阅读说明：本技术 一种智能睡眠时相检测与睡眠质量评估系统及方法 (Intelligent sleep time phase detection and sleep quality evaluation system and method ) 是由 陈卫 王娟 赖思渝 于 2019-09-26 设计创作，主要内容包括：本发明属于睡眠状态监测技术领域,公开了一种智能睡眠时相检测与睡眠质量评估系统及方法,脑电信号采集模块,用于采集睡眠脑电数据,通过脑电信号检测部件完成睡眠脑电数据的采集,后台服务器,用于接收采集的原始睡眠脑电数据,完成睡眠脑电信号处理,实现睡眠脑电自动分期,并对睡眠质量做出评价。网上服务平台,用于跟踪用户睡眠趋势,提供睡眠心理咨询及专家建议等服务。所述智能睡眠时相检测与睡眠质量评估方法包括：脑电信号获取、预处理、特征提取、模式识别分类、睡眠时相输出。本发明的评价结果可推送至移动终端,为用户提供专业睡眠心理咨询和专家建议,完全不依赖操作者工作经验,可作为临床诊断有益的补充和说明。(The invention belongs to the technical field of sleep state monitoring, and discloses an intelligent sleep time phase detection and sleep quality evaluation system and method. And the online service platform is used for tracking the sleep trend of the user and providing services such as sleep mental consultations, expert suggestions and the like. The intelligent sleep phase detection and sleep quality assessment method comprises the following steps: acquiring electroencephalogram signals, preprocessing, extracting features, identifying and classifying modes and outputting sleep time phases. The evaluation result of the invention can be pushed to the mobile terminal, provide professional sleep psychological consultation and expert suggestion for the user, completely do not depend on the working experience of the operator, and can be used as beneficial supplement and explanation of clinical diagnosis.)

1. An intelligent sleep time phase detection and sleep quality assessment system, characterized in that the intelligent sleep time phase detection and sleep quality assessment system comprises:

the electroencephalogram signal acquisition module is used for acquiring sleep electroencephalogram data and finishing acquisition of the sleep electroencephalogram data through the electroencephalogram signal detection component;

the background server is used for receiving the acquired original sleep electroencephalogram data, finishing sleep electroencephalogram signal processing, including signal preprocessing, feature extraction and feature classification, realizing automatic sleep electroencephalogram stage division and evaluating sleep quality;

and the online service platform is used for tracking the sleep trend of the user and providing services such as sleep mental consultations, expert suggestions and the like.

2. An intelligent sleep-time phase detection and sleep quality assessment method for executing the intelligent sleep-time phase detection and sleep quality assessment system of claim 1, wherein the intelligent sleep-time phase detection and sleep quality assessment method comprises: acquiring electroencephalogram signals, preprocessing, extracting features, identifying and classifying modes and outputting sleep time phases.

3. The intelligent sleep-time phase detection and sleep quality assessment method according to claim 2, wherein the sleep-brain electrical automatic staging of the intelligent sleep-time phase detection and sleep quality assessment method comprises the steps of:

dividing sleep data into small segments of 30s according to the sleep staging standard, and multiplying the sampling frequency by the time length of the data segments to obtain the total data point of each segment of data;

step two, zeroing signal baselines, and unifying the amplitudes into mV;

thirdly, preprocessing the signal by using a 0.5-35HZ Butterworth band-pass filter to remove interference signals;

selecting an FIR filter or a wavelet packet decomposition feature extraction method to enter a sleep automatic staging system; firstly, constructing an ideal FIR band-pass filter by using a window function, wherein the unit impulse response corresponding to the window function is obtained by Fourier inverse transformation, the response is truncated into a finite term by using a finite-length window function, the finite term is used as a unit impulse response sequence of the filter, and the sequence is finally used for approximating the ideal FIR filter; the parameters are selected, the allowable error of the passband ripple is set to be 0.1, the amplitude is set to be 1, and the stopband ripple is not more than 0.02dB of the allowable error; constructing 7 FIR band-pass filters according to the frequency band range of each characteristic wave of the electroencephalogram signal, and filtering out K synthesized wave, delta wave, theta wave, alpha wave, fusiform wave and beta wave₁Wave sum beta₂The percentage of each wave in the total energy of the wave band; adopting bio97 wavelet base and Shannon function as cost function, executing 5-layer wavelet packet decomposition to the EEG, taking wavelet packet coefficient corresponding to the most similar frequency segment of each characteristic wave to reconstruct waveform, and separating each characteristic wave; combining the total energy, the energy ratio of alpha waves to theta waves and the energy ratio of delta waves to theta waves with 3 energy characteristics to jointly form 10 energy characteristic values, carrying out staging on 10 characteristics of sleep data acquired by an FIR filter method and a wavelet packet decomposition method by using an LS-SVM model respectively, comparing the staging with standard experts, and selecting a characteristic extraction method with higher average accuracy;

step five, extracting the Lempel-Ziv complexity of the electroencephalogram signal, and performing coarse granulation processing on the electroencephalogram time sequence to obtain a sequence S consisting of 0 and 1; repeatedly executing copy and insert operations on the sequence to form a new sequence segmented by specific characters, wherein any subsequence at the end of the new sequence cannot appear in the original sequence; representing the relative KC complexity of the new sequence by the difference between the number of segments of the new sequence and the limit value thereof, and adding the new feature into the 10 energy features;

and step six, inputting characteristics of the LS-SVM multi-classifier. Performing feature dimensionality reduction by adopting a Principal Component Analysis (PCA) method, and representing original 11 feature vectors as linear combinations of a plurality of features;

step seven, constructing an LS-SVM multi-classifier, adding an error square sum term into a standard SVM objective function, adopting a radial basis kernel function RBF to replace high-dimensional inner product operation, mapping an input sample to a high-dimensional feature space, and converting the problem into the solution of a linear equation set under the KKT condition; the classification mode selects a one-to-one mode of setting an LS-SVM classifier between any two types of samples;

step eight, training the LS-SVM multi-classifier by using the characteristics of the electroencephalogram samples of the three groups of sc experimenters and the calibrated stages thereof; mixing the three groups of recorded samples, extracting 2/3 samples as a training set, extracting 1/3 samples as a test set, and stopping the step after the model training is finished;

step nine, respectively using the electroencephalogram sample characteristics of the three groups of st experimenters to send into a designed LS-SVM multi-classifier for automatic identification, recording the stage result, comparing the stage result with the originally calibrated stage, and testing the generalization capability of the designed LS-SVM multi-classifier;

and step ten, analyzing and outputting the automatic staging result.

