CN110151165A - A kind of monocardiogram classification method based on Nonlinear Dynamical Characteristics - Google Patents
A kind of monocardiogram classification method based on Nonlinear Dynamical Characteristics Download PDFInfo
- Publication number
- CN110151165A CN110151165A CN201910364525.6A CN201910364525A CN110151165A CN 110151165 A CN110151165 A CN 110151165A CN 201910364525 A CN201910364525 A CN 201910364525A CN 110151165 A CN110151165 A CN 110151165A
- Authority
- CN
- China
- Prior art keywords
- monocardiogram
- data sequence
- entropy
- sequence
- classification
- 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.)
- Granted
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
- A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
-
- 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
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
Abstract
The monocardiogram classification method based on Nonlinear Dynamical Characteristics that the invention discloses a kind of, belongs to ECG detecting technical field;This method comprises the following steps: acquisition tri- lead monocardiogram signal of Frank;Noise is removed using median filtering;10 Nonlinear Dynamical Characteristics of each lead are extracted respectively;Each feature is normalized, Fusion Features are carried out, using training electrocardial vector chart-pattern and the difference between electrocardial vector chart-pattern in nonlinear kinetics index is tested, realizes the classification of normal monocardiogram and abnormal electrocardiogram vectogram.The above method is for the first time by nonlinear dynamic analysis approach application in monocardiogram classification, extracted Nonlinear Dynamical Characteristics can characterize the dynamic attribute of monocardiogram, the internal characteristics of monocardiogram are preferably excavated, abnormal and normal monocardiogram is distinguished, is suitble to use in conventional electrocardiographic examination.
Description
Technical field
The invention belongs to mode identification technologies, and in particular to a kind of electrocardial vector based on Nonlinear Dynamical Characteristics
Figure classification method.
Background technique
Cardiovascular disease is common human diseases, seriously threatens the security of the lives and property of the mankind.The world defends within 2012
Raw tissue points out that cardiovascular disease becomes global first cause of death, death toll when counting the global ten big cause of death
Significantly more than tumour and the death toll of other diseases.Electrocardiosignal is researched and analysed as a kind of relatively morning and applies to examine
The bioelectrical signals of disconnected cardiovascular disease, compared with other biological informations, electrocardio is with more periodicity and is easier to extract and analyze,
And it has contained physiologic information abundant, therefore by the important evidence as evaluation human heart health status.
Monocardiogram is an important detection methods of diagnosis of cardiovascular diseases.Since clinical application, sufficiently prove
This auxiliary diagnosis means have very big value.Monocardiogram indicate be certain in a flash the electrocardio of heart depolarization and multipole to
The variation of amount.The dynamic actual conditions closer to cardiac electrical activity of the electric shock of heart are explained with it, while can the satisfactory explanation heart
The mechanism of electrograph waveform variation.Electrocardiogram is only capable of indicating electric current of heart size and positive and negative variation, so claiming quantity electrocardiogram;And
Monocardiogram can not only reflect electrocardio size, moreover it is possible to explain the potential change of moment.All electrocardiographic diagnosis are suspicious or unknown
It, can be with monocardiogram inspection when true.Monocardiogram is in the unascertainable myocardial infarction of electrocardiogram, myocardial ischemia, conduction
Retardance etc. has certain strong point.
The method of Current Diagnostic monocardiogram is cut to VCG as ring cutting according to waveform start-stop point, and three independences are formed
Vector ring P, QRS, T ring analyzes rotation direction, main ring, maximum vector, initial vector, the orientation of terminal vector, time limit, electricity
Pressure etc. is based on areal shape feature.These methods all do not extract electrocardio kinetic characteristics.
Summary of the invention
It is of the existing technology the purpose of the present invention is overcoming the problems, such as, propose the electrocardio based on Nonlinear Dynamical Characteristics
Nonlinear Dynamical Characteristics are used in monocardiogram classification by vectogram classification method for the first time, extract Embedded dimensions, delay
Time, Kolmogorov entropy, correlation dimension, Lyapunov maximal index spectrum, approximate entropy, Sample Entropy, fuzzy entropy, LC complexity and
10 Nonlinear Dynamical Characteristics of C0 complexity, the classification for monocardiogram are predicted.A kind of more accurate description is provided
The monocardiogram classification method based on Nonlinear Dynamical Characteristics of the feature of electrocardiosignal.
