CN108596043A - The automatic method by stages of sleep of single lead EEG signals based on set empirical mode decomposition - Google Patents
The automatic method by stages of sleep of single lead EEG signals based on set empirical mode decomposition Download PDFInfo
- Publication number
- CN108596043A CN108596043A CN201810273735.XA CN201810273735A CN108596043A CN 108596043 A CN108596043 A CN 108596043A CN 201810273735 A CN201810273735 A CN 201810273735A CN 108596043 A CN108596043 A CN 108596043A
- Authority
- CN
- China
- Prior art keywords
- signal
- parameter
- imf
- sleep
- single lead
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Signal Processing (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The automatic method by stages of sleep of single lead EEG signals based on set empirical mode decomposition, 1) multiple intrinsic mode function IMF are obtained into row set empirical mode decomposition to the single lead EEG signals collected；2) to the multiple intrinsic mode function signals of step 1) and original signal, the extraction of minimum 2 characteristic parameters, i.e. activity Ac and mobility Mo parameters are carried out respectively；3) characteristic parameter for the multiple intrinsic mode function signals and original signal for obtaining step 2) constitutes input matrix, is input in disaggregated model, obtains the result of sleep stage；Single lead EEG signal is resolved into multiple IMF signals by this method, by extracting multiple characteristic parameters, achieves the accuracy rate of higher sleep stage；Cross-validation experiments the result shows that this method have certain generalization ability, it is with a high credibility, can accurately complete sleep stage, for assessment sleep quality provide effective foundation.
Description
Technical field
The present invention relates to one kind based on set empirical mode decomposition (Ensemble Empirical Mode
Decomposition, EEMD) single lead brain electric (electroencephalogram, EEG) signal sleep it is automatic by stages
Method.
Background technology
With the increase of modern life pressure, more and more people by sleep disturbance, (especially exhale by insomnia, drowsiness, sleep
Inhale pause syndrome etc. illness) torment.According to the statistics of the World Health Organization, 27% people has sleep disturbance.Currently, sleep disturbance
A kind of disease with bulletin harmfulness is had been considered as, is paid much attention to by more and more people.Pass through each physiological signal pair
Sleep quality state carries out by stages, is a kind of effective ways of objective evaluation sleep quality.
At present the typical method of clinically monitoring sleep be with lead hypnotic instrument (Polysomnography, PSG) and acquire sleep
Physiological signal during dormancy, including brain electric (EEG), eye electric (EOG), myoelectricity (EMG), electrocardio (ECG), blood oxygen saturation (SpO2)
And breath signal.Sleep stage is to carry out artificial inspectional analysis to PSG signals according to R＆K rules by expert.And PSG signals need to be by
Professional operates, and technology is more demanding, somewhat expensive, and manually to sleep carry out this mode by stages it is very cumbersome,
It takes, accuracy rate and efficiency are very low.Therefore, as can establishing the automatic side by stages of sleep for the EEG signal for being based only on single lead
Method and model, will clinically for the assessment of sleep quality provide it is more easy, quick, accurately assess approach.
Invention content
The object of the present invention is to provide a kind of automatic methods by stages of sleep, are based only on the EEG signal of single lead, lead to
Cross set empirical mode decomposition (Ensemble Empirical Mode Decomposition, EEMD) obtain multiple performances compared with
Good intrinsic mode function (Intrinsic Mode Functions, IMF), to raw EEG signal and obtained IMF signals point
Not carry out characteristic parameter (including statistics parameter, time domain parameter and nonlinear parameter) extraction and constitute input matrix, then
It is input in disaggregated model, obtains reliably sleeping automatically by stages as a result, realizing for sleep stage.
The object of the present invention is achieved like this：One kind is based on set empirical mode decomposition (Ensemble Empirical
Mode Decomposition, EEMD) electric (electroencephalogram, the EEG) signal of single lead brain sleep it is automatic
Method by stages, includes the following steps：
1) the single lead EEG signals collected are obtained multiple intrinsic into row set empirical mode decomposition (EEMD)
Mode function IMF；
2) the multiple intrinsic mode function signals (Intrinsic Mode Functions, IMF) step 1) obtained with
Original signal carries out the extraction of minimum 2 characteristic parameters respectively；Activity Ac and mobility Mo；
3) characteristic parameter for the multiple intrinsic mode function signals and original signal for obtaining step 2) constitutes input matrix,
It is input in disaggregated model, obtains the result of sleep stage.
In the step 1), when to single lead EEG signals into row set empirical mode decomposition, n are obtained admittedly wherein decomposing
There is mode function signal.Preferably, it is described to single lead electric signal into row set empirical mode decomposition when, wherein decompose consolidated
Having mode function signal IMF) number n is 5-10, especially n=7, the standard deviation ε of white Gaussian noise is 0.05-0.5, in particular
0.1, integrated quantity N are 100-1000, especially 200；
In step 2), the extraction of characteristic parameter further includes secondary characteristic parameter：Kurtosis K, complexity Co, zero are passed through
Rate ZCT, Sample Entropy SE；
For original signal or IMF signals, computational statistics parameter kurtosis；
For original signal or IMF signals, calculating time domain parameter Hjorth indexes, (including activity, mobility are complicated
Property) and zero crossing rate；
For original signal or IMF signals, nonlinear parameter Sample Entropy is calculated；
It is described to single lead electric signal into row set empirical mode decomposition when, gather in empirical mode decomposition (EEMD) 3
Parameter area：Intrinsic mode function signal (IMF) number n is 5-10；The standard deviation ε of white Gaussian noise is 0.05-0.5；It is integrated
Quantity N be 100-1000.