4. The intelligent sleep-phase detection and sleep quality assessment method according to claim 2, wherein the sleep quality assessment employs EEG-based sleep assessment, and the sleep measurement indicators mainly include sleep progress indicators, sleep structure indicators, and REM measurement values; wherein the sleep process index comprises total sleep time, waking frequency, sleep ratio, sleep efficiency, sleep maintenance rate, sleep latency, wake-up time, and exercise wake-up time; sleep structure index refers to the percentage of total sleep time for each sleep stage N1, N2, N3, and REM; REM measured values refer to REM sleep latency and REM activity, REM intensity, REM density, REM cycle number;

sleep efficiency:

in formula (1): t is_REM+NREMIs the sum of the time of the REM phase and the NREM phase, T_allTime, T, of the data recorded in the experiment_REM+NREMAnd sum of wake time;

time ratio of sleep stages:

in formula (2): t is_xFor a certain sleep period T_XSum of occupied time, T_allTime, T, of the data recorded in the experiment_REM+NREMAnd sum of wake time;

sleep ratio:

in the formula (3), the normal adult REM sleep time accounts for 20-25% of the total sleep time; REM time analyzed by the intelligent sleep time phase detection and quality evaluation system accounts for 21.53% of total sleep time, and the time proportion of each sleep period is within a normal value range, so that the user is judged to have normal sleep, good sleep quality and no sleep disorder disease.

Technical Field

The invention belongs to the technical field of sleep state monitoring, and particularly relates to an intelligent sleep time phase detection and sleep quality evaluation system and method.

Background

Currently, the closest prior art:

in the field of sleep medicine, Polysomnography (PSG) is a relatively common sleep phase detection and sleep quality evaluation criterion. The system can synchronously acquire multi-conductive physiological signals such as electroencephalogram (EEG), eye movement (EOG), heart rhythm (ECG), muscle activity (EMG) and the like, and integrates the signals to realize sleep quality evaluation. PSG can eliminate or diagnose sleep disorders such as narcolepsy, pathological somnolence, periodic limb movement disorders, insomnia, and sleep apnea. Although the advantages of PSG are clear, it also has application limitations. The patient must be monitored in exclusive diagnosis and treatment institution, and for the patient whose sleep quality is not high originally, the adaptability in strange environment is poor, and the measured data accuracy is not high. In addition, PSG is not suitable for mid-and long-term sleep monitoring, and it is not practical for patients to purchase a specialized kit, and also requires professional skill from the operator. Meanwhile, a plurality of biosensors are required to be worn by a patient, so that the comfort level is poor, and the patient is not favorable to falling asleep to a certain extent.

Some commercial sleep monitoring devices rely on head-mounted sensors to monitor brain and muscle signals through the PSG for sleep, but this invasive monitoring method is detrimental to the health of the subject and must be based on the PSG. Some wristband-mounted sleep monitoring devices perform sleep-wake assessment using event logging, and while effective at monitoring sleep activity, analysis of wake time for patients with mild quality of sleep is not accurate. The patient is required to preset a schedule to inform the sleep device of the time to fall asleep and the time to wake up in advance, otherwise the accuracy of sleep quality evaluation is greatly affected. Other non-invasive devices perform sleep quality assessment by collecting environmental data, but these instrumented devices ignore physiological factors of the patient and improve the accuracy of the assessment.

Some researchers developed sleep quality monitoring applications that utilized smart sensors on cell phones that were placed at bed to provide sleep quality assessment by collecting signals collected by on-board accelerometers and microphones. However, such procedures require a duration of time to assess the sleep stages and sleep quality, rather than merely recording the corresponding sleep-related activity, but also place certain restrictions on the brand and hardware configuration of the handset. Other researchers have predicted their long-term sleep trends, sleep state perception schedules, and quality classifications by analyzing the "handset-constrained" behavior of different populations. However, these programs require users to place the mobile phone at a reasonable position in a matching manner, and the problems of mobile phone disconnection, background information push and battery use all become adverse factors affecting sleep quality monitoring, so that the programs are not suitable for long-term sleep tracking.

Electroencephalograms are modern auxiliary examination methods which help diagnose diseases by recording weak bioelectricity of the brain itself in an electroencephalograph in an amplifying manner to form a curve. In clinical application, the ear lobe, the nose tip or the mastoid part is generally used as a zero potential point on the body, and the potential difference between the electrode placed at the point and the electrodes at other parts on the scalp is the recorded electroencephalogram signal.

Electroencephalographic waveforms can vary greatly when the brain is exposed to different conditions (e.g., activation, drowsiness, sleep, etc.). Electroencephalographic waveforms can be classified into four basic types, mainly according to their frequencies.

Delta wave: the frequency is O.5-3.5 times per second, and the amplitude is 20-200 μ v. Normal adults have little delta wave while awake, but delta waves can occur during sleep. Generally considered, a slow wave of high amplitude.

(delta or theta waves) may be the primary manifestation of electrical activity when the cerebral cortex is in a state of inhibition.

θ wave: the frequency is 4-7 times per second, and the amplitude is 20-150 μ v. Theta waves may appear when an adult is drowsy. In the infancy stage, theta waves are commonly seen, and definite alpha waves do not appear until the age of ten.

