CN109674468A - It is a kind of singly to lead brain electricity sleep mode automatically method by stages - Google Patents
It is a kind of singly to lead brain electricity sleep mode automatically method by stages Download PDFInfo
- Publication number
- CN109674468A CN109674468A CN201910088652.8A CN201910088652A CN109674468A CN 109674468 A CN109674468 A CN 109674468A CN 201910088652 A CN201910088652 A CN 201910088652A CN 109674468 A CN109674468 A CN 109674468A
- Authority
- CN
- China
- Prior art keywords
- wave
- feature
- data
- formula
- frequency
- 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
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/369—Electroencephalography [EEG]
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/48—Other medical applications
- A61B5/4806—Sleep evaluation
- A61B5/4809—Sleep detection, i.e. determining whether a subject is asleep or not
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/48—Other medical applications
- A61B5/4806—Sleep evaluation
- A61B5/4812—Detecting sleep stages or cycles
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/48—Other medical applications
- A61B5/4806—Sleep evaluation
- A61B5/4815—Sleep quality
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7225—Details of analog processing, e.g. isolation amplifier, gain or sensitivity adjustment, filtering, baseline or drift compensation
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/725—Details of waveform analysis using specific filters therefor, e.g. Kalman or adaptive filters
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Veterinary Medicine (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- Physics & Mathematics (AREA)
- Public Health (AREA)
- Pathology (AREA)
- Signal Processing (AREA)
- Psychiatry (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physiology (AREA)
- Psychology (AREA)
- Power Engineering (AREA)
- Anesthesiology (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
It is a kind of singly to lead brain electricity sleep mode automatically method by stages the invention belongs to sleep monitor technical field, comprising the following steps: (1) that Signal Pretreatment, (2) characteristic of division extract, (3) sleep stage.The invention has the following advantages that acquisition quality preferably singly leads EEG signals first is that devising Preprocessing Algorithm;Second is that the present invention extracts multiple features from time domain, frequency domain, non-linear field, representative feature is filtered out;Third is that the present invention uses Random Forest model, the method does not have to concern overfitting, has good noise resisting ability, and can be derived that each decision tree by stages as a result, to the fiducial probability that obtains random forest to each sleep stage;Fourth is that present invention combination D-S evidence theory further increases the accuracy rate of sleep stage.
Description
Technical field
Brain electricity sleep mode automatically method by stages is singly led the present invention relates to a kind of, belongs to sleep monitor technical field.
Background technique
Sleep is the needs of life, is important one of the physiological activity of people.It can AASM mark according to newest U.S.'s sleep medicine
Standard, sleep can be divided into lucid interval W, non-rapid eye movement phase N1, N2, N3, and rapid eye movement phase REM, wherein N1 and N2 is shallowly to sleep the phase, N3
For the deep sleep phase.Enough deep sleep durations have very big facilitation for relieving fatigue degree, improving working efficiency;
Also there is certain relationships between different sleep stages for cardiovascular and cerebrovascular disease incidence rate;There is the disease in terms of some brains
Disease is not easy to be discovered under waking state, but as the increase of Depth of sleep, brain activity degree reduce, these lesions are begun to
It displays, and also different in the symptom that different sleep stages displays.Therefore, accurate sleep stage seems especially
It is important.
Current clinical sleep monitor technology mostly uses sleep analysis monitor equipment (Polysomnography, PSG).PSG
It is general to require to carry out in sleep laboratory, synchronous acquisition EEG signals (Eletroencephalogram, EEG), electrocardiosignal
(Electrocardiograph, ECG), electro-ocular signal (Electrooculogram, EOG), electromyography signal
More conductive physiological signals such as (Electromyography, EMG), in summary electricity physiological signal realizes sleep stage.It is insufficient
Place is: (1) being worn on subject head using a large amount of electrode, subject is made not feel good, influence its normal sleep
Activity;(2) PSG price costly, and need profession acquisition equipment and suitable experimental situation;(3) it needs experienced
Associated specialist carries out sleep stage, and consuming time is long, increase the workload of expert;(4) result contains certain subjectivity by stages
Factor.Therefore, the sleep mode automaticallies of EEG signals is led based on less lead or singly, and technology becomes the important of sleep monitor field by stages
One of developing direction.
Summary of the invention
For the deficiencies in the prior art, brain electricity sleep mode automatically side by stages is singly led it is an object of the present invention to provide a kind of
Method.This method is extracted and is screened representative brain electrical feature by handling to singly leading EEG signals, establish sleep point
Phase model is realized automatically by stages in conjunction with Finite State Machine and D-S evidence theory, can be widely applied to sleep monitor technology neck
Domain.
In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is that: it is a kind of singly to lead brain electricity sleep mode automatically by stages
Method, comprising the following steps:
Step 1, Signal Pretreatment, it is contemplated that original EEG signals contain various noise jammings, so first having to original
EEG signals are pre-processed, and following sub-step is specifically included:
(A) baseline drift is removed, it is a kind of nonstationary random signal that baseline drift, which belongs to low-frequency noise, is become based on small echo
The multiresolution property changed removes baseline drift using the method for Wavelet decomposing and recomposing;
(B) Hz noise is removed, Hz noise intensity is big, and is ubiquitous in unshielded environments, it is necessary to logical
It crosses dedicated trapper and filters out Hz noise, good effect can be reached using 4 rank IIR trappers, trap frequency is set as
50Hz;
(C) it goes unless EEG signals ingredient, sleep cerebral electricity signal include following several brain electrical feature waves: α wave frequency range
For 8~13Hz, β wave frequency range be 13~30Hz, δ wave frequency range is 1~4Hz, θ wave frequency range is 4~8Hz, sawtooth
Wave frequency range is 2~6Hz, spindle wave frequency range is 12~14Hz, 1 wave frequency range of β is 13~17Hz, 2 wave frequency rate model of β
It encloses for 17~30Hz;For the signal component except these frequencies, it is that other bioelectrical activities generate by human body, needs to filter
It removes;Using 10 rank IIR Butterworth bandpass filters, it is contemplated that the effective component of EEG signals sets bandpass filtering frequency
For 0.2~40Hz;
(D) signature waveform is extracted, is prepared to extract time-frequency domain characteristic of division in next step, using WAVELET PACKET DECOMPOSITION
EEG signals are decomposed into α wave, β wave, δ wave, θ wave, sawtooth by the method for wavelet packets decomposition, that is, WPD
Wave, spindle wave, 1 wave of β and each species rhythm ingredient of 2 wave of β, the i.e. careful division on time-frequency plane by EEG signals;In wavelet analysis
On the basis of theory, the concept of optimal base selection is introduced, after frequency band is passed through multi-level division, according to analyzed brain telecommunications
Number feature, adaptively choose best basic function, be allowed to match with EEG signals, to improve the analysis energy of EEG signals
Power;
Step 2, characteristic of division extract, including temporal signatures extract, frequency domain character extracts and Nonlinear feature extraction, by institute
After having feature extraction, Feature Selection is carried out, high-quality, representative feature is filtered out and forms characteristic of division vector;
Consider each sleep stage duration and analyze the relationship of data information amount, setting is once slept with the data of every τ=30s
Dormancy Installment calculation, eeg data point total number per treatment are N=fsτ, wherein fsIndicate the sampling of eeg data to be analyzed
Frequency, characteristic of division extraction specifically include following sub-step:
(A) temporal signatures extract, including variance var, zero-crossing rate, Hjorth Mobility, Hjorth Complexity,