It is n=7 especially to decompose and obtain intrinsic mode function signal number, and the standard deviation of white Gaussian noise is 0.1, is integrated
Quantity be 200.
The time domain parameter Hjorth index acquisition patterns are as follows：
Activity describes the amplitude of signal in the time domain, the i.e. variance of signal function；
Mobility describes the slope of signal in the time domain, i.e. square of the ratio between variance of signal first derivative and signal variance
Root；
Complexity describes the slope variation rate of signal in the time domain, the i.e. movement of the mobility of signal first derivative and signal
The ratio between property；
The acquisition pattern of the wherein time domain parameter zero crossing rate is as follows：
2 quadratures adjacent to signal x (t), obtain sequences yi:1≤i≤Q-1, and acquire the point that new sequence intermediate value is less than 0
Number q；Zero crossing rate value by the ratio between q and signal length Q as signal.
In step 2), the extraction of 6 characteristic parameters, including：
For original signal or IMF signals, computational statistics parameter kurtosis；
For original signal or IMF signals, calculating time domain parameter Hjorth indexes, (including activity, mobility are complicated
Property) and zero crossing rate；
For original signal or IMF signals, nonlinear parameter Sample Entropy is calculated；
Wherein the time domain parameter Hjorth index acquisition patterns are as follows：
Activity describes the amplitude of signal in the time domain, the i.e. variance of signal function；
Mobility describes the slope of signal in the time domain, i.e. square of the ratio between variance of signal first derivative and signal variance
Root；
Complexity describes the slope variation rate of signal in the time domain, the i.e. movement of the mobility of signal first derivative and signal
The ratio between property.
The acquisition pattern of the wherein time domain parameter zero crossing rate is as follows：
2 quadratures adjacent to signal x (t), obtain sequences yi:1≤i≤Q-1, and acquire the point that new sequence intermediate value is less than 0
Number q；Zero crossing rate value by the ratio between q and signal length Q as signal.
When wherein the nonlinear parameter Sample Entropy calculates, 3 parameter areas in characteristic parameter Sample Entropy SE：
Embedded dimensions m is 2-3；Similar tolerance r is 0.1-0.2 times of signal standards deviation；Data length is institute in practice
Use data length.Especially Embedded dimensions m is 2, and similar tolerance r is 0.1 times of signal standards deviation；Data length can be
3750。
In the step 3), the characteristic parameter by obtained multiple intrinsic mode function signals and original signal is constituted
Input matrix is input in disaggregated model, and the result for obtaining sleep stage is as follows：
3.1) multiple intrinsic mode function signals of all subjects and the characteristic parameter of original signal are constituted one big
Input matrix；
3.2) all characteristic parameters in input matrix are subjected to respective maximin normalized, obtain normalizing
Change matrix；
3.3) sample of each sleep stage of normalization matrix is randomly divided into five parts, is tested by 5 foldings intersection
Demonstrate,prove composing training sample and test sample；
3.4) training sample is inputted in training pattern and corresponding sleep stage type label trains network, by each
Secondary repetitive exercise model, makes loss function minimize.
6 parameters of the present invention are respectively：Kurtosis, the activity of Hjorth, the mobility of Hjorth, Hjorth's
Complexity, zero crossing rate and Sample Entropy.Other parameters can certainly be excavated again, in general, the inspection of above six parameters
Survey weight or bigger.
Beneficial effects of the present invention：Present invention only requires the EEG signals of single lead, meet convenient, comfortable sleep monitor
It is required that；The intrinsic mode function of multiple better performances is obtained by gathering empirical mode decomposition, by original signal and inherently
The space-time characteristic of EEG signal has fully been excavated in the feature extraction of mode function；Cross-validation experiments are the result shows that this method has
Certain generalization ability, it is with a high credibility, sleep stage can be accurately completed, effective foundation is provided for assessment sleep quality, has
There is good application prospect.
Description of the drawings
Fig. 1 is the automatic method by stages of sleep the present invention is based on single lead EEG signals of set empirical mode decomposition
Functional block diagram.
Fig. 2 is 7 IMF signals that EEG signal of the subject under some DS sleep stage state is decomposed.(a)-
(g) IMF1-7 is indicated respectively.
Specific implementation mode
In order to know more about the technology contents of the present invention, below in conjunction with specific embodiment, and with reference to attached drawing, to the present invention make into
The detailed description of one step.
Fig. 1 is the automatic method by stages of sleep the present invention is based on single lead EEG signals of set empirical mode decomposition
Functional block diagram.
