Comparative analysis of different characteristics of automatic sleep stages
Introduction
Sleep is a recurrent physical and mental state of consciousness and body characteristics. And it also is one of the basic patterns of brain activity. There is about one-third of the time for adults to sleep, so proper quality and quantity of sleep have a major impact on human emotions and health. However, sleep problems in modern society are increasing due to mechanical and stressful life. Sleep disorders such as insomnia, drowsiness or obstructive sleep apnea (OSA) may lead to daytime sleepiness, insomnia at night, irritability, depression or anxiety, and even death. Effective diagnosis and treatment of patients with sleep-related diseases has become an important issue in the medical community. Sleep has become a branch of medicine and plays an important role in many clinical problems [1]. Sleep staging is the first step in the diagnosis of sleep-related diseases. Physiologically, assessing sleep quality depends on many aspects, including the duration and composition of sleep and so on. Most of the current studies about analysis of sleep are based on the sleep EEG signal [2], [3]. Accurate staging of sleep based on sleep EEG signals is the basis for studying sleep quality and diagnosis of sleep disorders. Rechtchaffen and Kales proposed the R&K sleep staging rule in 1968 [4]. The sleep process can be divided into awake period, non-rapid eye movement sleep period (NREM) and rapid eye movement sleep period (REM). NREM includes four stages: S1∼S4. The American Society of Sleep Medicine combines the S3 and S4 phases into deep sleep or slow wave sleep [5]. There are two kinds of the method of sleep staging. One is the early application of a wide range of manual sleep staging, but the process is not only complicated and cumbersome, but also depends on the level of experts, and the staging result is easily influenced by subjective factors. The other is automatic stage method, which overcomes the defects of manual stage to some extent and is an important development process of pattern recognition applied to sleep stages. The method of automatic staging of sleep is also continuously developed from shallow learning (such as SVM [6], maximum entropy algorithm [7], Boosting [8], etc.) to deep learning (such as artificial neural network(ANN) [9], [10], [11], deep belief network [12], etc.).
The automatic sleep stage classification usually takes the following steps: data set preparation, signal preprocessing, feature extraction and classification prediction. The data set preparation includes dividing the original EEG signal into 30 s [13]; the signal preprocessing mainly includes filtering and noise reduction processing to obtain relatively pure sleep EEG signals.
The key step of the sleep EEG staging lies in feature extraction, mainly linear and nonlinear, and Thiago used such as variance, Kurtosis, skewness, etc. to deal with the EEG [14]. Later, Ahmadlou etc. used the sequence connectivity based detrended fluctuation analysis (DFA) and the visible graph (VG) to analyze the EEG signals and achieved good results [15]. In addition to time domain, the parameters of transform domain have also been proved to be useful in EEG research. Rodríguezsotelo et al. proposed a single channel EEG scheme by using the power spectral density (PSD) of EEG [16]. Huang et al. used two channels for forehead EEG based on short time Fourier transform (STFT) [17]. In recent years, the research of feature extraction mainly focuses on nonlinear algorithm. Nonlinear dynamics, based on the concept of chaos, have been used in many fields, including medical and biological fields, and are often used to detect some arrhythmias, such as ventricular fibrillation etc. [18]. Studies have shown that this method can be used for pathological detection effectively, and further promote the research and development of EEG. Freeman proposed the neocortex dynamics EEG model, which has been extended to identify different types of anomalies [19]. In EEG data analysis, there are many different chaotic methods, such as correlation dimension (CD), Lyapunov exponent and entropy [20]. The application of chaos theory and nonlinear time series method has provided an in-depth understanding of brain dynamics reflected by EEG. Patidar et al. commonly used nonlinear algorithms, such as CD, approximate entropy(APEN) and Hearst index, are used in sleep EEG processing [21]. Şen et al. used time domain, frequency and nonlinear analysis to deal with sleep EEG simultaneously [22].