75 percentiles, α wave energy, β wave energy, δ wave energy, θ wave energy, sawtooth wave energy, spindle wave energy, 1 wave energy of β and β 2
Wave energy;Variance var and 75 percentiles in temporal signatures is common, is not further described;Zero-crossing rate is that the symbol of signal becomes
Change ratio, i.e. eeg data becomes negative from positive number and becomes the total number of positive number and the ratio of eeg data point total number from negative
Value, zero-crossing rate ZCR are described by formula (1),
Wherein, xin(1≤i≤N) indicates that the signal amplitude of i-th n eeg data, sgn () are sign function, passes through public affairs
Formula (2) is described,
Hjorth Mobility indicates average frequency, is calculated by formula (3),
Wherein, m0Indicate the variance of eeg data x, m2Indicate the variance of eeg data x first derivative;
Hjorth Complexity indicates frequency variation, is calculated by formula (4),
Wherein m4Indicate the variance of eeg data x second dervative;
All kinds of feature wave energies, i.e. α wave energy, β wave energy, δ wave energy, θ wave energy, sawtooth wave energy, spindle wave energy
Amount, 1 wave energy of β and 2 wave energy Energy of β are calculated by formula (5) respectively,
(B) frequency domain character extracts, including α wave, β wave, δ wave, θ wave absolute power spectrum, α wave, β wave, δ wave, θ wave relative power
Spectrum, brain electrical feature wave power ratio, be its absolute power spectrum ratio, i.e. δ/α, δ/β, δ/θ, θ/α, θ/β, α/β, (δ+θ)/(α+
β)、α×β1/δ、θ×β2/ δ, relative power spectrum is the ratio of its absolute power spectrum with general power spectrum, and spectra calculation uses
Eeg data x is divided into L sections by Welch method, and every section of length is M, finds out each section of power spectrum respectively, then average, in turn
Calculate power spectrum;Welch method allows each Duan Douyou partial data to be overlapped, and using Hamming window, 50% overlapping carries out cyclic graph
Power estimation, if l sections of power spectrum isIt is described by formula (6),
Wherein, 1≤m≤M, 1≤l≤L,Indicate m-th of eeg data point, w in l segment datamIndicate added window
Function, f are frequency, and window function average energy U is calculated by formula (7),
Power spectrum after average is described by formula (8),
The power spectral density and frequency of all kinds of characteristic waves is calculated, and calculates its area under the curve, obtains absolute power
Spectrum, relative power spectrum, power ratio;
(C) Nonlinear feature extraction, for the nonlinear characteristic used for Sample Entropy, the value of Sample Entropy is bigger, and sample sequence is just
More complicated, the calculating of Sample Entropy SampleEnt specifically includes following sub-step:
(a) eeg data x is imported, x is formed into the Vector Groups X that one group of Embedded dimensions is dim in sequence1,X2,…,
Xi,…,XN-dim+1, wherein XiIt is described by formula (9), xiIndicate the signal amplitude of i-th of eeg data of eeg data x;
Xi=[xi,xi+1,…,xi+dim-1] (9)
(b)XiWith XjI-th and j-th of vector in Vector Groups are indicated, by XiWith XjBetween distance d [Xi,Xj] it is defined as two
Maximum one of element difference in a vector, is described, x by formula (10)jIndicate j-th of eeg data of eeg data x
Signal amplitude;
(c) for meeting the i and j of formula (10), d [X is countedi,Xj] be less than " similarity " metric r number, that is, mould
Plate coupling number calculates the ratio of this number Yu vector total number, is denoted as average template coupling numberBy formula (11) into
Row description,
(d)Average value Adim(r), it is calculated by formula (12),
(e) Embedded dimensions dim+1 forms dim+1 n dimensional vector n, repeats sub-step (a)~(d) in above-mentioned sub-step (C),
Obtain Adim+1(r);
(f) Sample Entropy is defined, is calculated by formula (13),
(g) when N is finite value, Sample Entropy is calculated by formula (14),
(D) feature extracted and artificial sleep stage result composition characteristic Vector Groups are based on filtration method by Feature Selection
In minimal redundancy maximum correlation carry out Feature Selection, score according to correlation each feature, in the feature, sample
Correlation is big between this entropy and classification marker, therefore retains this feature, and other feature is mainly carried out from time-domain and frequency-domain feature
Screening;Set the feature quantity of quasi- selection, maximize the correlation between feature and classified variable, minimize feature and feature it
Between correlation, to filter out representative feature;
Brain electrical feature after feature extraction and Feature Selection is reformulated training set and surveyed by step 3, sleep stage
Examination collection, obtains the model of sleep stage using the method for random forest, and carry out sleep stage verifying, obtains each sleep stage
Fiducial probability, finally by fiducial probability carry out D-S evidence theory fusion, singly led brain electrosleep by stages as a result, sleep
Following sub-step is specifically included by stages:
(A) data set is divided, by temporal signatures, nonlinear characteristic and sleep stage result form first data set
Frequency domain character and sleep stage result are formed second data set Data2 by Data1;
(B) above-mentioned two data set is randomly selected 80% data as training set, 20% data by model training respectively
It as test set, training Random Forest model model1 and model2, and is verified by test set, obtains each sleep point
The fiducial probability of phase specifically includes following sub-step:
(a) random sampling has in the slave training set put back to using bootstrap method and extracts TN times, constitutes TN sample,
The sample not being pumped to every time forms the outer data Out-of-bag of TN bag, and OBB is used to estimate the prediction accuracy of model;
(b) TN decision tree is constructed, nf feature is equipped with, randomly selects FN feature at each node of every one treeThe information content contained by calculating each feature, the most classification capacity of selection one in FN feature
Feature carries out node split, and each tree is grown to the maximum extent, do not do it is any cut out, to construct TN decision tree;
(c) decision tree sets number optimizing, and the range of usual TN is 100~800, and sub-step is utilized in value range
(B) bag not being pumped in sub-step (a) outer data estimate decision-tree model accuracy, when accuracy highest, it is corresponding most
Excellent decision tree number TN*;
(d) random forest is constructed, fiducial probability is calculated, TN* decision tree is combined into composition Random Forest model,
Application data set Data1 obtains model1, repeats the above method and is obtained with data set Data2 training Random Forest model
Test data is input to the model by model2, obtains the probability P by stages to W, N1, N2, N3, REM fivetw, wherein t=
1,2, w=1,2 ... 5, PtwIndicate that sleep stage is w-th of number by stages and total decision tree number in t-th of Random Forest model
Purpose ratio;
(C) D-S evidence theory data fusion, D-S evidence theory data fusion method refer to Dempster rule of combination,
Referred to as Evidence Combination Methods formula, if Θ is identification framework, by sleep stage W, N1, N2, N3, REM, it is uncertain by stages with empty set Φ group
At single element proposition identification framework, A represents any proposition in identification framework, and two Random Forest models are defined on identification framework
Basic probability assignment function masst(A) (0,1) ∈, t=1,2 meet:
masst(Φ)=0 (15)
ForMass on identification framework Θ1,mass2The Dempster composition rule of function passes through formula
(17) it is described,
Wherein, 1≤p≤7,1≤q≤7, A1Represent the W in identification framework, A2Represent the N1 in identification framework, A3It represents and knows
N2, A in other frame4Represent the N3 in identification framework, A5Represent the REM in identification framework, A6It represents not true in identification framework
Determine by stages, A7Represent the Φ in identification framework;K is normaliztion constant:
Belief assignment Bel (A) for proposition A is,
The sum of the basic probability assignment for indicating all subset B in proposition A, i.e., to total degree of belief of A, when A is single element
When proposition,
Bel (A)=masst(A) (20)
Two Random Forest models are to the probability of each sleep stage as Basic Probability As-signment, masst(A1) i.e. Pt1,
masst(A2) i.e. Pt2, masst(A3) i.e. Pt3, masst(A4) i.e. Pt4, masst(A5) i.e. Pt5, masst(A6) and masst(A7) it is 0,
With above-mentioned Dempster composition rule, by one new evidence body of each combining evidences, by the substantially credible of different evidence bodies
Degree distribution merges the belief assignment Bel (A) for generating a totality, and according to maximum trust value method, calculates the trust of each proposition
Functional value selects the result with maximum trust value as the sleep stage result finally identified.