Alpha wave: the frequency is 8-13 times per second, and the amplitude is 20-100 μ v. When normal people are awake, quiet and eye-closed, the alpha wave can appear, the amplitude of the alpha wave is changed from small to large and then from large to small, and the alpha wave is periodically changed repeatedly in such a way to form a fusiform shape of the alpha wave. Each alpha wave is in the form of a shuttle lasting about 1-2 seconds. When the subject opens his eyes or receives other excitatory stimuli (e.g., performing a mental calculation), the alpha wave disappears immediately and turns into a fast wave, called "alpha wave block". Thus, alpha waves are considered to be the primary manifestation of electrical activity when the cerebral cortex is in a conscious and quiescent state. The factors such as frequency, amplitude and spatial distribution of alpha wave are important indexes for reflecting the brain function state.

Beta wave: the frequency is 14 to 30 times per second, and the amplitude is 5 to 20 μ v. When the subject opens his eyes to see objects and performs a thinking activity, a beta wave appears. It is generally accepted that the beta wave causes the primary manifestation of electrical activity when the cerebral cortex is in a state of tension.

During sleep, the electroencephalogram undergoes a variety of different changes, which vary with the depth of sleep. Sleep can be divided into two states according to different characteristics of electroencephalogram: non-rapid eye movement sleep (NREM sleep) and rapid eye movement sleep (REM sleep).

The rapid movement sleep stage of the non-eyeball, the muscle of the whole body is relaxed, the eyeball does not move, and the visceral parasympathetic nerve activity is dominant. Heart rate and respiration are slowed, blood pressure is reduced, gastrointestinal motility is increased, basal metabolic rate is low, brain temperature is slightly reduced when the brain is more awake, and total cerebral blood flow is reduced when the brain is more awake. The rapid non-eyeball movement sleep is divided into four stages by electroencephalogram characteristics, namely a sleep onset stage, a light sleep stage, a moderate sleep stage and a deep sleep stage. In the first stage of electroencephalogram, the wave is mainly theta wave, spindle wave or K comprehensive wave does not appear, actually, the reaction to external stimulation is weakened in the transition stage from complete waking to sleeping, mental activities enter a floating boundary, and thinking and reality are disconnected; in the second stage, the brain waves are spindle waves and K combined waves, the delta waves are less than 20%, and actually, a person enters real sleep; in the third stage, the delta wave in the brain wave occupies 20 to 50 percent, and the sleep is in medium-depth sleep; in the fourth stage, delta waves in brain waves account for more than 50%, and people are in a deep sleep state and are not easy to wake up in the fourth stage. The 3-4 stage sleep is deep sleep in the general sense, and the arousal threshold value is the highest at the moment.

In the rapid eye movement sleep stage, desynchronized low-amplitude brain waves with mixed frequencies appear. The rapid movement of eyeballs, a lot of paroxysmal small twitching of facial and limb muscles, sometimes or when the sucking action of lips occurs, the throat makes a short sound, hands and feet shake, the activity of internal organs is highly unstable, breathing is irregular, heart rate often changes, gastric acid secretion is increased, cerebral blood flow and metabolism are increased, the discharge activity of cerebral neurons in most areas is increased, the temperature of brain tissues is increased, and the oxygen consumption of brain is obviously increased compared with that of waking. The arousal threshold value of the rapid eyeball movement sleep is higher than that of NREM1 sleep and is between NREM 2-3 sleep.

In the whole night sleep, REM sleep and NREM sleep alternate in intervals of about 90-100 minutes, and the change period is called a sleep period. Normal persons sleep first in NREM sleep stage, and rapidly in phase 2, 3, 4 and continuing from phase 1. The REM sleep occurs after the NREM sleep period lasts for 80-120 minutes, the next REM sleep is started after the NREM sleep period lasts for several minutes, a circulation period of the NREM sleep and the REM sleep is formed, the REM sleep occurs every 90 minutes on average, and the REM sleep duration is gradually prolonged as the time is closer to the later period of sleep. Each time lasts for 10-30 minutes. The NREM-REM sleep cycle is repeatedly circulated for 3-5 times in the whole sleep period, the periods of each period are not necessarily complete, but all start from the period 1, the sleep depth in each period becomes shallow in the morning and does not reach the period 4 any more, and as can be seen from the cycle transition of NREM sleep and REM sleep, the sleep process does not continue from shallow to deep to bright as soon as the sleep is started, but a deep burst, a shallow burst and deep and shallow sleep are continuously alternated.

In addition to studies on sleep regularity, studies on some neurotransmitters and chemicals inside the brain have found that: neurotransmitters inside the brain, such as: endogenous morphin (or endorphin), 5-HT (5 hydroxytryptamine), gamma-aminobutyric acid (GABA) and the like have the effects of calming and relaxing, and can restore the comfortable and healthy environment in the brain; neuronal released stimulants such as: dopamine, acetylcholine, serotonin and the like in the brain can improve the symptoms of listlessness, attention loss, thought loss and the like in the daytime caused by insomnia; stress hormones in the brain, including Adrenaline (ADR), Norepinephrine (NE), glucocorticoids (cortisol, corticosterone), angiotensin I (Aug positive), etc., can promote the brain to be in a state of tension and excitement, and the phenomena of accelerated heartbeat, vasoconstriction, etc.

According to the above rules, it can be found that the duration of NREM3 phase and 4 phase is longer, and the duration of REM phase is shorter, when a person is in deep sleep. In addition, when the components promoting the sedation and pleasure of people in the brain are increased, and the components making people feel nervous and excited are inhibited, the brain can have a good rest and is more helpful for deep sleep. The treatment of insomnia is basically based on this sleep principle.

Normal people experience several relatively stable states throughout their night's sleep, and in order to better describe sleep, Rechtschaffen and Kales classify sleep as stage 6 (i.e. R & K criteria) based on the appearance of Electroencephalogram (EEG), Electrooculogram (EOG), Electromyogram (EMG) during sleep:

wake period (W), rapid eye movement sleep period (REM), and 4 non-rapid eye movement sleep periods (NREM). NREM is divided into sleep 1 (S1), sleep 2 (S2), sleep 3 (S3) and sleep 4 (S4), where sleep 1 and 2 are Light Sleep (LS) and sleep 3 and 4 are deep sleep (also called Slow Wave Sleep (SWS)). Currently, international American Academy of Sleep Medicine (AASM) modified R & K sleep staging standards are commonly used internationally, which merge periods S3, S4 of the R & K guidelines into one period, and denote the awake period, rapid eye movement sleep period, light sleep period, and deep sleep period by W, R, N1, N2, and N3, respectively.