The specific technical solution of the present invention is achieved by the steps of:
Step 1 obtains three lead monocardiogram signals;
Three-dimensional monocardiogram signal is extracted, is stored in the matrix form, one group of monocardiogram signal variable is constituted;
Step 2, data prediction;
Median filter process carried out to the monocardiogram signal that obtains in step 1, removal baseline drift, muscle noise,
Power supply disturbance;
Step 3, Nonlinear Dynamical Characteristics are extracted;
The pretreated monocardiogram signal of step 2 is calculated separately into ten nonlinear kinetics indexs: Embedded dimensions,
Delay time, Kolmogorov entropy, correlation dimension, Lyapunov maximal index spectrum, approximate entropy, Sample Entropy, fuzzy entropy, LZ are complicated
Degree and C0 complexity;
Step 4: normalized and Fusion Features;
Calculate in step 3 ten nonlinear kinetics indexs are normalized respectively, part is carried out to feature
With whole fusions;
Step 5: Classification and Identification;
Feature fused in step 4 is exercised supervision the training of Study strategies and methods, according to training mode and test pattern
Between about the difference between nonlinear kinetics index, realize the classification of monocardiogram;
Monocardiogram signal acquisition described in step 1 refers to using tri- lead system of Frank and obtains three-dimensional electrocardio
Vectogram, and stored in a manner of matrix, it indicates are as follows: xi(n), i=1,2,3;N=1,2 ..., N, wherein i indicates the
Several dimension datas, N indicate the sequence length of monocardiogram data sequence;
Data prediction described in step 2 refers to and carries out median filter process, input to the data obtained in step 1
Monocardiogram data sequence xk={ xk(n) | n=1,2 ..., N }, k=1,2,3;Window ranges M is defined, to data sequence xk
(n-M),...,xk(n),...,xk(n+M) intermediate value is taken to substitute xk(n), i.e.,
yk(n)=med [xk(n-M),...,xk(n),...,xk(n+M)],
After wherein med [] indicates that all numbers are by sequence sequence from small to large in window, median is taken;
Embedded dimensions described in step 3 and delay time are referred to and are extracted using improved C_C method;Extraction process
It is as follows:
4-1. inputs monocardiogram data sequence xk={ xk(n) | n=1,2 ..., N }, k=1,2,3, N be sequence
Length;The section of delay time t and Embedded dimensions m are set, optimal value is found in section;With t, m phase space reconstruction Xk={ Xki
(n) }, XkiIt (n) is the point in phase space;
4-2. defines the correlation integral of the monocardiogram data sequence:
Wherein, M=N- (m-1) t, dkij=| | Xki-Xkj||(∞), θ is Heaviside function:R is
Phase space suprasphere radius;
4-3. is by monocardiogram data sequence xk={ xki| i=1,2 ..., N } resolve into t mutually nonoverlapping sub- sequences
Column;According to correlation integral, two test statistics are defined:
4-4.Sk1(m, r, t)~t reflects the autocorrelation performance of monocardiogram data sequence;It selects minimum and maximum
Two radius r define residual quantity:
ΔSk1(m, t)=max { Sk1(m,rj,t)}-min{Sk1(m,rj,t)}
ΔSk1(m, t) has measured Sk1Maximum deviation of (m, r, the t)~t to all radius r;Sk1The first of (m, r, t)~t
A local minimum point's Δ Sk1Optimal delay, τ corresponding to (m, t)dFor the delay time t of the monocardiogram data sequence;
The monocardiogram data sequence that 4-5 is T for the period, as fixed m, r, N → ∞, t=aT is both Sk1(m,
N, r, t) Local modulus maxima be S againk2The zero point of (m, N, r, t), a are the integer greater than zero;Therefore find | Sk1(t)-Sk2
(t) | periodic point as optimal embedding window τω;The Embedded dimensions M of the monocardiogram data sequencek:
Kolmogorov entropy and correlation dimension described in step 3, are referred to and are extracted using G_P algorithm;Extraction process
It is as follows:
5-1 determines the monocardiogram data sequence x of input according to step 4k={ xk(n) | n=1,2 ..., N } delay
The section Embedded dimensions m and spacing value s is arranged, with t, m phase space reconstruction X in time tk={ Xki(n) }, Xki(n) in phase space
Point;
Define correlation integral
Wherein, l is scale, and θ is Heaviside function:
5-2 is known in l → 0, correlation integralThere are following relationships with l:
Wherein DkFor correlation dimension;
5-3 defines the K entropy of monocardiogram data sequence are as follows:
Wherein
5-4 Embedded dimensions by s is continuously increased at equal intervals in the case where, make the equal slopes line of above formula in dimensionless interzone
Property return, can simultaneously obtain correlation dimension D and the Kolmogorov entropy of monocardiogram data sequence stablize estimation;
?In the dimensionless interzone of relationship, enableThere is yij=axij-bi;Benefit
With least square method, a and b are askediOptimal estimation
?Wherein
The correlation dimension D of monocardiogram data sequence kth dimensionk:
Kolmogorov entropy:
The spectrum of Lyapunov maximal index described in step 3, refers to and extracts with the following method:
6-1 can be considered as x according to 4-1, the phase space orbit evolution of the monocardiogram data sequencek→XkMapping Fk
(x), the differential equation of the monocardiogram data sequence:Wherein xk∈Rn, and cut space midpoint xk(t) place is cut