Classification methods mainly include ANN [9], SVM [23], decision tree (DT) [24], random forest (RF) [25], and fuzzy system [14] and so on. Saha et al. used wavelet feature and feedforward neural network for automatic sleep stages [26]. Khandoker et al. used SVM, which uses 28 features extracted from heart rate variability (HRV) data and the electrocardiogram (ECG) derived from wavelet decomposition to detect OSA [27]. Many earlier studies have reported the performance of sleep stage classification by HRV indices [28]. Aktaruzzaman et al. have recently reported classification performance by a combination of actigraphy and HRV in 18 subjects with no previous history of sleep disorders [29]. And the accuracy between sleep and waking were 78% by four features derived only from wrist actigraphy. Hsu et al. have implemented a real-time sleep apnea and hypopnea syndrome detection system using 111 features extracted from ECG data in the time domain and spectral domain [30].
In the study of sleep, in addition to EEG signals and ECG signals, there are many physiological signals that are related to sleep. However, in practical applications, there are certain difficulties in the collection of physiological signals. On the one hand, the physiological signal is weak, and it is difficult to collect due to the large influence of noise; on the other hand, it is because the device is not portable. Xia et al. researched the feasibility of a low-cost respiratory motion monitoring system based on the Microsoft(MS) KINECT sensor [31]. The result shows that it is feasible to use the KINECT for respiratory motion tracking. Traces are similar to those of a clinically used strain gauge system. The KINECT-based system provides a new and economical way to monitor respiratory motion. Procházka et al. used a simple thermal imaging camera to study the temperature change in the facial region and compared its temporal evolution to the chest region motion recorded by the MS Kinect depth sensor [32]. The proposed methods verify that thermal and depth cameras can be used as additional tools for multimodal detection of breathing patterns. Besides, researches showed that the amplitude of the breathing effort signal is more regular during non-REM sleep than during REM sleep [33]. Procházka and other researchers used MS KINECT to extract the breathing features for sleep staging [34]. Mikkelsen et al. combined the actigraphic and heart rate signals for sleep staging. The accuracy between REM and working reached 75.8% [35]. For the time being, more research on sleep is still using EEG signals. Mikkelsen et al. investigate automated sleep scoring based on a low-cost, mobile EEG platform consisting of a lightweight EEG amplifier combined with flex-printed cEEGrid electrodes placed around the ear, which can be implemented as a fully self‐applicable sleep system. However, cEEGrid signals have different amplitude characteristics to normal scalp PSG signals, which might be challenging for visual scoring [36].
This paper has five chapters. The first chapter introduces the basic knowledge of sleep staging based on EEG signals. And the next chapter includes some methods of feature extraction and classification. The third chapter mainly describes automatic staging process and results of sleep EEG signal. The fourth chapter mainly analyzes different features. The last chapter has a summary about this paper.
Section snippets
Methods
The sleep database used in this article comes from the data in the Sleep-EDF database of PhysioBank, namely Sleep-EDF Database [extended]. More details about the database will be explained in Chapter 3. The data for the whole night includes more than 20 h of sleep EEG signals. We used WT to preprocess the signal and divided into sections for every 30 s, there are more than 2000 sleep periods. We use wavelet analysis to preprocess the signal. After feature extraction, we randomly selected the
Results
This article mainly uses a total of 22 types of features for sleep EEG. Parameters in the time domain include kurtosis, skewness, and Hjorth parameters. Time-frequency parameters include the energy of each rhythm wave and the standard deviation of each rhythm wave. The analysis parameters include SampEn, fuzzy entropy, Tsallis entropy, FD, and complexity. The following text will use the feature number to indicate the corresponding feature. The correspondences are as the following: 1-Fuzzy; 2,
Analysis of different features
According to the results of the automatic staging of sleep electroencephalograms in the previous chapter, it can be seen that the staging results of different characteristics are different. Combining the differences in feature values during the automatic staging of sleep, this chapter will analyze and discuss various types of feature data.
ANOVA, also called ``analysis of variances,” is for the significance test of differences in the mean of two or more samples [113], [114]. The basic idea is to
Conclusion
This paper summarized many methods for analyzing sleep EEG in Section 2.1, mainly including time domain, frequency domain, time-frequency and nonlinear analysis methods, and use some of them as feature values of automatic sleep staging. The time domain analysis methods (such as skewness, kurtosis, and Hjorth parameters) are not very accurate in the automatic stages of sleep, and the accuracy of time frequency analysis methods (such as the energy and standard deviation of each rhythmic wave) is
Conflicts of interest
The authors declared that they have no conflicts of interest to this work.