During the sleep process, NREM and REM are alternately carried out, so that 4-6 NREM-REM sleep cycles are formed, and each sleep cycle lasts for 90-120 minutes. During normal adult sleep, stages of sleep regularly occur in sequence from W-N1-N2-N3-N2-REM, followed by repeating N2-N3-N2-REM, and so on, connected to each other, and so on. Wherein the REM period is about 100 minutes and accounts for 20% -25% of the total sleep time.

A sleep aid instrument is theoretically designed, through detecting the change curve of brain waves of a human body, a chip is used for adjusting the parameter intensity and duration of electronic pulses to stimulate acupuncture points such as the head part of the general meeting, the sleep aiding and the mental, and meanwhile, comprehensive means such as stimulation of adjusting sound frequency, smell and the like are combined to improve the brain waves and help people to enter deep sleep.

Standard of brain waves during sleep:

the four basic brain waves are: DELTA waves (DELTA/DELTA wave), THETA waves (THETA/THETA wave), ALPHA waves (ALPHA/ALPHA wave), and BETA waves (BETA/BETA wave). These four brain waves constitute an electroencephalogram (EEG). Electroencephalograms are the display of waves in the brain, but the voltages of waves in the brain are small, being only a few millionths of a volt.

(1) Waking eyes open, and beta wave when brain activity is stressed.

(2) Clear head, quiet, alpha wave when closing eyes; .

(3) Absentmindedness theta waves.

(4) Delta wave after falling asleep.

(5) The sleep includes slow wave sleep and fast wave sleep.

(6) The slow wave I period is the sleep period, alpha wave is gradually reduced, low-amplitude theta wave and beta wave G are irregularly mixed together, and brain wave is in a flat trend.

(7) Phase II is a light sleep phase with sigma waves and a small number of delta waves.

(8) Stage iii is a moderate sleep stage, with high amplitude delta waves, or kappa waves (a complex of delta and sigma waves).

(9) The IV phase is a deep sleep phase, and delta waves appear.

(10) Fast wave sleep is manifested as an irregular beta wave.

Sleep staging table

According to the standard of the american society for sleep medicine 2007, the electroencephalogram standard for sleep staging is as follows:

according to electroencephalographic studies, sleep can be divided into four stages. In 2007, the american society for sleep medicine incorporated the third and fourth stages into one stage, the third stage. Thus, it can also be said that the sleep stage includes three stages.

First stage

First stage (about 10 minutes): a mild sleep stage. This is a transition sleep where the alpha wave seen in the relaxed, awake state begins to become irregular and gradually disappear, while the eye rotates slowly. At this time, the theta wave gradually appears, and the frequency is lower by 4-7 cps. The sleeper's body is slowly relaxed and breathes slowly, but is easily awakened by an external stimulus.

Second stage

Second stage (about 20 minutes): sleep becomes deeper and its EEG shows occasional 8-14Hz concussions, called "sleep spindles" (sleep spindles), which are short bursts of high frequency, large amplitude brain waves. At this stage, the sleeper is difficult to wake up.

The third stage

Third stage (about 40 minutes): at this time, a high amplitude and slow delta wave appears, and the eye and body movements disappear, sometimes also referred to as "sleep spindles".

Fourth stage

Fourth stage (about 20 minutes): in the deep sleep stage, a broad EEG rhythm of 2Hz or less occurs, and phenomena such as sleeptalking, somntalking, bed wetting, and the like may occur.

Rapid eye movement sleep stage

The first four stages of sleep, after approximately 60-90 minutes, appear to have entered first stage sleep, but rather than repeating the above process, a completely new stage, the rapid eye movement sleep (REM) stage, is entered.

In the REM sleep stage, the electrophysiological activity of the brain changes rapidly, the delta wave disappears, and high-frequency, low-amplitude brain waves appear. The eyeball of the sleeper starts to move left, right, up and down rapidly, and dreams are accompanied. The heart rhythm and blood pressure become irregular and breathing becomes rapid, but the muscles remain relaxed.

The 1 st REM sleep generally lasts for 5-10 minutes, and after 4 stages of sleep, the 2 nd REM sleep will last for a longer time. The last REM sleep was for up to 1 hour.

Features of periodic circulation

Each cycle typically lasts 90 minutes and is repeated 4-6 times per night. Deep sleep occurs in far more time in the first half of the night than in the latter half of the night. When dawn is approaching, the third and fourth stages of sleep gradually disappear.

Sound: double sound racket (BB)

The simplest way to stimulate the brain is by sound, however, the frequency of sound sufficient to effectively stimulate the brain is too low for a person to hear. This requires the use of a special technique, known as the Binaural Beat Technology (BBT).

At the same time as a steady sound stimulation of 500Hz for the left ear and 510Hz for the right ear, 2 similar but different tones will be integrated in the brain, and a frequency difference of 10Hz (the so-called third sound) will be sensed by the brain, while the brain waves are very effectively loaded, thus shifting the EEG to the 10Hz alpha mode as well. When stereo headphones are used, the sound of the left and right channels is integrated only to the brain. This frequency difference, when perceived by the brain, is called a Binaural Beat (BB).

Acupoint selection: baihui point

Baihui acupoint is located at the highest point of human body, and is the meeting acupoint of governor vessel, bladder meridian of foot taiyang, gallbladder meridian of foot shaoyang, triple energizer meridian of hand shaoyang and liver meridian of foot jueyin, and yang qi passed through all meridians meet this point, so it is also called "Sanyang Wuhui". Therefore, Baihui points can reach yin and yang meridians and link up the channels and collaterals of the whole body; when applied to Baihui, it acts on one acupoint to regulate the balance between yin and yang.

Research shows that acupuncture on Baihui acupoint has therapeutic effect on insomnia from the perspective of neurophysiology. Modern researches also show that acupuncture on Baihui acupoint has the effects of increasing brain blood supply, repairing neurons, enhancing activity of intracerebral acetylcholinesterase, regulating content of 5-hydroxyindaceneacetic acid (5-HIAA), and the like, and can regulate the body state of insomnia patients from multiple angles.