The automatic method by stages of sleep of single lead EEG signals based on set empirical mode decomposition, step include：
1) it to the single lead EEG signals collected, is decomposed into row set empirical mode decomposition, obtains shown in Fig. 27
A intrinsic mode function, calculating include the following steps：
1.1) white Gaussian noise signal w (t) is added to original signal x (t), generates the signal X (t) for including noise；
1.2) local maximum and local minimum of X (t) are determined, and utilizes Cubic Spline Fitting curve by the office of X (t)
Portion's maximum and local minizing point connect, and constitute the coenvelope line u of X (t)1With lower envelope line l1；
1.3) the mean value m of envelope up and down is calculated1=(u1+l1)/2；
1.4) m is subtracted with X (t)1, obtain new sequences h1=X (t)-m1；
If 1.5) h1Meet 2 features of IMF, i.e.,：1. extreme point is equal with the number of zero crossing or at most differs 1
It is a；2. at any time, the mean value of lower envelope line is necessary for 0 thereon, then h1It is exactly an IMF, this season c1=h1；If
It is unsatisfactory for, then makees h1Envelope up and down, acquire its mean value m according to step 2.3)11, calculate h11=h1-m11, then judge h11It is
It is no to meet the condition of IMF, if then enabling c1=h11, otherwise repeatedly step 1.2), 1.3) and 1.4), until h1kMeet the item of IMF
Until part.At this moment c1=h1kFor first IMF component of sequence X (t)；
1.6) first IMF components c is subtracted from X (t)1, obtain sequence of differences：r1=X (t)-c1, with r1To be decomposed
Object, by step 1.2)~1.5) find second IMF components c2, until residue sequence rnOnly there are one Local Extremum or
Become a monotonic function, can not extract IMF again at this time.So far, X (t) can be expressed as：
In formula (1), n is the number of IMF；rn(t) it is final residual error item；ci(t) it is each layer IMF components.
1.7) different white Gaussian noise signals is added to original signal x (t) respectively, repeats step 1.2)~1.6)；
1.8) obtained IMF is integrated into mean value as last decomposition result.By integrated average, can will add into
White Gaussian noise signal influence offset.And increased white Gaussian noise signal can be controlled by statistical law：
In formula (2), ε is the standard deviation for increasing white Gaussian noise, εnFor the standard deviation of final error, N is the quantity integrated.
It is 7 wherein to decompose and obtain intrinsic mode function signal number n, and the standard deviation ε of white Gaussian noise is 0.1, integrated
Quantity N is 200.
2) the 7 intrinsic mode function signals and original signal obtained to step 1), carry out the extraction of 6 characteristic parameters,
Including：Computational statistics parameter kurtosis；Time domain parameter Hjorth indexes (including activity, mobility, complexity) and zero are worn
More rate；Nonlinear parameter Sample Entropy.
Each sensitive features parameter is respectively calculated as follows：
2.1) statistics parameter kurtosis K is calculated as follows：
2.2) time domain parameter Hjorth indexes calculate as follows：
2.2.1) activity Ac describes the amplitudes of signal x (t) in the time domain, the i.e. variance of signal function：
In formula (3), Q indicates the length of signal；
2.2.2) mobility Mo describes the slopes of signal x (t) in the time domain, the i.e. variance of signal first derivative and signal side
The square root of the ratio between difference：
2.2.3) complexity Co describes the slope variation rates of signal x (t) in the time domain, the i.e. mobility of signal first derivative
The ratio between with the mobility of signal：
2.3) time domain parameter zero crossing rate ZCT calculating includes the following steps：
2.3.1) 2 quadratures adjacent to signal x (t), obtain sequences yi:1≤i≤Q-1；
2.3.2) acquire sequences yi:In 1≤i≤Q-1, the number q of point of the value less than 0；
2.3.3) the ZCT values by the ratio between q and signal length Q as signal.
2.4) nonlinear parameter Sample Entropy SE calculating includes the following steps：
2.4.1 signal x (t)) is reconstructed into one group of 2 dimensional vector according to previously given Embedded dimensions 2：
X2(t)={ x (t), x (t+1) } (t=1,2 ..., Q-1)；
2.4.2) according to each t value, X is calculated2(t) with other vectors X2The distance between (k) d [X2(t),X2(k)]；
2.4.3) statistics d [X2(t),X2(k)] it is less than the number B of similar tolerance r=0.15*SD2(t), and calculate and distance
The average value C of the ratio of sum2(r)；
2.4.4 dimension) is increased to 3, repeats step 2.4.1)~2.4.3), obtain C3(r)；
2.4.5) Sample Entropy SE estimated values are：SE (x (t), 2, r)=- ln (C3(r)/C2(r))。
3) characteristic parameter of the multiple intrinsic mode function signals and original signal that in the step 2), obtain is constituted defeated
Enter matrix, be input in disaggregated model, wherein disaggregated model includes one of following models：
3.1) support vector machines (Support Vector Machine, SVM) disaggregated model；
3.2) random forest (Random Forest, RF) disaggregated model；
3.3) extreme gradient boosting algorithm (eXtreme Gradient Boosting, XGB) disaggregated model；
The automatic application by stages slept below to the single lead EEG signal of actual acquisition with this method, in conjunction with
The present invention is further detailed in chart.
Data come from SHHS databases, we according to the demographic information of subject, sleep quality scale and scoring,
The various aspects such as signal quality scoring have filtered out 111 satisfactory Healthy subjects, as shown in table 1.In the signal of acquisition
It is C3-A2 and C4-A1 respectively including two lead EEG signal data, embodiment only chooses C4-A1 leads, sample rate 125Hz,
And by whole section of eeg data according to American Academy of Sleep Medicine (American Academy of Sleep in 2007
Medicine, AASM) rule is divided into mono- section of 30s, per one piece of data on all correspond to the label by stages of an expert evaluation.AASM is advised
Then, sleep was divided into for five phases：Awakening phase (W phases), 1 phase of non-rapid eye movement (N1 phases), 2 phase of non-rapid eye movement (N2 phases), non-rapid
3 phase of eye movement (N3 phases) or sound sleep phase (DS phases), rapid eye movement phase (REM phases).There are a large amount of scholars to merge N1 phases and N2 phases at present
It is slept the phase (LS phases) at shallow.Table 2 lists 111 each quantity by stages of subject sleep.