The medicine have the advantages that a kind of singly lead brain electricity sleep mode automatically method by stages, comprising the following steps: (1) signal is pre-
Processing, (2) characteristic of division extract, (3) sleep stage.Compared with the prior art, the invention has the following advantages that first is that devising
Preprocessing Algorithm obtains quality and preferably singly leads EEG signals;Second is that the present invention is extracted from time domain, frequency domain, non-linear field
A feature more than 30, and design feature filtering algorithm, filter out representative feature;Third is that the present invention uses random forest
Model, the method does not have to concern overfitting, has good noise resisting ability, and can be derived that each decision tree
By stages as a result, to obtain random forest to the fiducial probability of each sleep stage;Fourth is that present invention combination D-S evidence theory
Further increase the accuracy rate of sleep stage.In short, the present invention design a kind of singly lead brain electricity sleep mode automatically method have by stages
The advantages that algorithm is readily appreciated that, is explanatory strong, generalization ability is strong.
Detailed description of the invention
Fig. 1 is the method for the present invention flow chart of steps.
Fig. 2 is Signal Pre-Processing Method flow chart of steps.
Fig. 3 is 3 layers of WAVELET PACKET DECOMPOSITION structural schematic diagram.
Fig. 4 is to extract brain electrical feature wave flow chart.
Fig. 5 is characteristic of division extraction step flow chart.
Fig. 6 is sleep stage detailed step flow chart.
Specific embodiment
The present invention will be further explained below with reference to the attached drawings.
As shown in Figure 1, a kind of singly lead brain electricity sleep mode automatically method by stages, comprising the following steps:
Step 1, Signal Pretreatment, it is contemplated that original EEG signals contain various noise jammings, so first having to original
EEG signals are pre-processed, and specifically include following sub-step, as shown in Figure 2.
(A) baseline drift is removed, it is a kind of nonstationary random signal that baseline drift, which belongs to low-frequency noise, is become based on small echo
The multiresolution property changed removes baseline drift using the method for Wavelet decomposing and recomposing;Eeg data x is imported, x is carried out discrete
Wavelet decomposition, first layer decompose to obtain approximation coefficient C1With detail coefficients D1, wherein approximation coefficient is mainly low frequency part, details
Coefficient is mainly high frequency section, to C1Wavelet decomposition is carried out again, obtains approximation coefficient C2With detail coefficients D2, and so on;This
Invention carries out 8 layer scattering wavelet decompositions to x using ' db4 ' wavelet basis, obtains approximation coefficient C8With detail coefficients D1、D2、D3、D4、
D5、D6、D7、D8, wherein C8Frequency range beIn sample frequency fsFor C can be obtained under conditions of 100Hz8Frequency band model
It encloses for 0~0.1953Hz, this frequency range corresponds to baseline drift;Ignore C8According only to D1~D8It is reconstructed, can be obtained
EEG signals after removing baseline drift;
(B) Hz noise is removed, Hz noise intensity is high, and is ubiquitous in unshielded environments, it is necessary to logical
It crosses trapper and filters out Hz noise, good effect can be reached using 4 rank IIR trappers, trap frequency is set as 50Hz;
(C) it goes unless EEG signals ingredient, sleep cerebral electricity signal include following several brain electrical feature waves: α wave frequency range
For 8~13Hz, β wave frequency range be 13~30Hz, δ wave frequency range is 1~4Hz, θ wave frequency range is 4~8Hz, sawtooth
Wave frequency range is 2~6Hz, spindle wave frequency range is 12~14Hz, 1 wave frequency range of β is 13~17Hz, 2 wave frequency rate model of β
It encloses for 17~30Hz;For the signal component except these frequencies, it is that other bioelectrical activities generate by human body, needs to filter
It removes;Using 10 rank IIR Butterworth bandpass filters, it is contemplated that the effective component of EEG signals sets bandpass filtering frequency
For 0.2~40Hz;
(D) signature waveform is extracted, is prepared to extract time-frequency domain characteristic of division in next step, using WAVELET PACKET DECOMPOSITION
EEG signals are decomposed into α wave, β wave, δ wave, θ wave, sawtooth by the method for wavelet packets decomposition, that is, WPD
Each species rhythm ingredients such as wave, 2 wave of spindle wave, 1 wave of β and β, the careful division on time-frequency plane by EEG signals improve brain telecommunications
Number analysis ability;As shown in figure 3, this is three layers of WAVELET PACKET DECOMPOSITION structural schematic diagram.Wavelet packet point is carried out to eeg data x
Solution, first layer decompose to obtain approximation coefficient (1,0) and detail coefficients (1,1), and wherein approximation coefficient is mainly low frequency part, details
Coefficient is mainly high frequency section, carries out WAVELET PACKET DECOMPOSITION again to (1,0), obtain approximation coefficient (2,0) and detail coefficients (2,
1), same (1,1) can also be decomposed, and so on, 6 layers of WAVELET PACKET DECOMPOSITION are carried out to x, share 26=64 nodes.Often
A node can indicate that wherein u represents the WAVELET PACKET DECOMPOSITION number of plies, and v represents u layers of the v+1 node, Fig. 3 with a coordinate (u, v)
Shown in WAVELET PACKET DECOMPOSITION tree be a full binary tree, for convenience of calculation, for v since 0, such WAVELET PACKET DECOMPOSITION tree is any non-
The left subtree coordinate of leaf node is (u+1,2v).