The brain-computer interface bci (brain computer interface) is a novel human-computer interaction mode, and realizes the interaction between the human brain and a computer or other electronic equipment based on electroencephalogram (eeg), which is independent of peripheral nerves and muscle tissues outside the human body. Therefore, the technology shows good development in the fields of medical treatment, games, industrial control and the like. In the field of computers, brain-computer interface technology is widely applied to communication and control technology and the like. Brain wave therapy techniques are also increasingly being advanced.

The sleep quality evaluation can be performed by using a psychological scale method and a physiological parameter detection method, and the scale method has the defects of strong subjectivity, objective physiological parameter method, large amount of signal processing and analysis, and limitation of the accuracy of sleep staging on the result, so that the most accurate method is to perform comprehensive evaluation on the sleep quality by combining sleep electroencephalogram data analysis, a Pittsburgh sleep quality index scale (PSQI) and the chief complaint sleep condition.

In summary, the problems of the prior art are as follows:

(1) polysomnography analysis has clear requirements on monitoring places, needs professional acquisition equipment and fixed experimental environments, and is not beneficial to acquisition of patient data. Wear more biosensor, make the person who is testee feel uncomfortable, influence its normal sleep activity. The technical level of an operator has certain requirements, the equipment debugging is time-consuming, medical staff mainly carries out manual staging, the working experience of the operator is excessively depended, and the labor capacity is easily increased.

(2) Invasive commercial monitoring devices require polysomnography as a background for analysis and are detrimental to the health of the subject. Some devices, while effective for sleep monitoring, are not suitable for determining wake-up times for patients with mild quality of sleep. Other devices relying on environmental data for sleep quality assessment do not fully consider physiological signals of patients and are not high in accuracy.

(3) The sleep quality monitoring program developed by the mobile phone sensor not only limits the brand and hardware configuration of the mobile phone, but also regulates the monitoring duration, and the short time can not meet the data acquisition requirement. When the sleep state and the sleep trend are predicted through the application program, the requirements on the positioning of the mobile phone, the network state and other interference factors are higher, and the method is not suitable for long-term sleep tracking.

The difficulty of solving the technical problems is as follows:

patients with sleep disorders generally cannot bear professional clinical sleep monitoring equipment, and cannot train the patients to independently control the equipment through professional skills, and clinical sleep monitoring generally needs 7 to 8 hours, and the specific conditions are determined according to the specific efficiency of each hospital. Some commercial monitoring devices can replace PSG for data acquisition, but do not fully consider physiological signals of patients, and the monitoring precision is not enough to meet medical standards. The sleep monitoring program developed by the mobile phone sensor not only meets the requirements on the mobile phone brand and hardware configuration, but also refers to a psychological scale method for analyzing the sleep quality too much, so a large number of subjective factors are introduced, and the objectivity of analysis is influenced.

The significance of solving the technical problems is as follows:

sleep disorder seriously affects the health of people and becomes a worldwide problem. Therefore, the research and development of a simple and reliable sleep quality tracking analysis system has important clinical significance for improving the diagnosis and treatment effect of the sleep disorder. With the miniaturization and household development direction of the sleep monitoring equipment, the sleep time phase detection and the sleep quality evaluation by using the single-channel electroencephalogram signal become an important development direction of sleep disorder analysis. If the sleep analysis system can be matched with a background data processing platform and the participation of a sleep medical expert, the household intelligent sleep analysis system has higher professional degree, can quickly analyze the sleep data, gives a relevant report through the mobile terminal and provides professional sleep psychological consultation and suggestion. The method is beneficial supplement and attempt of intelligent sleep electroencephalogram analysis, and provides an important means for sleep health condition evaluation.

Disclosure of Invention

Aiming at the problems in the prior art, the invention provides an intelligent sleep time phase detection and sleep quality evaluation system.

The invention is realized in such a way, and provides an intelligent sleep time phase detection and sleep quality evaluation system. The intelligent sleep phase detection and sleep quality assessment system comprises:

the system comprises an electroencephalogram signal acquisition module, a background server and an online service platform;

the electroencephalogram signal acquisition module is used for acquiring sleep electroencephalogram data and finishing acquisition of the sleep electroencephalogram data through the electroencephalogram signal detection component;

the background server is used for receiving the acquired original sleep electroencephalogram data, finishing sleep electroencephalogram signal processing, including signal preprocessing, feature extraction, feature classification and the like, realizing automatic sleep electroencephalogram stage division and evaluating sleep quality;

and the online service platform is used for tracking the sleep trend of the user and providing services such as sleep mental consultations, expert suggestions and the like.

Further, the system for automatically staging the sleep brain electricity comprises: acquiring electroencephalogram signals, preprocessing, extracting characteristics, identifying and classifying modes, and outputting sleep time phases.

Further, the sleep electroencephalogram automatic staging method comprises the following steps:

dividing sleep data into small segments of 30s according to the sleep staging standard, and multiplying the sampling frequency by the time length of the data segments to obtain the total data point of each segment of data;

step two, zeroing signal baselines, and unifying the amplitudes into mV;

thirdly, preprocessing the signal by using a 0.5-35HZ Butterworth band-pass filter to remove interference signals;

and step four, preferentially selecting an FIR filter or a wavelet packet decomposition feature extraction method to enter a sleep automatic staging system. Firstly, an ideal FIR band-pass filter is constructed by using a window function, the corresponding unit impulse response of the filter is obtained by Fourier inversion, the response is truncated into a finite term by using a finite length window function, the finite term is used as a unit impulse response sequence of the filter, and finally the sequence is used for approximating the ideal FIR filter. The parameters are selected such that the passband ripple tolerance is set to 0.1, the amplitude is set to 1, and the stopband ripple is not greater than 0.02dB of tolerance. Constructing 7 FIR band-pass filters according to the frequency band range of each characteristic wave of the electroencephalogram signal, and filtering out K synthesized wave, delta wave, theta wave, alpha wave, fusiform wave and beta wave₁Wave sum beta₂The waves each account for a percentage of the total energy of the wave band. Then adopting bio97 wavelet base and Shannon function as cost function to apply 5-layer wavelet packet decomposition to the EEG signal, taking wavelet packet coefficient corresponding to the most similar frequency segment of each characteristic wave to reconstruct waveform, and separating outThe characteristic waves are described above. Combining the total energy, the energy ratio of alpha waves to theta waves and the energy ratio of delta waves to theta waves with 3 energy characteristics to jointly form 10 energy characteristic values, carrying out staging on 10 characteristics of sleep data acquired by an FIR filter method and a wavelet packet decomposition method by using an LS-SVM model respectively, comparing the staging with standard experts, and selecting a characteristic extraction method with higher average accuracy;