1. demographic information of table is measured with sleep quality
Pay attention to：If data fit normal distribution, value indicates median [lower quartile, upper quartile]；Otherwise, it is worth
Indicate mean+SD.
2.111 each quantity by stages of subject sleep of table
By 111 subjects, 97014 sections of EEG signal data into row set empirical mode decomposition, every section of EEG signal is decomposed
To 7 intrinsic mode function signals；The extraction of 6 characteristic parameters is carried out simultaneously to original signal and 7 intrinsic mode function signals
Constitute the input matrix of a 97014*48；All characteristic parameters in input matrix are subjected to respective maximin normalizing
Change is handled, and obtains normalization matrix；The sample of each sleep stage of normalization matrix is randomly divided into five parts, is passed through
5 folding cross validation composing training samples and test sample.We split data into two kinds of periodical modes, and one kind including five sleeps
By stages：W phases, N1 phases, N2 phases, DS phases, REM phases, another kind comprising four by stages：W phases, LS phases, DS phases, REM phases.
Training sample is inputted in training pattern and corresponding sleep stage type label trains network, by changing each time
For training pattern, loss function is made to minimize, to obtain two kinds of periodical modes three kinds of disaggregated models as a result, such as 3 institute of table
Show.
The automatic result by stages of the 5 folding cross validations of 3. 3 kinds of models of table
By table 3 the results show that the result of 5 folding cross validations of each disaggregated model under each periodical mode is basic
Unanimously, illustrate that this method has certain generalization ability.For three kinds of disaggregated models, the result of 4 periodical modes is better than 5
The result of periodical mode.For 4 or 5 periodical modes, the classifying quality of XGB disaggregated models is best, svm classifier model
It, RF disaggregated models are worst.The classification average accuracy of XGB disaggregated models can reach 86.0% and 84.6%.Take three kinds of models
Highest one group of accuracy is as a result, calculate each accuracy by stages, as shown in table 4 and 5.
Each automatic result by stages by stages of three 4 periodical modes of disaggregated model in 4. test set of table
Each automatic result by stages by stages of three 5 periodical modes of disaggregated model in 5. test set of table
As can be seen from Table 4, three kinds of disaggregated models show the accuracy highest of W phases, followed by LS phases, are DS later
It is phase, worst for the REM phases.And XGB disaggregated models four by stages on all show highest, W phases 89.9%, LS phases
87.4%, DS phase 87.4%, REM phases 78.2%.
As can be seen from Table 5, three kinds of disaggregated models show the accuracy highest of W phases, followed by N2 phases, are DS later
It is phase, followed by REM phases, worst for the N1 phases.And XGB disaggregated models N1, N2 and DS tri- by stages on all show highest,
N1 phases 9.4%, N1 phases 89.2%, DS phases 86.4%；RF disaggregated models show highest on the W phases, are 90.7%；Svm classifier
Model shows highest on the REM phases, is 82.5%.Other than the result of N1 phases is poor less than 10%, other are tied by stages automatically
Fruit reaches 80% or more.It may be only to account for sleep since data are less by stages for this that we, which analyze the poor reason of N1 phase results,
The 3.6% of overall data, in model training, no the characteristics of preferably learning to this by stages.
In the present invention, the pretreatment of single lead EEG signal uses set empirical mode decomposition.Gather empirical modal point
Solution has become useful non-linear, non-stationary signal the means of analysis at present.Its major advantage is that performance is bad
Signal decomposition is the intrinsic mode function of one group of better performances, can the more acurrate characteristic information for effectively holding former data.So
And it is fairly limited to the research of sleep scoring using set empirical mode decomposition in current research.Therefore, the present invention is innovatively
Sleep EEG signal is handled using set empirical mode decomposition, realize sleep it is automatic by stages.
In the present invention, the extraction of parameter includes statistics parameter, time domain parameter and nonlinear parameter.It is anti-from different sides
The dynamic characteristic for having reflected signal has close be associated with by stages with sleep.It is found in previous research, statistics parameter
(kurtosis), time domain parameter (Hjorth indexes and zero crossing rate) and nonlinear parameter (Sample Entropy), for having for sleep by stages
Separating capacity more outstanding.Therefore, in the present invention, its parameter is combined, obtain relatively reliable sleep it is automatic by stages
Criterion.
Since the EEG signal of single lead is used only in this method, the drawbacks of avoiding conventional sleep by stages.It will not shadow
The sleep quality of subject is rung, operation becomes simple, and expense becomes cheap.Artificial judgment is avoided by stages automatically simultaneously, is no longer consumed
When effort, and there is no subjective factor influence.Meanwhile cross-validation experiments are the result shows that this method has centainly extensive
Ability, it is with a high credibility, sleep stage can be accurately completed, effective foundation is provided for assessment sleep quality, there is good answer
Use foreground.
It these are only the preferred embodiment of the present invention, be not intended to limit the scope of the invention, it is every to utilize this hair
Equivalent structure or equivalent flow shift made by bright specification and accompanying drawing content is applied directly or indirectly in other relevant skills
Art field, is included within the scope of the present invention.