If the sample frequency of signal is fs, the frequency range of all z layers of WAVELET PACKET DECOMPOSITION coefficients is as follows:
When sample frequency is 100Hz, corresponding 6th layer of frequency range are as follows:
[0,0.78125],[0.78125,1.5625],[1.5625,2.34375],…,[49.21875,50]
EEG signal include following several brain electrical feature waves, 8~13Hz of α wave frequency range, 13~30Hz of β wave frequency range,
1~4Hz of δ wave frequency range, 4~8Hz of θ wave frequency range, 12~14Hz of spindle wave frequency range, sawtooth wave frequency range 2~
1 wave frequency range 12~18 of 6Hz, β, 1 wave frequency range 18~30 of β, and people is in sleep procedure, different sleep stages is corresponding
Different characteristic waves play a leading role, and when people is in lucid interval W, α wave, β wave occupy leading position, rapid eye movement phase REM
With α wave, β wave characteristic, also there is the feature of sawtooth wave, the sound sleep phase sleeps the phase without α wave, β wave characteristic with shallow, therefore can basis
These features distinguish W phase and REM phase;Shallowly sleeping the phase is N1, N2, and the sound sleep phase is N3, and δ wave is only present in the sound sleep phase, and shallowly the phase of sleeping can
There are the series of features waves such as θ wave, spindle wave, K complex wave, can be slept by these spies to distinguish deep sleep phase and either shallow
Phase.Therefore, sleep stage is carried out to the EEG signals of acquisition, needs to obtain the corresponding characteristic wave of EEG signals, this project is adopted
The method for taking 6 layers of WAVELET PACKET DECOMPOSITION obtains a series of brain electrical feature waves such as α wave, β wave, θ wave, δ wave with ' db4 ' for wavelet basis.
As shown in figure 4, this is to extract brain electrical feature wave flow chart.Input EEG signals are subjected to 6 layers of WAVELET PACKET DECOMPOSITION first, obtain 64
A node finds out frequency range further according to the brain electrical feature wave to be extracted, and selects corresponding node;By taking α wave as an example, the frequency of α wave
For 8~13Hz, selecting frequency range is the node of 7.8125~12.5Hz, extracts α wave characteristic wave, obtained node be (6,8),
(6,9), (6,10), (6,11), (6,14), (6,15), since WAVELET PACKET DECOMPOSITION tree is a full binary tree, any non-leaf section
The ordinate with the node left subtree is put into 2 times of relationships, therefore some merging can be carried out to above-mentioned node, such as (6,8) and (6,9)
(5,4) are merged into, (6,10) and (6,11) are merged into (5,5), and (6,14) and (6,15) are merged into (5,7), (5,4) and (5,5)
(4,2) are merged into, finally obtaining α wave and corresponding to wavelet packet decomposition node is (4,2)+(5,7), these nodes are carried out wavelet packet point
Solution reconstruct, the α wave brain electrical feature wave after reconstruct can be obtained.Similarly, remaining brain electrical feature wave is reconstructed, is mentioned for next step feature
It takes and prepares.
Step 2, characteristic of division extract.As shown in figure 5, to after bandpass filtering treatment signal carry out temporal signatures and
Nonlinear feature extraction carries out time domain to the brain electrical feature wave of extraction and frequency domain character extracts, by all feature extractions after,
Feature Selection is carried out, high-quality, representative feature is filtered out and forms characteristic of division vector.Comprehensively consider each sleep point
Phase duration and the relationship for analyzing data information amount, the present invention are set in terms of the data of an every τ=30s sleep stage of progress
It calculates.In the case where the sample frequency of eeg data to be analyzed is 100Hz, eeg data point total number per treatment is N=
fsτ=3000, explanation is i.e. by taking 3000 data points as an example below.It includes following sub-step that the characteristic of division, which specifically calculates:
(A) temporal signatures extract, including variance var, zero-crossing rate, Hjorth Mobility, Hjorth Complexity,
75 percentiles, α wave energy, β wave energy, δ wave energy, θ wave energy, sawtooth wave energy, spindle wave energy, 1 wave energy of β and β 2
Wave energy;Variance var and 75 percentiles in temporal signatures is common, is not further described;Zero-crossing rate is that the symbol of signal becomes
Change ratio, i.e. eeg data becomes negative from positive number and becomes the total number of positive number and the ratio of eeg data point total number from negative
Value, zero-crossing rate ZCR are described by formula (1),
Wherein, xin(1≤i≤N) indicates that the signal amplitude of i-th n eeg data, sgn () are sign function, passes through public affairs
Formula (2) is described,
Hjorth Mobility indicates average frequency, is calculated by formula (3),
Wherein, m0Indicate the variance of eeg data x, m2Indicate the variance of eeg data x first derivative;
Hjorth Complexity indicates frequency variation, is calculated by formula (4),
Wherein m4Indicate the variance of eeg data x second dervative;
All kinds of feature wave energies, i.e. α wave energy, β wave energy, δ wave energy, θ wave energy, sawtooth wave energy, spindle wave energy
Amount, 1 wave energy of β and 2 wave energy Energy of β are calculated by formula (5) respectively,
(B) frequency domain character extracts, including α wave, β wave, δ wave, θ wave absolute power spectrum, α wave, β wave, δ wave, θ wave relative power
Spectrum, brain electrical feature wave power ratio, be its absolute power spectrum ratio, i.e. δ/α, δ/β, δ/θ, θ/α, θ/β, α/β, (δ+θ)/(α+
β)、α×β1/δ、θ×β2/ δ, relative power spectrum is the ratio of its absolute power spectrum with general power spectrum, and spectra calculation uses
Eeg data x is divided into L sections by Welch method, and every section of length is M, finds out each section of power spectrum respectively, then average, in turn
Calculate power spectrum;Welch method allows each Duan Douyou partial data to be overlapped, and using Hamming window, 50% overlapping carries out cyclic graph
Power estimation, if l sections of power spectrum isIt is described by formula (6),
Wherein, 1≤m≤M, 1≤l≤L,Indicate m-th of eeg data point, w in l segment datamIndicate added window
Function, f are frequency, and window function average energy U is calculated by formula (7),
Power spectrum after average is described by formula (8),
The present invention uses Hamming window, and window width 512 is as partitioned into the length of every segment signal, has 50% between every section
Power spectral density and frequency is calculated by above formula in overlapping, and the range of frequency is 0~50Hz;To frequency range be 8~
For the α wave of 13Hz, the α brain electrical feature wave extracted by WAVELET PACKET DECOMPOSITION is subjected to periodogram spectral estimation, obtains power spectrum
Density and frequency, and 8~13Hz power spectral density part is intercepted out, its area under the curve is calculated, absolute power spectrum is obtained;It will be former
Beginning EEG signals carry out periodogram spectral estimation, and calculate the general power spectrum that its area under the curve obtains, and relative power spectrum, are them
The ratio of absolute power spectrum and general power spectrum;The brain electrical feature wave power ratio is the ratio of its absolute power spectrum.