and fifthly, extracting the Lempel-Ziv complexity (KC complexity) of the electroencephalogram signal. The brain electrical time sequence is subjected to coarse graining treatment to obtain a sequence S consisting of (0) and (1). The 'copy' and 'insert' operations are repeatedly executed on the sequence to form a new sequence segmented by specific characters, and any sub-sequence at the end of the new sequence cannot appear in the original sequence. Representing the relative KC complexity of the new sequence by the difference between the number of segments of the new sequence and the limit value of the new sequence, taking the new sequence as a new characteristic and adding the new characteristic into the 10 energy characteristics;

and step six, inputting characteristics of the LS-SVM multi-classifier. Performing feature dimensionality reduction by adopting a Principal Component Analysis (PCA) method, and representing original 11 feature vectors as linear combinations of a plurality of features;

and step seven, constructing the LS-SVM multi-classifier. Adding an error square sum term into a standard SVM objective function, adopting a Radial Basis Function (RBF) to replace high-dimensional inner product operation, mapping an input sample to a high-dimensional feature space, and converting the problem into the solution of a linear equation set under the KKT condition. The classification mode selects a one-to-one mode of setting an LS-SVM classifier between any two types of samples;

and step eight, training the LS-SVM multi-classifier by using the characteristics of the electroencephalogram samples of the three groups of sc experimenters and the calibrated stages of the electroencephalogram samples. Mixing the three groups of recorded samples, extracting 2/3 samples as a training set, extracting 1/3 samples as a test set, and stopping the step after the model training is finished;

step nine, respectively using the electroencephalogram sample characteristics of the three groups of st experimenters to send into a designed LS-SVM multi-classifier for automatic identification, recording the stage result, comparing the stage result with the originally calibrated stage, and testing the generalization capability of the designed LS-SVM multi-classifier;

and step ten, analyzing and outputting the automatic staging result.

Furthermore, the sleep quality assessment adopts EEG-based sleep assessment, and the sleep measurement indexes mainly comprise a sleep progress index, a sleep structure index and an REM measurement value. The sleep progress index includes total sleep time (time taken to wake up during the period subtracted from the time taken to go to sleep to the last wake up), number of awakenings (cumulative sum of the total sleep awakening times), ratio of sleep to total sleep time, sleep efficiency (i.e., ratio of total sleep time to total recorded time), sleep maintenance rate (i.e., ratio of total sleep time to time taken to go to sleep to the last wake up), sleep latency (i.e., time taken to go to sleep until S1), wake-up time (i.e., time taken to wake up from the last wake up), and exercise wake-up time (time taken to be incompletely awakened due to body movement during sleep), etc. The sleep structure index refers to the percentage of total sleep time (i.e., the time ratio of sleep stages) for each sleep stage N1, N2, N3, and REM. REM measurements refer to REM sleep latency (time to fall asleep until the first REM appears) and REM activity, REM intensity, REM density, number of REM cycles, etc. The following is the calculation of several key sleep index parameters.

Sleep efficiency:

in formula (1): t is_REM+NREMIs the sum of the time of the REM phase and the NREM phase, T_allTime, T, of the data recorded in the experiment_REM+NREMAnd the sum of the wake time.

Time ratio of sleep stages:

in formula (2): t is_xFor a certain sleep period T_XSum of occupied time, T_allTime, T, of the data recorded in the experiment_REM+NREMAnd the sum of the wake time.

Sleep ratio (ratio of wake time to total sleep time):

the REM sleep time of the normal adult in the formula (3) is 20 to 25% of the total sleep time, and is about 100 minutes. The REM time analyzed by the sleep electroencephalogram automatic staging system approximately accounts for the total sleep time, and whether the time proportion of each sleep stage is larger or not can be judged, so that whether the sleep of the user is normal or not, whether the sleep quality is better or not, whether sleep disorder diseases such as insomnia exist or not can be judged.

Furthermore, the intelligent sleep time phase detection and sleep quality evaluation system finishes sleep electroencephalogram data acquisition through the electroencephalogram signal detection component and transmits the acquired data to a background for processing. And (3) processing sleep electroencephalogram signals on a background server, including signal preprocessing, feature extraction, feature classification and the like, realizing automatic staging of sleep electroencephalogram and evaluating the sleep quality. In addition, after each experiment is finished, the testee is required to complete a Pittsburgh sleep quality index scale (PSQID, the PSQID simultaneously inquires about the tested chief complaint sleep condition, the sleep quality result analyzed by electroencephalogram signals is combined with the score of the scale and the self complaint condition to give a comprehensive sleep condition evaluation result, and finally the analyzed result is sent to an online service platform and a mobile terminal, the mobile terminal provides service functions of timing, alarm clock, awakening, inquiring, setting, suggesting and displaying a sleep stage map, a sleep quality evaluation result and the like, and the online service platform tracks the sleep trend of the user, and provides services of sleep psychological information inquiry, expert suggestion and the like.

In summary, the advantages and positive effects of the invention are: the invention provides an intelligent sleep time phase detection and sleep quality evaluation system, which collects electroencephalogram signals of three contacts of Fp1, Fpz and Fp2 through a forehead single-channel electroencephalogram collector, and transmits initial data to a background for processing after being dried by a Butterworth band-pass filter through a data preprocessing flow (figure 5).