Claims (8)
1. a kind of automatic method by stages of sleep of single lead EEG signals based on set empirical mode decomposition, it is characterized in that packet
Include the following steps：
1) multiple intrinsic mode functions are obtained into row set empirical mode decomposition to the single lead EEG signals collected
IMF；
2) to the multiple intrinsic mode functions of step 1) (Intrinsic Mode Functions, IMF) signal and original signal, divide
The extraction of minimum 2 characteristic parameters, i.e. activity Ac and mobility Mo parameters are not carried out；
3) characteristic parameter for the multiple intrinsic mode function signals and original signal for obtaining step 2) constitutes input matrix, input
Into disaggregated model, the result of sleep stage is obtained；
In the step 1), when to single lead EEG signals into row set empirical mode decomposition EEMD, n are obtained admittedly wherein decomposing
There is mode function signal；
In step 2), the extraction of characteristic parameter further includes secondary characteristic parameter：Kurtosis K, complexity Co, zero crossing rate
ZCT, Sample Entropy SE；
For original signal or IMF signals, computational statistics parameter kurtosis；
For original signal or IMF signals, time domain parameter Hjorth indexes (including activity, mobility, complexity) are calculated
With zero crossing rate；
For original signal or IMF signals, nonlinear parameter Sample Entropy is calculated.
2. the sleep of single lead EEG signals based on set empirical mode decomposition according to claim 1 is automatically by stages
Method, it is characterized in that it is described to single lead electric signal into row set empirical mode decomposition when, gather empirical mode decomposition
(EEMD) 3 parameter areas in：
Intrinsic mode function signal (IMF) number n is 5-10；
The standard deviation ε of white Gaussian noise is 0.05-0.5；
Integrated quantity N is 100-1000.
3. the sleep of single lead EEG signals based on set empirical mode decomposition according to claim 1 is automatically by stages
Method, it is characterized in that the time domain parameter Hjorth index acquisition patterns are as follows：
Activity describes the amplitude of signal in the time domain, the i.e. variance of signal function；
Mobility describes the slope of signal in the time domain, the i.e. square root of the ratio between variance of signal first derivative and signal variance；
Complexity describes the slope variation rate of signal in the time domain, i.e., the mobility of signal first derivative and the mobility of signal it
Than；
The acquisition pattern of the wherein time domain parameter zero crossing rate is as follows：
2 quadratures adjacent to signal x (t), obtain sequences yi:1≤i≤Q-1, and acquire of point of the new sequence intermediate value less than 0
Number q；Zero crossing rate value by the ratio between q and signal length Q as signal.
4. the sleep of single lead EEG signals based on set empirical mode decomposition according to claim 1 is automatically by stages
Method, it is characterized in that when the wherein described nonlinear parameter Sample Entropy calculates, 3 parameter models in characteristic parameter Sample Entropy SE
It encloses：Embedded dimensions m is 2-3；Similar tolerance r is 0.1-0.2 times of signal standards deviation；Data length is data used in practice
Length.
5. the sleep of single lead EEG signals based on set empirical mode decomposition according to claim 1 is automatically by stages
Method, it is characterized in that in the step 3), the feature by obtained multiple intrinsic mode function signals and original signal
Parameter constitutes input matrix, is input in disaggregated model, the result for obtaining sleep stage is as follows：
3.1) multiple intrinsic mode function signals of all subjects and the characteristic parameter of original signal are constituted into a big input
Matrix；
3.2) all characteristic parameters in input matrix are subjected to respective maximin normalized, obtain normalized moments
Battle array；
3.3) sample of each sleep stage of normalization matrix is randomly divided into five parts, passes through 5 folding cross validation structures
At training sample and test sample；
3.4) training sample is inputted in training pattern and corresponding sleep stage type label trains network, by changing each time
For training pattern, loss function is made to minimize.
6. the sleeps of single lead EEG signals according to claim 3 based on set empirical mode decomposition is automatic by stages
Method, characterized in that when wherein the nonlinear parameter Sample Entropy calculates, Embedded dimensions m is 2, and similar tolerance r is signal post
0.1 times of quasi- deviation.
7. the sleeps of single lead EEG signals according to claim 1 based on set empirical mode decomposition is automatic by stages
Method, characterized in that
Step 1) decomposes into row set empirical mode decomposition to the single lead EEG signals collected, obtains 7 natural mode of vibration
Function includes the following steps：
1.1) white Gaussian noise signal w (t) is added to original signal x (t), generates the signal X (t) for including noise；
1.2) local maximum and local minimum of X (t) are determined, and utilizes Cubic Spline Fitting curve by the local pole of X (t)
Big value and local minizing point connect, and constitute the coenvelope line u of X (t)1With lower envelope line l1；
1.3) the mean value m of envelope up and down is calculated1=(u1+l1)/2；
1.4) m is subtracted with X (t)1, obtain new sequences h1=X (t)-m1；
If 1.5) h1Meet 2 features of IMF, i.e.,：1. extreme point it is equal with the number of zero crossing or it is most difference 1；2.