(C) Nonlinear feature extraction, for the nonlinear characteristic used for Sample Entropy, the value of Sample Entropy is bigger, and sample sequence is just
More complicated, the calculating of Sample Entropy SampleEnt specifically includes following sub-step:
(a) eeg data x is imported, x is formed into the Vector Groups X that one group of Embedded dimensions is dim in sequence1,X2,…,
Xi,…,XN-dim+1, wherein XiIt is described by formula (9), xiIndicate the signal amplitude of i-th of eeg data of eeg data x;
Xi=[xi,xi+1,…,xi+dim-1] (9)
(b)XiWith XjI-th and j-th of vector in Vector Groups are indicated, by XiWith XjBetween distance d [Xi,Xj] it is defined as two
Maximum one of element difference in a vector, is described, x by formula (10)jIndicate j-th of eeg data of eeg data x
Signal amplitude;
(c) for meeting the i and j of formula (10), d [X is countedi,Xj] be less than " similarity " metric r number, that is, mould
Plate coupling number calculates the ratio of this number Yu vector total number, is denoted as average template coupling numberBy formula (11) into
Row description,
(d)Average value Adim(r), it is calculated by formula (12),
(e) Embedded dimensions dim+1 forms dim+1 n dimensional vector n, repeats sub-step (a)~(d) in above-mentioned sub-step (C),
Obtain Adim+1(r);
(f) Sample Entropy is defined, is calculated by formula (13),
(g) when N is finite value, Sample Entropy is calculated by formula (14),
In above-mentioned formula, the Sample Entropy of eeg data x is calculated, N takes 3000, m that 2, r is taken to indicate similar tolerance, if r
Value is too big, can lose many details, and the present invention takes r=0.2 × std (X), and the std (X) is input eeg data
Standard deviation.
(D) feature extracted and artificial sleep stage result composition characteristic Vector Groups are based on filtration method by Feature Selection
In minimal redundancy maximum correlation carry out Feature Selection, score according to correlation each feature, in the feature, sample
Correlation is big between this entropy and classification marker, therefore retains this feature, and other feature is mainly carried out from time-domain and frequency-domain feature
Screening;Set the feature quantity of quasi- selection, maximize the correlation between feature and classified variable, minimize feature and feature it
Between correlation, to filter out representative feature;
Step 3, sleep stage, specific steps are as shown in fig. 6, by the brain electrical feature after feature extraction and Feature Selection
Training set and test set are reformulated, the model of sleep stage is obtained using the method for random forest, and carry out sleep stage and test
Card, obtains the fiducial probability of each sleep stage, and the fiducial probability is finally carried out D-S evidence theory fusion, is singly led
The result of brain electrosleep by stages.It includes following sub-step that the sleep stage, which specifically calculates:
(A) data set is divided, by temporal signatures, nonlinear characteristic and sleep stage result form first data set
Frequency domain character and sleep stage result are formed second data set Data2 by Data1;
(B) above-mentioned two data set is randomly selected 80% data as training set, 20% data by model training respectively
It as test set, training Random Forest model model1 and model2, and is verified by test set, obtains each sleep point
The fiducial probability of phase specifically includes following sub-step:
(a) random sampling has in the slave training set put back to using bootstrap method and extracts TN times, constitutes TN sample,
The sample not being pumped to every time forms the outer data Out-of-bag of TN bag, and OBB is used to estimate the prediction accuracy of model;
(b) TN decision tree is constructed, nf feature is equipped with, randomly selects FN feature at each node of every one treeThe information content contained by calculating each feature, the most classification capacity of selection one in FN feature
Feature carries out node split, and each tree is grown to the maximum extent, do not do it is any cut out, to construct TN decision tree;
(c) decision tree sets number optimizing, and the range of usual TN is 100~800, and sub-step is utilized in value range
(B) bag not being pumped in sub-step (a) outer data estimate decision-tree model accuracy, when accuracy highest, it is corresponding most
Excellent decision tree number TN*;
(d) random forest is constructed, fiducial probability is calculated, TN* decision tree is combined into composition Random Forest model,
Application data set Data1 obtains model1, repeats the above method and is obtained with data set Data2 training Random Forest model
Test data is input to the model by model2, obtains the probability P by stages to W, N1, N2, N3, REM fivetw, wherein t=
1,2, w=1,2 ... 5, PtwIndicate that sleep stage is w-th of number by stages and total decision tree number in t-th of Random Forest model
Purpose ratio;
(C) D-S evidence theory merges, and D-S evidence theory data fusion method refers to Dempster rule of combination, also referred to as
Evidence Combination Methods formula by sleep stage W, N1, N2, N3, REM, does not know to form list with empty set Φ by stages if Θ is identification framework
Element proposition identification framework, A represent any proposition in identification framework, and the base of two Random Forest models is defined on identification framework
This probability distribution function masst(A) (0,1) ∈, t=1,2 meet:
masst(Φ)=0 (15)
ForMass on identification framework Θ1,mass2The Dempster composition rule of function passes through formula
(17) it is described,
Wherein, 1≤p≤7,1≤q≤7, A1Represent the W in identification framework, A2Represent the N1 in identification framework, A3It represents and knows
N2, A in other frame4Represent the N3 in identification framework, A5Represent the REM in identification framework, A6It represents not true in identification framework
Determine by stages, A7Represent the Φ in identification framework;K is normaliztion constant:
Belief assignment Bel (A) for proposition A is,
The sum of the basic probability assignment for indicating all subset B in proposition A, i.e., to total degree of belief of A, when A is single element
When proposition,
Bel (A)=masst(A) (20)
Two Random Forest models are to the probability of each sleep stage as Basic Probability As-signment, masst(A1) i.e. Pt1,
masst(A2) i.e. Pt2, masst(A3) i.e. Pt3, masst(A4) i.e. Pt4, masst(A5) i.e. Pt5, masst(A6) and masst(A7) it is 0,
With above-mentioned Dempster composition rule, by one new evidence body of each combining evidences, by the substantially credible of different evidence bodies
Degree distribution merges the belief assignment Bel (A) for generating a totality, and according to maximum trust value method, calculates the trust of each proposition
Functional value selects the result with maximum trust value as the sleep stage result finally identified.