Discrete electroencephalogram signals are serialized through a processing program after being processed through a characteristic extraction and model construction flow (figure 6), the discrete electroencephalogram signals are input into a constructed LS-SVM multi-classifier after being extracted through characteristics, a supervised machine learning method is adopted to train a classification model, an optimal sample set is divided through cross validation to improve the generalization capability of the model, and finally, sleep time phase analysis results are output according to the model, so that automatic stage division of electroencephalogram data is realized. The model precision is further optimized along with the accumulation of the number of samples, and the prediction model is further stabilized along with the research depth. The intelligent sleep time phase detection and quality evaluation system is suitable for various sleep therapeutic instruments, can provide patient sleep partition data on the basis of the prior art, can also be used as a decision standard for carrying out real-time auxiliary treatment, and can be matched with an appropriate control means to construct an acupoint, sound and smell auxiliary treatment system which is autonomously excited according to partition conclusions. The evaluation result can be pushed to the mobile terminal, professional sleep psychological consultation and expert suggestion are provided for the user, the working experience of an operator is completely not depended on, and the evaluation result can be used as beneficial supplement and explanation of clinical diagnosis.

Drawings

Fig. 1 is a diagram illustrating classification accuracy evaluation of intelligent sleep phase detection and sleep quality evaluation according to an embodiment of the present invention.

Fig. 2 is a flowchart of a system for intelligent sleep phase detection and sleep quality assessment according to an embodiment of the present invention.

Fig. 3 is a flowchart of a sleep electroencephalogram automatic staging system provided by the embodiment of the invention.

FIG. 4 is a method for automatically staging sleep electroencephalogram according to an embodiment of the present invention;

in the figure: 1. an electroencephalogram signal acquisition module; 2. a background server; 3. and (4) an online service platform.

Fig. 5 is a partial code screenshot of data preprocessing provided by an embodiment of the present invention.

Fig. 6 is a code screenshot of a feature extraction and model building part provided by the embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

The following detailed description of the principles of the invention is provided in connection with the accompanying drawings.

As shown in fig. 2, the system for detecting sleep time phase and evaluating sleep quality in an intelligent sleep mode according to an embodiment of the present invention includes:

the system comprises an electroencephalogram signal acquisition module 1, a background server 2 and an online service platform 3;

the electroencephalogram signal acquisition module 1 is used for acquiring sleep electroencephalogram data and completing acquisition of the sleep electroencephalogram data through an electroencephalogram signal detection component;

the background server 2 is used for receiving the acquired original sleep electroencephalogram data, finishing sleep electroencephalogram signal processing, including signal preprocessing, feature extraction, feature classification and the like, realizing automatic sleep electroencephalogram stage division and evaluating sleep quality;

and the online service platform 3 is used for tracking the sleep trend of the user and providing services such as sleep mental consultations, expert suggestions and the like.

As shown in fig. 3, the system for automatically staging sleep electroencephalogram provided by the embodiment of the present invention includes: acquiring electroencephalogram signals, preprocessing, extracting characteristics, identifying and classifying modes, and outputting sleep time phases.

As shown in fig. 4, the sleep electroencephalogram automatic staging method provided by the embodiment of the present invention includes the following steps:

s101: according to the foregoing sleep staging standard, the sleep data is first divided into 30s segments (3000 points since fs is 100 HZ);

s102: zeroing signal baselines, and unifying amplitudes into mV;

s103: preprocessing the signal by using a 0.5-35HZ Butterworth band-pass filter to remove interference signals;

s104: extracting electroencephalogram characteristic waves by using an FIR band-pass filter and wavelet packet decomposition to obtain 10 energy characteristics of the electroencephalogram, and selecting a better one of the two methods to enter a final sleep automatic staging system;

s105: extracting the complexity (namely KC complexity) of the EEG signal Lempel-Ziv, and combining the complexity with 10 energy characteristics;

s106: input features of the LS-SVM multi-classifier;

s107: constructing an LS-SVM multi-classifier, and setting relevant parameters in the LS-SVM multi-classifier before training;

s108: training the LS-SVM multi-classifier by using the characteristics of the electroencephalogram samples of the three groups of sc experimenters and the calibrated stages thereof, wherein the step is not performed after the training is finished;

s109: respectively sending the electroencephalogram sample characteristics of the three groups of st experimenters into a designed LS-SVM multi-classifier for automatic identification, recording a stage result, comparing the stage result with an original calibrated stage, and testing the generalization capability of the designed LS-SVM multi-classifier;

s110: and analyzing and outputting the automatic staging result.

The sleep electroencephalogram automatic staging method provided by the embodiment of the invention specifically comprises the following steps:

dividing sleep data into small segments of 30s according to the sleep staging standard, and multiplying the sampling frequency by the time length of the data segments to obtain the total data point of each segment of data;

step two, zeroing signal baselines, and unifying the amplitudes into mV;

thirdly, preprocessing the signal by using a 0.5-35HZ Butterworth band-pass filter to remove interference signals;

and step four, preferentially selecting an FIR filter or a wavelet packet decomposition feature extraction method to enter a sleep automatic staging system. Firstly, an ideal FIR band-pass filter is constructed by using a window function, the corresponding unit impulse response of the filter is obtained by Fourier inversion, the response is truncated into a finite term by using a finite length window function, the finite term is used as a unit impulse response sequence of the filter, and finally the sequence is used for approximating the ideal FIR filter. The parameters are selected such that the passband ripple tolerance is set to 0.1, the amplitude is set to 1, and the stopband ripple is not greater than 0.02dB of tolerance. Constructing 7 FIR band-pass filters according to the frequency band range of each characteristic wave of the electroencephalogram signal, and filtering out K synthesized wave, delta wave, theta wave, alpha wave, fusiform wave and beta wave₁Wave sum beta₂Wave of eachAs a percentage of the total energy of the band. And then, adopting a bio97 wavelet basis and a Shannon function as a cost function, performing 5-layer wavelet packet decomposition on the electroencephalogram signal, taking wavelet packet coefficients corresponding to the most similar frequency bands of the characteristic waves to reconstruct a waveform, and separating the characteristic waves. Combining the total energy, the energy ratio of alpha waves to theta waves and the energy ratio of delta waves to theta waves with 3 energy characteristics to jointly form 10 energy characteristic values, carrying out staging on 10 characteristics of sleep data acquired by an FIR filter method and a wavelet packet decomposition method by using an LS-SVM model respectively, comparing the staging with standard experts, and selecting a characteristic extraction method with higher average accuracy;