Any time, the mean value of lower envelope line is necessary for 0 thereon, then h1It is exactly an IMF, this season c1=h1；If conditions are not met,
Then make h1Envelope up and down, acquire its mean value m11, calculate h11=h1-m11, then judge h11Whether the condition of IMF is met, if
Then enable c1=h11, otherwise repeatedly step 1.2), 1.3) and 1.4), until h1kUntil the condition for meeting IMF；At this moment c1=h1kFor sequence
Arrange first IMF component of X (t)；
1.6) first IMF components c is subtracted from X (t)1, obtain sequence of differences：r1=X (t)-c1, with r1For the object that is decomposed,
By step 1.2)~1.5) find second IMF components c2, until residue sequence rnOnly there are one Local Extremum or become one
A monotonic function can not extract IMF again at this time；So far, X (t) is expressed as：
In formula (1), n is the number of IMF；rn(t) it is final residual error item；ci(t) it is each layer IMF components；
1.7) different white Gaussian noise signals is added to original signal x (t) respectively, repeats step 1.2)~1.6)；
1.8) obtained IMF is integrated into mean value as last decomposition result；By integrated average, the white Gaussian that will be added into
The influence of noise signal is offset；And increased white Gaussian noise signal is controlled by statistical law：
In formula (2), ε is the standard deviation for increasing white Gaussian noise, εnFor the standard deviation of final error, N is the quantity integrated；
It is 7 wherein to decompose and obtain intrinsic mode function signal number n, and the standard deviation ε of white Gaussian noise is 0.1, integrated quantity N
It is 200；
2) the 7 intrinsic mode function signals and original signal obtained to step 1) carry out the extraction of 6 characteristic parameters, including：
Computational statistics parameter kurtosis；Time domain parameter Hjorth indexes (including activity, mobility, complexity) and zero crossing rate；
Nonlinear parameter Sample Entropy；
Each sensitive features parameter is respectively calculated as follows：
2.1) statistics parameter kurtosis K is calculated as follows：
2.2) time domain parameter Hjorth indexes calculate as follows：
2.2.1) activity Ac describes the amplitudes of signal x (t) in the time domain, the i.e. variance of signal function：
In formula (3), Q indicates the length of signal；
2.2.2) mobility Mo describes the slopes of signal x (t) in the time domain, i.e., the variance of signal first derivative and signal variance it
The square root of ratio：
2.2.3) complexity Co describes the slope variation rates of signal x (t) in the time domain, the i.e. mobility and letter of signal first derivative
Number the ratio between mobility：
2.3) time domain parameter zero crossing rate ZCT calculating includes the following steps：
2.3.1) 2 quadratures adjacent to signal x (t), obtain sequences yi:1≤i≤Q-1；
2.3.2) acquire sequences yi:In 1≤i≤Q-1, the number q of point of the value less than 0；
2.3.3) the ZCT values by the ratio between q and signal length Q as signal；
2.4) nonlinear parameter Sample Entropy SE calculating includes the following steps：
2.4.1 signal x (t)) is reconstructed into one group of 2 dimensional vector according to previously given Embedded dimensions 2：
X2(t)={ x (t), x (t+1) } (t=1,2 ..., Q-1)；
2.4.2) according to each t value, X is calculated2(t) with other vectors X2The distance between (k) d [X2(t),X2(k)]；
2.4.3) statistics d [X2(t),X2(k)] it is less than the number B of similar tolerance r=0.15*SD2(t), and calculate with apart from sum
Ratio average value C2(r)；
2.4.4 dimension) is increased to 3, repeats step 2.4.1)~2.4.3), obtain C3(r)；
2.4.5) Sample Entropy SE estimated values are：SE (x (t), 2, r)=- ln (C3(r)/C2(r))。
8. the sleeps of single lead EEG signals according to claim 1 based on set empirical mode decomposition is automatic by stages
Method, characterized in that by the characteristic parameter structure of the multiple intrinsic mode function signals and original signal that in the step 2), obtain
It at input matrix, is input in disaggregated model, wherein disaggregated model includes one of following models：
3.1) support vector machines (Support Vector Machine, SVM) disaggregated model；
3.2) random forest (Random Forest, RF) disaggregated model；
3.3) extreme gradient boosting algorithm (eXtreme Gradient Boosting, XGB) disaggregated model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810273735.XA CN108596043A (en) | 2018-03-29 | 2018-03-29 | The automatic method by stages of sleep of single lead EEG signals based on set empirical mode decomposition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810273735.XA CN108596043A (en) | 2018-03-29 | 2018-03-29 | The automatic method by stages of sleep of single lead EEG signals based on set empirical mode decomposition |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108596043A true CN108596043A (en) | 2018-09-28 |
Family
ID=63623959
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810273735.XA Pending CN108596043A (en) | 2018-03-29 | 2018-03-29 | The automatic method by stages of sleep of single lead EEG signals based on set empirical mode decomposition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108596043A (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109558863A (en) * | 2019-01-07 | 2019-04-02 | 哈尔滨工业大学（深圳） | EEG signals character representation and extracting method, device and storage medium |