Claims (1)
1. a kind of singly lead brain electricity sleep mode automatically method by stages, it is characterised in that the following steps are included:
Step 1, Signal Pretreatment, it is contemplated that original EEG signals contain various noise jammings, so first having to original brain electricity
Signal is pre-processed, and following sub-step is specifically included:
(A) baseline drift is removed, it is a kind of nonstationary random signal, based on wavelet transformation that baseline drift, which belongs to low-frequency noise,
Multiresolution property removes baseline drift using the method for Wavelet decomposing and recomposing;
(B) Hz noise is removed, Hz noise intensity is big, and is ubiquitous in unshielded environments, it is necessary to by special
Trapper filters out Hz noise, can reach good effect using 4 rank IIR trappers, trap frequency is set as 50Hz;
(C) it going unless EEG signals ingredient, sleep cerebral electricity signal includes following several brain electrical feature waves: α wave frequency range is 8~
13Hz, β wave frequency range are 13~30Hz, δ wave frequency range is 1~4Hz, θ wave frequency range is 4~8Hz, sawtooth wave frequency rate
Range is 2~6Hz, spindle wave frequency range is 12~14Hz, 1 wave frequency range of β is 13~17Hz, 2 wave frequency range of β is 17
~30Hz;For the signal component except these frequencies, it is that other bioelectrical activities generate by human body, needs to filter out;It adopts
With 10 rank IIR Butterworth bandpass filters, it is contemplated that bandpass filtering frequency is set as 0.2 by the effective component of EEG signals
~40Hz;
(D) signature waveform is extracted, is prepared to extract time-frequency domain characteristic of division in next step, using WAVELET PACKET DECOMPOSITION wavelet
EEG signals are decomposed into α wave, β wave, δ wave, θ wave, sawtooth wave, spindle by the method for packets decomposition, that is, WPD
Wave, 1 wave of β and each species rhythm ingredient of 2 wave of β, the i.e. careful division on time-frequency plane by EEG signals, to improve point of EEG signals
Analysis ability;
Step 2, characteristic of division extract, including temporal signatures extract, frequency domain character extracts and Nonlinear feature extraction, by all spies
After sign is extracted, Feature Selection is carried out, high-quality, representative feature is filtered out and forms characteristic of division vector;Consider
Each sleep stage duration and the relationship for analyzing data information amount, setting carry out primary sleep point with the data of every τ=30s
Phase calculates, and eeg data point total number per treatment is N=fsτ, wherein fsIndicate the sampling frequency of eeg data to be analyzed
Rate, characteristic of division extraction specifically include following sub-step:
(A) temporal signatures extract, including variance var, zero-crossing rate, Hjorth Mobility, Hjorth Complexity, and 7,500
Quantile, α wave energy, β wave energy, δ wave energy, θ wave energy, sawtooth wave energy, 2 wave energy of spindle wave energy, 1 wave energy of β and β
Amount;Variance var and 75 percentiles in temporal signatures is common, is not further described;Zero-crossing rate is the sign change ratio of signal
Rate, i.e. eeg data become negative from positive number and become the total number of positive number and the ratio of eeg data point total number from negative,
Zero-crossing rate ZCR is described by formula (1),
Wherein, xin(1≤i≤N) indicates that the signal amplitude of i-th n eeg data, sgn () are sign function, passes through formula (2)
It is described,
Hjorth Mobility indicates average frequency, is calculated by formula (3),
Wherein, m0Indicate the variance of eeg data x, m2Indicate the variance of eeg data x first derivative;
Hjorth Complexity indicates frequency variation, is calculated by formula (4),
Wherein m4Indicate the variance of eeg data x second dervative;
All kinds of feature wave energies, i.e. α wave energy, β wave energy, δ wave energy, θ wave energy, sawtooth wave energy, spindle wave energy, β 1
Wave energy and 2 wave energy Energy of β are calculated by formula (5) respectively,
(B) frequency domain character extracts, including α wave, β wave, δ wave, θ wave absolute power spectrum, α wave, β wave, δ wave, θ wave relative power spectrum,
Brain electrical feature wave power ratio is the ratio of its absolute power spectrum, i.e. δ/α, δ/β, δ/θ, θ/α, θ/β, α/β, (δ+θ)/(alpha+beta), α
×β1/δ、θ×β2/ δ, relative power spectrum is the ratio of its absolute power spectrum with general power spectrum, and spectra calculation uses Welch
Eeg data x is divided into L sections by method, and every section of length is M, finds out each section of power spectrum respectively, then average, and then is calculated
Power spectrum;Welch method allows each Duan Douyou partial data to be overlapped, and using Hamming window, 50% overlapping carries out period map and estimates
Meter, if l sections of power spectrum isIt is described by formula (6),
Wherein, 1≤m≤M, 1≤l≤L,Indicate m-th of eeg data point, w in l segment datamIndicate added window function, f
For frequency, window function average energy U is calculated by formula (7),
Power spectrum after average is described by formula (8),
The power spectral density and frequency of all kinds of characteristic waves is calculated, and calculates its area under the curve, obtains absolute power spectrum, phase
To power spectrum, power ratio;
(C) Nonlinear feature extraction, the nonlinear characteristic used is Sample Entropy, and the value of Sample Entropy is bigger, and sample sequence is more multiple
Miscellaneous, the calculating of Sample Entropy SampleEnt specifically includes following sub-step:
(a) eeg data x is imported, x is formed into the Vector Groups X that one group of Embedded dimensions is dim in sequence1,X2,…,Xi,…,
XN-dim+1, wherein XiIt is described by formula (9), xiIndicate the signal amplitude of i-th of eeg data of eeg data x;
Xi=[xi,xi+1,…,xi+dim-1] (9)
(b)XiWith XjI-th and j-th of vector in Vector Groups are indicated, by XiWith XjBetween distance d [Xi,Xj] it is defined as two arrows
Maximum one of element difference in amount, is described, x by formula (10)jIndicate the letter of j-th of eeg data of eeg data x
Number amplitude;
(c) for meeting the i and j of formula (10), d [X is countedi,Xj] be less than " similarity " metric r number, that is, template
With number, the ratio of this number Yu vector total number is calculated, is denoted as average template coupling numberIt is retouched by formula (11)
It states,
(d)Average value Adim(r), it is calculated by formula (12),
(e) Embedded dimensions dim+1 forms dim+1 n dimensional vector n, repeats sub-step (a)~(d) in above-mentioned sub-step (C), obtains
Adim+1(r);
(f) Sample Entropy is defined, is calculated by formula (13),
(g) when N is finite value, Sample Entropy is calculated by formula (14),
(D) Feature Selection, by the feature extracted and artificial sleep stage result composition characteristic Vector Groups, based in filtration method
Minimal redundancy maximum correlation carries out Feature Selection, scores according to correlation each feature, in the feature, Sample Entropy
Correlation is big between classification marker, therefore retains this feature, and other feature is mainly screened from time-domain and frequency-domain feature;
The feature quantity for setting quasi- selection, maximizes the correlation between feature and classified variable, minimizes between feature and feature
Correlation, to filter out representative feature;
Brain electrical feature after feature extraction and Feature Selection is reformulated training set and test by step 3, sleep stage
