benefit from it.
“…S2 ). A similar proportion has been reported in other studies in mice with the same genetic background 13 . Figure 1 While EEG signals of wakefulness and REMS share some visual similarities, their imbalanced frequencies makes REMS detection difficult for machine learning.…”
Section: Resultssupporting
confidence: 90%
“…They reported a macro-averaging F1 score of 69% to 86% for REMS between different cohorts of mice and rats. Tezuka et al developed a real-time sleep stage classification system with a CNN, named the universal time-series network, processing raw EEG, FFT, and zeitgeber time together 13 for 91% sensitivity and 98% specificity for REMS. Using supervised methods, both experiments reported very successful results; however, both assessed EEG frequencies up to 100 Hz and did not look at the possible role of higher frequencies in REMS detection, which would be affected by vigilance levels 14 and could be used as a feature to score sleep stages in an entirely unsupervised manner.…”
Rapid eye movement sleep (REMS) is characterized by the appearance of fast, desynchronized rhythms in the cortical electroencephalogram (EEG), similar to wakefulness. The low electromyogram (EMG) amplitude during REMS distinguishes it from wakefulness; therefore, recording EMG signal seems to be imperative for discriminating between the two states. The present study evaluated the high frequency components of the EEG signal from mice (80–500 Hz) to support REMS detection during sleep scoring without an EMG signal and found a strong positive correlation between waking and the average power of 80–120 Hz, 120–200 Hz, 200–350 Hz and 350–500 Hz. A highly negative correlation was observed with REMS. Furthermore, our machine learning approach demonstrated that simple EEG time-series features are enough to discriminate REMS from wakefulness with sensitivity of roughly 98 percent and specificity of around 92 percent. Interestingly, assessing only the higher frequency bands (200–350 Hz as well as 350–500 Hz) gives significantly greater predictive power than assessing only the lower end of the EEG frequency spectrum. This paper proposes an approach that can detect subtle changes in REMS reliably, and future unsupervised sleep-scoring approaches could greatly benefit from it.
“…50 Hz EEG and EMG signals were also filtered since the power supply frequency in Japan (50 Hz) possibly causes artificial noise signals. The EEG and EMG signals were acquired using SleepSignRecorder (KISSEI COMTEC, Nagano, Japan) at a sampling rate of 128 Hz, which is the same frequency as used in the previous study for the neural network model described later (Tezuka et al, 2021). Eight young mice and 10 aged mice were kept in the LD condition for 4 days after the measurement and then transferred to DD.…”
Aging disrupts circadian clocks, as evidenced by a reduction in the amplitude of circadian rhythms. Because the circadian clock strongly influences sleep–wake behavior in mammals, age-related alterations in sleep–wake patterns may be attributable, at least partly, to functional changes in the circadian clock. However, the effect of aging on the circadian characteristics of sleep architecture has not been well assessed, as circadian behaviors are usually evaluated through long-term behavioral recording with wheel-running or infrared sensors. In this study, we examined age-related changes in circadian sleep–wake behavior using circadian components extracted from electroencephalography (EEG) and electromyography (EMG) data. EEG and EMG were recorded from 12 to 17-week-old and 78 to 83-week-old mice for 3 days under light/dark and constant dark conditions. We analyzed time-dependent changes in the duration of sleep. Rapid eye movement (REM) and non-REM (NREM) sleep significantly increased during the night phase in old mice, whereas no significant change was observed during the light phase. The circadian components were then extracted from the EEG data for each sleep–wake stage, revealing that the circadian rhythm in the power of delta waves during NREM sleep was attenuated and delayed in old mice. Furthermore, we used machine learning to evaluate the phase of the circadian rhythm, with EEG data serving as the input and the phase of the sleep–wake rhythm (environmental time) as the output. The results indicated that the output time for the old mice data tended to be delayed, specifically at night. These results indicate that the aging process significantly impacts the circadian rhythm in the EEG power spectrum despite the circadian rhythm in the amounts of sleep and wake attenuated but still remaining in old mice. Moreover, EEG/EMG analysis is useful not only for evaluating sleep–wake stages but also for circadian rhythms in the brain.
“…Mice serve as essential model organisms for investigating sleep patterns and their implications in humans 4 . Sleep stages in mice can be categorized into wake, non-REM, and REM stages [5][6][7] . During the wake stage, mice exhibit active brain and body functions, resulting in mixed frequency electroencephalography (EEG) signals and large electromyography (EMG) amplitudes 8 .…”
Sleep research is fundamental to understanding health and well-being, as proper sleep is essential for maintaining optimal physiological function. Here we present SlumberNet, a novel deep learning model based on residual network (ResNet) architecture, designed to classify sleep states in mice using electroencephalogram (EEG) and electromyogram (EMG) signals. Our model was trained and tested on data from mice undergoing baseline sleep, sleep deprivation, and recovery sleep, enabling it to handle a wide range of sleep conditions. Employing k-fold cross-validation and data augmentation techniques, SlumberNet achieved high levels of accuracy (∼98%) in predicting sleep stages and showed robust performance even with a small and diverse training dataset. Comparison of SlumberNet’s performance to manual sleep stage classification revealed a significant reduction in analysis time (∼50x faster), without sacrificing accuracy. Our study showcases the potential of deep learning to facilitate sleep research by providing a more efficient, accurate, and scalable method for sleep stage classification. Our work with SlumberNet demonstrates the power of deep learning in sleep research, and looking forward, SlumberNet could be adapted to human EEG analysis and sleep stage classification. Thus, SlumberNet could be a valuable tool in understanding both sleep physiology and disorders in mammals.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