| CN109598222A (en) * | 2018-11-26 | 2019-04-09 | 南开大学 | Wavelet neural network Mental imagery brain electricity classification method based on the enhancing of EEMD data |
| CN109602417A (en) * | 2018-11-23 | 2019-04-12 | 杭州妞诺科技有限公司 | Sleep stage method and system based on random forest |
| CN109674468A (en) * | 2019-01-30 | 2019-04-26 | 大连理工大学 | It is a kind of singly to lead brain electricity sleep mode automatically method by stages |
| CN111466910A (en) * | 2020-04-30 | 2020-07-31 | 电子科技大学 | Sleep monitoring method, system, storage medium, computer program and device |
| CN111493822A (en) * | 2020-03-23 | 2020-08-07 | 济南国科医工科技发展有限公司 | Sleep electroencephalogram based rapid eye movement period sleep behavior disorder classification method |
| CN111772631A (en) * | 2020-07-06 | 2020-10-16 | 天津大学 | Early breast cancer feature extraction method based on integrated empirical mode decomposition |
| CN112545535A (en) * | 2020-12-07 | 2021-03-26 | 杭州沃维医疗科技有限公司 | Sleep-wake cycle analysis method based on amplitude integrated electroencephalogram |
| CN112842363A (en) * | 2021-02-07 | 2021-05-28 | 湖南万脉医疗科技有限公司 | Sleep electroencephalogram detection method and system |
| CN113392412A (en) * | 2021-05-11 | 2021-09-14 | 杭州趣链科技有限公司 | Data receiving method, data sending method and electronic equipment |
| CN113855049A (en) * | 2021-10-22 | 2021-12-31 | 上海电机学院 | Electroencephalogram sleep staging method based on EMD-XGboost |
| CN114305343A (en) * | 2022-01-18 | 2022-04-12 | 山东大学 | Single-lead electroencephalogram sleep staging method based on complementary set empirical mode decomposition |
| CN116808391A (en) * | 2023-06-29 | 2023-09-29 | 北京理工大学 | Sleep awakening method and system based on physiological signal decoding |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102274022A (en) * | 2011-05-10 | 2011-12-14 | 浙江大学 | Sleep state monitoring method based on electroencephalogram signals |
| US20140316230A1 (en) * | 2013-04-22 | 2014-10-23 | Personal Neuro Devices Inc. | Methods and devices for brain activity monitoring supporting mental state development and training |
| CN104809434A (en) * | 2015-04-22 | 2015-07-29 | 哈尔滨工业大学 | Sleep staging method based on single-channel electroencephalogram signal ocular artifact removal |
| US20150313535A1 (en) * | 2014-05-02 | 2015-11-05 | University Health Network | Method and system for sleep detection |
| CN105615834A (en) * | 2015-12-22 | 2016-06-01 | 深圳创达云睿智能科技有限公司 | Sleep stage classification method and device based on sleep EEG (electroencephalogram) signals |
| CN106778865A (en) * | 2016-12-13 | 2017-05-31 | 重庆邮电大学 | A kind of semi-supervised EEG signals sleep stage method under multi-domain characteristics |
| CN107007263A (en) * | 2017-04-25 | 2017-08-04 | 中国科学院计算技术研究所 | The sleep quality assessment method and system of a kind of generalization |
| CN107495962A (en) * | 2017-09-18 | 2017-12-22 | 北京大学 | A kind of automatic method by stages of sleep of single lead brain electricity |
-
2018
- 2018-03-29 CN CN201810273735.XA patent/CN108596043A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102274022A (en) * | 2011-05-10 | 2011-12-14 | 浙江大学 | Sleep state monitoring method based on electroencephalogram signals |
| US20140316230A1 (en) * | 2013-04-22 | 2014-10-23 | Personal Neuro Devices Inc. | Methods and devices for brain activity monitoring supporting mental state development and training |
| US20150313535A1 (en) * | 2014-05-02 | 2015-11-05 | University Health Network | Method and system for sleep detection |
| CN104809434A (en) * | 2015-04-22 | 2015-07-29 | 哈尔滨工业大学 | Sleep staging method based on single-channel electroencephalogram signal ocular artifact removal |
| CN105615834A (en) * | 2015-12-22 | 2016-06-01 | 深圳创达云睿智能科技有限公司 | Sleep stage classification method and device based on sleep EEG (electroencephalogram) signals |
| CN106778865A (en) * | 2016-12-13 | 2017-05-31 | 重庆邮电大学 | A kind of semi-supervised EEG signals sleep stage method under multi-domain characteristics |
| CN107007263A (en) * | 2017-04-25 | 2017-08-04 | 中国科学院计算技术研究所 | The sleep quality assessment method and system of a kind of generalization |
| CN107495962A (en) * | 2017-09-18 | 2017-12-22 | 北京大学 | A kind of automatic method by stages of sleep of single lead brain electricity |
Non-Patent Citations (9)
| Title |
|---|
| AHNAF RASHIK HASSAN等: "Automated identification of sleep states from EEG signals by means of ensemble empirical mode decomposition and random under sampling boosting", 《COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE》 * |
| B. KOLEY等: "An ensemble system for automatic sleep stage classification using single channel EEG signal", 《COMPUTERS IN BIOLOGY AND MEDICINE》 * |
| J. ZHANG等: "Assessing severity of obstructive sleep apnea by fractal dimension sequence analysis of sleep EEG", 《PHYSICA A》 * |
| JANTO SKOWRONEK等: "Features for audio classification: Percussiveness of sounds", 《INTELLIGENT ALGORITHMS IN AMBIENT AND BIOMEDICAL COMPUTING》 * |
| 刘跃雷: "一种新的人睡眠EEG自动分期方法研究", 《中国优秀博硕士学位论文全文数据库(硕士) 医药卫生科技辑》 * |
| 李敏等: "《信号分析与处理的软件硬件实现》", 31 January 2014 * |
| 葛家怡: "睡眠分期及低频磁场睡眠诱导的研究", 《中国优秀博硕士学位论文全文数据库(博士) 医药卫生科技辑》 * |
| 闫佳运等: "基于单路脑电的睡眠分期判别方法研究", 《中国医疗设备》 * |
| 陈尚勤等: "《近代语音识别》", 31 March 1993 * |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109602417A (en) * | 2018-11-23 | 2019-04-12 | 杭州妞诺科技有限公司 | Sleep stage method and system based on random forest |