Collection, obtains the model of sleep stage using the method for random forest, and carry out sleep stage verifying, obtains each sleep stage
Fiducial probability, finally by fiducial probability carry out D-S evidence theory fusion, singly led brain electrosleep by stages as a result, sleep point
Phase specifically includes following sub-step:
(A) data set is divided, by temporal signatures, nonlinear characteristic and sleep stage result form first data set Data1,
Frequency domain character and sleep stage result are formed into second data set Data2;
(B) above-mentioned two data set is randomly selected 80% data as training set, 20% data conduct by model training respectively
Test set, training Random Forest model model1 and model2, and verified by test set, obtain each sleep stage
Fiducial probability specifically includes following sub-step:
(a) random sampling has in the slave training set put back to using bootstrap method and extracts TN times, constitutes TN sample, every time
The sample not being pumped to forms the outer data Out-of-bag of TN bag, and OBB is used to estimate the prediction accuracy of model;
(b) TN decision tree is constructed, nf feature is equipped with, randomly selects FN feature at each node of every one treeThe information content contained by calculating each feature, the most classification capacity of selection one in FN feature
Feature carries out node split, and each tree is grown to the maximum extent, do not do it is any cut out, to construct TN decision tree;
(c) decision tree sets number optimizing, and the range of usual TN is 100~800, and sub-step (B) son is utilized in value range
The bag that is not pumped in step (a) outer data estimate decision-tree model accuracy, when accuracy highest, corresponding optimizing decision
Set number TN*;
(d) random forest is constructed, fiducial probability is calculated, TN* decision tree is combined into composition Random Forest model, is applied
Data set Data1 obtains model1, repeats the above method with data set Data2 training Random Forest model and obtains model2, will
Test data is input to the model, obtains the probability P by stages to W, N1, N2, N3, REM fivetw, wherein t=1,2, w=1,
2 ... 5, PtwIndicate that sleep stage is the ratio of w-th of number by stages and total decision tree number in t-th of Random Forest model;
(C) D-S evidence theory data fusion, D-S evidence theory data fusion method refer to Dempster rule of combination, also referred to as
Evidence Combination Methods formula by sleep stage W, N1, N2, N3, REM, does not know to form list with empty set Φ by stages if Θ is identification framework
Element proposition identification framework, A represent any proposition in identification framework, and the base of two Random Forest models is defined on identification framework
This probability distribution function masst(A) (0,1) ∈, t=1,2 meet:
masst(Φ)=0 (15)
ForMass on identification framework Θ1,mass2The Dempster composition rule of function by formula (17) into
Row description,
Wherein, 1≤p≤7,1≤q≤7, A1Represent the W in identification framework, A2Represent the N1 in identification framework, A3Represent identification frame
N2 in frame, A4Represent the N3 in identification framework, A5Represent the REM in identification framework, A6Represent uncertain point in identification framework
Phase, A7Represent the Φ in identification framework;K is normaliztion constant:
Belief assignment Bel (A) for proposition A is,
The sum of the basic probability assignment for indicating all subset B in proposition A, i.e., to total degree of belief of A, when A is single element proposition
When,
Bel (A)=masst(A) (20)
Two Random Forest models are to the probability of each sleep stage as Basic Probability As-signment, masst(A1) i.e. Pt1, masst
(A2) i.e. Pt2, masst(A3) i.e. Pt3, masst(A4) i.e. Pt4, masst(A5) i.e. Pt5, masst(A6) and masst(A7) it is 0, it uses
Above-mentioned Dempster composition rule, by one new evidence body of each combining evidences, by the basic confidence level of different evidence bodies point
With the belief assignment Bel (A) for merging one totality of generation, and according to maximum trust value method, the belief function of each proposition is calculated
Value selects the result with maximum trust value as the sleep stage result finally identified.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910088652.8A CN109674468A (en) | 2019-01-30 | 2019-01-30 | It is a kind of singly to lead brain electricity sleep mode automatically method by stages |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910088652.8A CN109674468A (en) | 2019-01-30 | 2019-01-30 | It is a kind of singly to lead brain electricity sleep mode automatically method by stages |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109674468A true CN109674468A (en) | 2019-04-26 |
Family
ID=66194936
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910088652.8A Pending CN109674468A (en) | 2019-01-30 | 2019-01-30 | It is a kind of singly to lead brain electricity sleep mode automatically method by stages |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109674468A (en) |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110742599A (en) * | 2019-11-01 | 2020-02-04 | 广东工业大学 | Electrocardiosignal feature extraction and classification method and system |
CN111110221A (en) * | 2019-12-26 | 2020-05-08 | 天津大学 | Time-frequency-nonlinear multidimensional body surface gastric electrical feature extraction method |
CN111202516A (en) * | 2020-01-19 | 2020-05-29 | 北京道贞健康科技发展有限责任公司 | Method and system for judging sleep depth based on precordial brain waves of sternum |
CN111329475A (en) * | 2020-03-17 | 2020-06-26 | 杭州妞诺科技有限公司 | Single-channel electroencephalogram sleep staging method and system based on integrated model |
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 |
CN112641449A (en) * | 2020-12-18 | 2021-04-13 | 浙江大学 | EEG signal-based rapid evaluation method for cranial nerve functional state detection |
CN112674779A (en) * | 2020-12-25 | 2021-04-20 | 西华大学 | Automatic sleep staging method based on ECG (electrocardiogram) signals |
CN112842279A (en) * | 2021-03-01 | 2021-05-28 | 中山大学 | Sleep quality evaluation method and device based on multi-dimensional characteristic parameters |
CN113116307A (en) * | 2021-04-26 | 2021-07-16 | 西安领跑网络传媒科技股份有限公司 | Sleep staging method, computer-readable storage medium and program product |
CN113397560A (en) * | 2021-06-21 | 2021-09-17 | 北京脑陆科技有限公司 | Training method, device, terminal and medium for sleep stage recognition model |
CN113576410A (en) * | 2021-07-20 | 2021-11-02 | 电子科技大学 | Dynamic continuous analysis method for sleep process |
CN113812965A (en) * | 2021-08-19 | 2021-12-21 | 杭州回车电子科技有限公司 | Sleep state recognition method, device, electronic device and storage medium |
CN113842111A (en) * | 2020-06-28 | 2021-12-28 | 珠海格力电器股份有限公司 | Sleep staging method and device, computing equipment and storage medium |
CN113876339A (en) * | 2021-10-11 | 2022-01-04 | 浙江工业大学 | Method for constructing sleep state electroencephalogram characteristic signal feature set |
CN116369866A (en) * | 2023-06-05 | 2023-07-04 | 安徽星辰智跃科技有限责任公司 | Sleep stability quantification and adjustment method, system and device based on wavelet transformation |
CN116671867A (en) * | 2023-06-06 | 2023-09-01 | 中国人民解放军海军特色医学中心 | Sleep quality evaluation method and system for underwater operators |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018027141A1 (en) * | 2016-08-05 | 2018-02-08 | The Regents Of The University Of Colorado, A Body Corporate | In-ear sensing systems and methods for biological signal monitoring |
CN108451505A (en) * | 2018-04-19 | 2018-08-28 | 广西欣歌拉科技有限公司 | The In-Ear sleep stage system of light weight |