Acknowledgments
This paper is funded by “Chongqing Research and Innovation Project of Graduate Students” (CYS17242) and Natural Science Foundation of Chongqing, China (cstc2018jcyjAX0163).
References (137)
- et al.
Complex networks approach for EEG signal sleep stages classification
Expert Syst. Appl.
(2016) - et al.
A novel, fast and efficient single-sensor automatic sleep-stage classification based on complementary cross-frequency coupling estimates
Clin. Neurophysiol.
(2018) Computer-aided obstructive sleep apnea detection using normal inverse Gaussian parameters and adaptive boosting
Biomed. Signal Process. Control
(2016)- et al.
Application of empirical mode decomposition and artificial neural network for the classification of normal and epileptic EEG signals
Biocybern. Biomed. Eng.
(2016) - et al.
Perspective: a dynamics-based classification of ventricular arrhythmias
J. Mol. Cell. Cardiol.
(2015) - et al.
Automatic sleep scoring using statistical features in the EMD domain and ensemble methods
Biocybern. Biomed. Eng.
(2016) - et al.
Automated epileptic seizure detection using improved correlation-based feature selection with random forest classifier
Neurocomputing
(2017) - et al.
Performance comparison between wrist and chest actigraphy in combination with heart rate variability for sleep classification
Comput. Biol. Med.
(2017) - et al.
Evaluating the sleep quality of obstructive sleep apnea patients after continuous positive airway pressure treatment
Comput. Biol. Med.
(2013) - et al.
Investigating the contribution of distance-based features to automatic sleep stage classification
Comput. Biol. Med.
(2018)
A new automatic sleep staging system based on statistical behavior of local extrema using single channel EEG signal
Expert Syst. Appl.
Comparison of FFT and marginal spectra of EEG using empirical mode decomposition to monitor anesthesia
Comput. Methods Programs Biomed.
Learning to short-time Fourier transform in spectrum sensing
Phys. Commun.
Fusing time, frequency and shape-related information: introduction to the Discrete Shapelet Transform's second generation (DST-II)
Inf. Fusion
EEG signal classification for BCI applications by wavelets and interval type-2 fuzzy logic systems
Expert Syst. Appl.
Automatic epileptic EEG detection using DT-CWT-based non-linear features
Biomed. Signal Process. Control
Dimensional regularization of Renyi's statistical mechanics
Phys. A Stat. Mech. Appl.
Financial time series modeling using the Hurst exponent
Phys. A Stat. Mech. Appl.
A convolutional neural network for sleep stage scoring from raw single-channel eeg
Biomed. Signal Process. Control
A rule-based automatic sleep staging method
J. Neurosci. Methods
Automated identification of sleep states from EEG signals by means of ensemble empirical mode decomposition and random under sampling boosting
Comput. Methods Programs Biomed.
Computer-aided obstructive sleep apnea detection using normal inverse Gaussian parameters and adaptive boosting
Biomed. Signal Process. Control
Sleep scoring using artificial neural networks
Sleep Med. Rev.
Epileptic seizure prediction using relative spectral power features
Clin. Neurophysiol.
Perceptual pain classification using ANFIS adapted RBF kernel support vector machine for therapeutic usage
Appl. Soft Comput.
Learning machines and sleeping brains: automatic sleep stage classification using decision-tree multi-class support vector machines
J. Neurosci. Methods
A comparison of performance of K-complex classification methods using feature selection
Inf. Sci.
A two-step automatic sleep stage classification method with dubious range detection
Comput. Biol. Med.
Computer-aided obstructive sleep apnea screening from single-lead electrocardiogram using statistical and spectral features and bootstrap aggregating
Biocybern. Biomed. Eng.
Assess sleep stage by modern signal processing techniques
IEEE Trans. Biomed. Eng.
A comparison study on stages of sleep: quantifying multiscale complexity using higher moments on coarse-graining
Commun. Nonlinear Sci. Numer. Simul.