| CN109598222B (en) * | 2018-11-26 | 2023-04-07 | 南开大学 | EEMD data enhancement-based wavelet neural network motor imagery electroencephalogram classification method |
| CN109598222A (en) * | 2018-11-26 | 2019-04-09 | 南开大学 | Wavelet neural network Mental imagery brain electricity classification method based on the enhancing of EEMD data |
| CN109558863A (en) * | 2019-01-07 | 2019-04-02 | 哈尔滨工业大学（深圳） | EEG signals character representation and extracting method, device and storage medium |
| CN109674468A (en) * | 2019-01-30 | 2019-04-26 | 大连理工大学 | It is a kind of singly to lead brain electricity sleep mode automatically method by stages |
| CN111493822A (en) * | 2020-03-23 | 2020-08-07 | 济南国科医工科技发展有限公司 | Sleep electroencephalogram based rapid eye movement period sleep behavior disorder classification method |
| CN111466910A (en) * | 2020-04-30 | 2020-07-31 | 电子科技大学 | Sleep monitoring method, system, storage medium, computer program and device |
| CN111466910B (en) * | 2020-04-30 | 2023-11-21 | 电子科技大学 | Sleep monitoring method, system, storage medium, computer program and device |
| CN111772631A (en) * | 2020-07-06 | 2020-10-16 | 天津大学 | Early breast cancer feature extraction method based on integrated empirical mode decomposition |
| CN111772631B (en) * | 2020-07-06 | 2021-12-14 | 天津大学 | Early breast cancer feature extraction method based on integrated empirical mode decomposition |
| CN112545535A (en) * | 2020-12-07 | 2021-03-26 | 杭州沃维医疗科技有限公司 | Sleep-wake cycle analysis method based on amplitude integrated electroencephalogram |
| CN112842363A (en) * | 2021-02-07 | 2021-05-28 | 湖南万脉医疗科技有限公司 | Sleep electroencephalogram detection method and system |
| CN113392412A (en) * | 2021-05-11 | 2021-09-14 | 杭州趣链科技有限公司 | Data receiving method, data sending method and electronic equipment |
| CN113855049A (en) * | 2021-10-22 | 2021-12-31 | 上海电机学院 | Electroencephalogram sleep staging method based on EMD-XGboost |
| CN114305343A (en) * | 2022-01-18 | 2022-04-12 | 山东大学 | Single-lead electroencephalogram sleep staging method based on complementary set empirical mode decomposition |
| CN116808391A (en) * | 2023-06-29 | 2023-09-29 | 北京理工大学 | Sleep awakening method and system based on physiological signal decoding |
| CN116808391B (en) * | 2023-06-29 | 2023-12-26 | 北京理工大学 | Sleep awakening method and system based on physiological signal decoding |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108596043A (en) | The automatic method by stages of sleep of single lead EEG signals based on set empirical mode decomposition | |
| CN110811609B (en) | Epileptic spike intelligent detection device based on self-adaptive template matching and machine learning algorithm fusion | |
| CN107569226B (en) | The method and application of HRV are obtained based on piezoelectric sensing | |
| CN109843163A (en) | For marking dormant method and system | |
| CN103584840B (en) | Automatic sleep stage method based on electroencephalogram, heart rate variability and coherence between electroencephalogram and heart rate variability | |
| CN104757975B (en) | Tidal breathing pattern is distinguished using oxygen saturation signal | |
| US7623912B2 (en) | Method, apparatus and system for characterizing sleep | |
| Belgacem et al. | ECG based human authentication using wavelets and random forests | |
| CN106388818B (en) | The characteristics information extraction method and system of sleep state monitoring model | |
| CN108143411A (en) | A kind of tranquillization state brain electricity analytical system towards Autism Diagnostic | |
| CN105147248A (en) | Physiological information-based depressive disorder evaluation system and evaluation method thereof | |
| CN109674468A (en) | It is a kind of singly to lead brain electricity sleep mode automatically method by stages | |
| CN204931634U (en) | Based on the depression evaluating system of physiologic information | |
| CN110074765A (en) | A kind of method of discrimination of SAHS screening and sleep stage | |
| CN103584872A (en) | Psychological stress assessment method based on multi-physiological-parameter integration | |
| CN109363670A (en) | A kind of depression intelligent detecting method based on sleep monitor | |
| CN109106350A (en) | The preliminary screening method of the Sleep Apnea-hypopnea Syndrome of low physiological stress | |
| KR102378278B1 (en) | The biological signal analysis system and biological signal analysis method for operating by the system | |
| CN110135285B (en) | Electroencephalogram resting state identity authentication method and device using single-lead equipment | |
| CN107545906A (en) | Lung Sounds processing method, processing equipment and readable storage medium storing program for executing | |
| CN103405225B (en) | A kind of pain that obtains feels the method for evaluation metrics, device and equipment | |
| CN106618611A (en) | Sleeping multichannel physiological signal-based depression auxiliary diagnosis method and system | |
| CN108420406A (en) | Method based on pulse wave sleep stage | |
| CN111317446B (en) | Sleep structure automatic analysis method based on human muscle surface electric signals | |
| CN106333676B (en) | The annotation equipment of EEG signals data type under waking state |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180928 |