and fifthly, extracting the Lempel-Ziv complexity (KC complexity) of the electroencephalogram signal. The brain electrical time sequence is subjected to coarse graining treatment to obtain a sequence S consisting of (0) and (1). The 'copy' and 'insert' operations are repeatedly executed on the sequence to form a new sequence segmented by specific characters, and any sub-sequence at the end of the new sequence cannot appear in the original sequence. Representing the relative KC complexity of the new sequence by the difference between the number of segments of the new sequence and the limit value of the new sequence, taking the new sequence as a new characteristic and adding the new characteristic into the 10 energy characteristics;

and step six, inputting characteristics of the LS-SVM multi-classifier. Performing feature dimensionality reduction by adopting a Principal Component Analysis (PCA) method, and representing original 11 feature vectors as linear combinations of a plurality of features;

and step seven, constructing the LS-SVM multi-classifier. Adding an error square sum term into a standard SVM objective function, adopting a Radial Basis Function (RBF) to replace high-dimensional inner product operation, mapping an input sample to a high-dimensional feature space, and converting the problem into the solution of a linear equation set under the KKT condition. The classification mode selects a one-to-one mode of setting an LS-SVM classifier between any two types of samples;

and step eight, training the LS-SVM multi-classifier by using the characteristics of the electroencephalogram samples of the three groups of sc experimenters and the calibrated stages of the electroencephalogram samples. Mixing the three groups of recorded samples, extracting 2/3 samples as a training set, extracting 1/3 samples as a test set, and stopping the step after the model training is finished;

step nine, respectively using the electroencephalogram sample characteristics of the three groups of st experimenters to send into a designed LS-SVM multi-classifier for automatic identification, recording the stage result, comparing the stage result with the originally calibrated stage, and testing the generalization capability of the designed LS-SVM multi-classifier;

and step ten, analyzing and outputting the automatic staging result.

Furthermore, the sleep quality assessment adopts EEG-based sleep assessment, and the sleep measurement indexes mainly comprise a sleep progress index, a sleep structure index and an REM measurement value. The sleep progress index includes total sleep time (time taken to wake up during the period subtracted from the time taken to go to sleep to the last wake up), number of awakenings (cumulative sum of the total sleep awakening times), ratio of sleep to total sleep time, sleep efficiency (i.e., ratio of total sleep time to total recorded time), sleep maintenance rate (i.e., ratio of total sleep time to time taken to go to sleep to the last wake up), sleep latency (i.e., time taken to go to sleep until S1), wake-up time (i.e., time taken to wake up from the last wake up), and exercise wake-up time (time taken to be incompletely awakened due to body movement during sleep), etc. The sleep structure index refers to the percentage of total sleep time (i.e., the time ratio of sleep stages) for each sleep stage N1, N2, N3, and REM. REM measurements refer to REM sleep latency (time to fall asleep until the first REM appears) and REM activity, REM intensity, REM density, number of REM cycles, etc. The following is the calculation of several key sleep index parameters.

Sleep efficiency:

in formula (1): t is_REM+NREMIs the sum of the time of the REM phase and the NREM phase, T_allTime, T, of the data recorded in the experiment_REM+NREMAnd the sum of the wake time.

Time ratio of sleep stages:

in formula (2): t is_xFor a certain sleep period T_XSum of occupied time, T_allTime, T, of the data recorded in the experiment_REM+NREMAnd the sum of the wake time.

Sleep ratio (ratio of wake time to total sleep time):

in the formula (3), the normal adult REM sleep time accounts for 20-25% of the total sleep time and is 100 minutes; REM time analyzed by the intelligent sleep time phase detection and quality evaluation system accounts for 21.53% of total sleep time, and the time proportion of each sleep period is within a normal value range, so that the user is judged to have normal sleep, good sleep quality and no sleep disorder diseases such as insomnia and the like.

Furthermore, the intelligent sleep time phase detection and sleep quality evaluation system finishes sleep electroencephalogram data acquisition through the electroencephalogram signal detection component and transmits the acquired data to a background for processing. And (3) processing sleep electroencephalogram signals on a background server, including signal preprocessing, feature extraction, feature classification and the like, realizing automatic staging of sleep electroencephalogram and evaluating the sleep quality. In addition, after each experiment is finished, the testee is required to complete a Pittsburgh sleep quality index scale (PSQID, the PSQID simultaneously inquires about the tested chief complaint sleep condition, the sleep quality result analyzed by electroencephalogram signals is combined with the score of the scale and the self complaint condition to give a comprehensive sleep condition evaluation result, and finally the analyzed result is sent to an online service platform and a mobile terminal, the mobile terminal provides service functions of timing, alarm clock, awakening, inquiring, setting, suggesting and displaying a sleep stage map, a sleep quality evaluation result and the like, and the online service platform tracks the sleep trend of the user, and provides services of sleep psychological information inquiry, expert suggestion and the like.

In the working process of the intelligent sleep time phase detection and sleep quality evaluation system, an LS-SVM multi-classifier is constructed through the working steps, a certain verification set is taken as an example, the verification result is shown in figure 1, the processing time of each sample (data of one person overnight) is only about 16s, the prediction accuracy rate reaches 92.9%, the prediction results of 0-4 sleep stages are quite effective as can be seen from a confusion matrix, and the interpretability of the prediction conclusion of the method is further improved along with the accumulation of the number of samples and the enhancement of the generalization capability of the model. According to the cooperation of the staging conclusion and the control system, when the sleep condition of a user is not ideal, the sleep-assisting equipment is automatically started to realize accurate control of sleep assistance, and sleep time phase detection and intelligent staging are carried out through the acquired electroencephalogram data, and auxiliary treatment is carried out automatically.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.