Dual tree complex wavelet transform for sleep state identification from single channel electroencephalogram
Classification of EEG signals using a multiple kernel learning support vector machine
Sensors
Dysnomia, a computer program for maximum-entropy method (MEM) analysis and its performance in the MEM-based pattern fitting
Powder Diffr.
Epilepsy and seizure detection using statistical features in the complete ensemble empirical mode decomposition domain
On the classification of sleep states by means of statistical and spectral features from single channel Electroencephalogram
Using deep belief network modelling to characterize differences in brain morphometry in schizophrenia
Sci. Rep.
Automatic sleep stage classification
Single-channel EEG sleep stage classification based on a streamlined set of statistical features in wavelet domain
Med. Biol. Eng. Comput.
EEG classification of adolescents with type I and type II of bipolar disorder
Australas. Phys. Eng. Sci. Med.
Automatic sleep stages classification using EEG entropy features and unsupervised pattern analysis techniques
Entropy
Knowledge-based identification of sleep stages based on two forehead electroencephalogram channels
Front. Neurosci.
Modulation of epileptic activity by deep brain stimulation: a model-based study of frequency-dependent effects
Front. Comput. Neurosci.
Chaotic analysis of epileptic EEG signals
IFMBE Proc.
An integrated alcoholic index using tunable-Q wavelet transform based features extracted from EEG signals for diagnosis of alcoholism
Appl. Soft Comput.
A comparative study on classification of sleep stage based on EEG signals using feature selection and classification algorithms
J. Med. Syst.
Improving time-frequency domain sleep EEG classification via singular spectrum analysis
J. Neurosci. Methods
Enhanced inter-subject brain computer interface with associative sensorimotor oscillations
Healthc. Technol. Lett.
A comparison of machine learning methods for classification using simulation with multiple real data examples from mental health studies
Stat. Methods Med. Res.
The addition of entropy-based regularity parameters improves sleep stage classification based on heart rate variability
Med. Biol. Eng. Comput.
Cited by (34)
Systemic modeling of chaotic EEG during human sleep
2023, Informatics in Medicine UnlockedAn EEG emotion recognition method based on transfer learning and echo state network for HilCPS
2021, Microprocessors and MicrosystemsCitation Excerpt :Wen et al. [36] proposed a facial expressions recognition method based on ensemble convolutional ESN, and improved the accuracy of facial expressions recognition. The nonlinearity of EEG samples brings challenges to emotion recognition [37]. Compared with traditional classification models, ESN possesses strong nonlinear processing capability [38], which can solve the problem of the nonlinearity of EEG samples.
An effective multi-model fusion method for EEG-based sleep stage classification
2021, Knowledge-Based SystemsCitation Excerpt :Moreover, a variety of signal decomposition and transformation methods can be used to obtain effective classification features of different sleep stages [10–13]. Furthermore, classification model is used to detect specific sleep stages with the extracted signal features, the commonly used models include support vector machine, random forest, hidden Markova model and other machine learning methods [14–17]. In addition, more and more methods based on deep learning have been applied to classify sleep stages in recent years [18–20].
End-to-end sleep staging using convolutional neural network in raw single-channel EEG
2021, Biomedical Signal Processing and ControlCitation Excerpt :This neglect is often the result of a major misunderstanding [1]. Sleep staging is the first and key step in the study of sleep-related illnesses [2,3]. Polysomnogram (PSG) is a clinically approved sleep-monitoring device, recording several physiological signals, such as electroencephalogram (EEG), electrooculogram, electromyogram, electrocardiogram, and other environmental signals [4].
Detection of Abnormal Respiratory Events with Single Channel ECG and Hybrid Machine Learning Model in Patients with Obstructive Sleep Apnea
2020, IRBMCitation Excerpt :In studies where deep learning methods are used, feature extraction and selection steps are generally performed by deep learning methods [23]. Respiratory scoring process by engineering-based methods can generally be divided into 4 steps [2,24]. In the first step, records labeled as apnea by the attending physician are taken.
Intelligent retrieval of wear rate prediction for hypereutectoid steel
2023, Multiscale and Multidisciplinary Modeling, Experiments and Design