CN108596043A (en) * | 2018-03-29 | 2018-09-28 | 中国药科大学 | The automatic method by stages of sleep of single lead EEG signals based on set empirical mode decomposition |
CN108742517A (en) * | 2018-03-27 | 2018-11-06 | 重庆邮电大学 | One kind is based on the mono- lead brain electrosleep of Stacking method by stages automatically |
-
2019
- 2019-01-30 CN CN201910088652.8A patent/CN109674468A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018027141A1 (en) * | 2016-08-05 | 2018-02-08 | The Regents Of The University Of Colorado, A Body Corporate | In-ear sensing systems and methods for biological signal monitoring |
CN108742517A (en) * | 2018-03-27 | 2018-11-06 | 重庆邮电大学 | One kind is based on the mono- lead brain electrosleep of Stacking method by stages automatically |
CN108596043A (en) * | 2018-03-29 | 2018-09-28 | 中国药科大学 | The automatic method by stages of sleep of single lead EEG signals based on set empirical mode decomposition |
CN108451505A (en) * | 2018-04-19 | 2018-08-28 | 广西欣歌拉科技有限公司 | The In-Ear sleep stage system of light weight |
Non-Patent Citations (5)
Title |
---|
MEHDI ABDOLLAHPOUR,ET AL: "Sleep Stage Classification Using Dempster-Shafer Theory for Classifier Fusion", 《IEEE》 * |
刘家宇: "便携式脑电系统的睡眠自动分期方法", 《中国优秀硕士学位论文全文数据库》 * |
李向新: "基于多生理参数的睡眠分期研究及其应用", 《中国优秀硕士学位论文全文数据库》 * |
李斐: "基于脑电信号特征提取的睡眠分期方法研究", 《中国优秀硕士学位论文全文数据库》 * |
魏云: "经颅磁刺激干预孤独症儿童及脑电评价", 《中国优秀硕士学位论文全文数据库》 * |
Cited By (23)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110742599A (en) * | 2019-11-01 | 2020-02-04 | 广东工业大学 | Electrocardiosignal feature extraction and classification method and system |
CN110742599B (en) * | 2019-11-01 | 2022-05-10 | 广东工业大学 | Electrocardiosignal feature extraction and classification method and system |
CN111110221A (en) * | 2019-12-26 | 2020-05-08 | 天津大学 | Time-frequency-nonlinear multidimensional body surface gastric electrical feature extraction method |
CN111202516A (en) * | 2020-01-19 | 2020-05-29 | 北京道贞健康科技发展有限责任公司 | Method and system for judging sleep depth based on precordial brain waves of sternum |
CN111329475A (en) * | 2020-03-17 | 2020-06-26 | 杭州妞诺科技有限公司 | Single-channel electroencephalogram sleep staging method and system based on integrated model |
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 |
CN111493822B (en) * | 2020-03-23 | 2023-01-31 | 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 |
CN113842111A (en) * | 2020-06-28 | 2021-12-28 | 珠海格力电器股份有限公司 | Sleep staging method and device, computing equipment and storage medium |
CN112641449A (en) * | 2020-12-18 | 2021-04-13 | 浙江大学 | EEG signal-based rapid evaluation method for cranial nerve functional state detection |
CN112674779A (en) * | 2020-12-25 | 2021-04-20 | 西华大学 | Automatic sleep staging method based on ECG (electrocardiogram) signals |
CN112842279B (en) * | 2021-03-01 | 2022-03-08 | 中山大学 | Sleep quality evaluation method and device based on multi-dimensional characteristic parameters |
CN112842279A (en) * | 2021-03-01 | 2021-05-28 | 中山大学 | Sleep quality evaluation method and device based on multi-dimensional characteristic parameters |
CN113116307A (en) * | 2021-04-26 | 2021-07-16 | 西安领跑网络传媒科技股份有限公司 | Sleep staging method, computer-readable storage medium and program product |
CN113397560A (en) * | 2021-06-21 | 2021-09-17 | 北京脑陆科技有限公司 | Training method, device, terminal and medium for sleep stage recognition model |
CN113576410B (en) * | 2021-07-20 | 2022-09-02 | 电子科技大学 | Dynamic continuous analysis method for sleep process |
CN113576410A (en) * | 2021-07-20 | 2021-11-02 | 电子科技大学 | Dynamic continuous analysis method for sleep process |
CN113812965A (en) * | 2021-08-19 | 2021-12-21 | 杭州回车电子科技有限公司 | Sleep state recognition method, device, electronic device and storage medium |
CN113812965B (en) * | 2021-08-19 | 2024-04-09 | 杭州回车电子科技有限公司 | Sleep state identification method, sleep state identification device, electronic device and storage medium |
CN113876339A (en) * | 2021-10-11 | 2022-01-04 | 浙江工业大学 | Method for constructing sleep state electroencephalogram characteristic signal feature set |
CN116369866A (en) * | 2023-06-05 | 2023-07-04 | 安徽星辰智跃科技有限责任公司 | Sleep stability quantification and adjustment method, system and device based on wavelet transformation |
CN116369866B (en) * | 2023-06-05 | 2023-09-01 | 安徽星辰智跃科技有限责任公司 | Sleep stability quantification and adjustment method, system and device based on wavelet transformation |
CN116671867A (en) * | 2023-06-06 | 2023-09-01 | 中国人民解放军海军特色医学中心 | Sleep quality evaluation method and system for underwater operators |
CN116671867B (en) * | 2023-06-06 | 2024-02-20 | 中国人民解放军海军特色医学中心 | Sleep quality evaluation method and system for underwater operators |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109674468A (en) | It is a kind of singly to lead brain electricity sleep mode automatically method by stages | |
CN108742517B (en) | Automatic sleep staging method based on Stacking single lead electroencephalogram | |
Wang et al. | Extracting and selecting distinctive EEG features for efficient epileptic seizure prediction | |
CN106709469B (en) | Automatic sleep staging method based on electroencephalogram and myoelectricity multiple characteristics | |
CN112690793B (en) | Emotion electroencephalogram migration model training method and system and emotion recognition method and equipment | |
CN101690659B (en) | Brain wave analysis method | |
CN105147248A (en) | Physiological information-based depressive disorder evaluation system and evaluation method thereof | |
CN204931634U (en) | Based on the depression evaluating system of physiologic information | |
CN112656427A (en) | Electroencephalogram emotion recognition method based on dimension model | |
CN109602417A (en) | Sleep stage method and system based on random forest | |
CN107260166A (en) | A kind of electric artefact elimination method of practical online brain | |
CN108309290A (en) | The automatic removal method of Muscle artifacts in single channel EEG signals | |
Murugappan | Electromyogram signal based human emotion classification using KNN and LDA | |
CN105342605B (en) | A kind of method of Muscle artifacts in removal EEG signals | |
CN113768519B (en) | Method for analyzing consciousness level of patient based on deep learning and resting state electroencephalogram data | |
CN111956221A (en) | Temporal lobe epilepsy classification method based on wavelet scattering factor and LSTM neural network model | |
CN115640827B (en) | Intelligent closed-loop feedback network method and system for processing electrical stimulation data | |
Pane et al. | Identifying rules for electroencephalograph (EEG) emotion recognition and classification | |
Islam et al. | Optimal IMF selection of EMD for sleep disorder diagnosis using EEG signals | |
CN117281479A (en) | Human lower limb chronic pain distinguishing method, storage medium and device based on surface electromyographic signal multi-dimensional feature fusion | |
Ang | A Novel SVM and K-NN Classifier Based Machine Learning Technique for Epileptic Seizure Detection. | |
CN108814595A (en) | EEG signals fear degree graded features research based on VR system | |
Nanda et al. | A quantitative classification of essential and Parkinson's tremor using wavelet transform and artificial neural network on sEMG and accelerometer signals | |
Tang et al. | Epileptic seizure detection based on path signature and bi-LSTM network with attention mechanism | |
CN115270847A (en) | Design decision electroencephalogram recognition method based on wavelet packet decomposition and convolutional neural network |
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 |
Application publication date: 20190426 |
|
RJ01 | Rejection of invention patent application after publication |