Vector ekEVOLUTION EQUATION are as follows:In formula, T is the Jacobi matrix of F, and solution can indicate are as follows:
ek(t)=U (t, ek(0)), wherein U:ek(0)→ekIt (t) is linear operator mapping, the progressive behavior of this mapping U can be with referring to
Number is portrayed are as follows:
The Lyapunov Index Definition of 6-2 monocardiogram data sequence is the average of above-mentioned repetitive process:
The Lyapunov maximal index spectrum of 6-3 monocardiogram data sequence kth dimension is defined as:
Approximate entropy described in step 3 refers to and extracts with the following method:
7-1 inputs monocardiogram data sequence xk={ xk(n) | n=1,2 ..., N }, k=1,2,3;Reconstruct m tie up to
Amount:
Xk(1), Xk(2),....,Xk(N-m+1)
7-2 calculates any vector X in monocardiogram data sequencek,The distance between d [Xk,X* k]:
Wherein, ukIt (a) is vector XkElement;
7-3 given threshold r is counted in the monocardiogram data sequence of reconstruct and is met d [Xk(i),Xk(j)]≤r condition
Xk(j) vector number S;
DefinitionWherein the value range of j is [1, N-m+1], including j=i;
7-4 noteThen the monocardiogram data sequence kth is tieed up
Approximate entropy (ApEn) is defined as:
Sample Entropy described in step 3 refers to and extracts with the following method:,
8-1 is counted in the monocardiogram data sequence of reconstruct according to 7-1 and 7-2, given threshold r and is met d [Xk(i),Xk
(j)]≤r condition Xk(j) vector number S;
DefinitionWherein the value range of j is [1, N-m+1], including j ≠ i;
8-2 is worth average value to the i that monocardiogram data sequence is all, is denoted as
8-3 repeats 8-1 and 8-2, given threshold r, counts in the monocardiogram data sequence of reconstruct and meets d [Xk(i),Xk
(j)]≤r condition Xk(j) vector number S;
Note
The Sample Entropy of 8-4 monocardiogram data sequence kth dimension:
Fuzzy entropy extracting method described in step 3, refers to following method:
9-1 inputs monocardiogram data sequence xk={ xk(n) | n=1,2 ..., N }, k=1,2,3;It reconstructs m and ties up electrocardio
The phase space of vectogram data sequence:
Xk(i)=[uk(i),uk(i+1),...,uk(i+m-1)]-uk0(i), i=1,2 ..., N-m+1
Wherein,
9-2 introduces fuzzy membership function:
For i=1,2 ..., N-m+1, calculate
WhereinFor window
Mouth vector Xk(i) and Xk(j) the maximum absolute distance between;
9-3 is worth average value to the i that monocardiogram data sequence is all
9-4 definitionThe fuzzy entropy estimate of monocardiogram data sequence kth dimension
Are as follows:
LZ complexity extracting method described in step 3, refers to following method:
10-1 inputs monocardiogram data sequence xk={ xk(n) | n=1,2 ..., N }, k=1,2,3;
It is the number of mode newly occur in the monocardiogram data sequence that 10-2, which defines c,;
The 10-3 monocardiogram data sequence kth ties up Lempel-Ziv complexity:
Wherein, l is coarse number of segment, and n is the length of the monocardiogram data sequence of input;
C0 complexity extracting method described in step 3, refers to following method:
11-1 inputs monocardiogram data sequence xk={ xk(n) | n=1,2 ..., N }, k=1,2,3;
11-2 calculates the Fast Fourier Transform (FFT) D of monocardiogram data sequence kth dimensionk(z):
11-3 calculates gained Fast Fourier Transform (FFT) item Dk(z) mean-square value Gk:
And obtain a new sequence Yk(z):
11-4 calculates Yk(z) inverse fast Fourier transform of sequence:
The C0 complexity of monocardiogram data sequence kth dimension:
The processing of feature normalization described in step 4 refers to, place is normalized to characteristic with min-max method
Reason, calculation method are as follows:
Assuming that for sequence a1,a2,···,anIt is converted:Then new sequence b1,
b2,···,bn∈[0,1]。
Classification and Identification described in step 55 refers to and carries out abnormal electrocardiogram vectogram and normal monocardiogram together
The training of supervised learning classifier is tested with the data for having neither part nor lot in trained.
Compared with the prior art, the invention has the following advantages and beneficial effects:
1, compared with traditional monocardiogram classification method based on areal shape feature, the method for the present invention for the first time will be non-
Linear dynamics analysis thinking apply to monocardiogram classification, extract 10 Nonlinear Dynamical Characteristics for normally with it is different
The classification of normal monocardiogram.
2, compared with traditional monocardiogram classification method based on areal shape feature, what the method for the present invention was proposed
Nonlinear Dynamical Characteristics have biggish advantage, can more comprehensively extract the internal characteristics of monocardiogram, accuracy
Higher than 90%, classifying quality is more preferable.
Detailed description of the invention
Fig. 1 is the flow chart of the monocardiogram classification method based on Nonlinear Dynamical Characteristics of the embodiment of the present invention.
Fig. 2 a is heart disease patients monocardiogram three-dimensional data schematic diagram used in embodiment.
Fig. 2 b is heart disease patients monocardiogram three-dimensional visualization display schematic diagram used in embodiment.
Fig. 2 c is Normal group monocardiogram schematic diagram used in embodiment.
Fig. 2 d is Normal group monocardiogram three-dimensional visualization display schematic diagram used in embodiment.
Fig. 3 is that 10 Fusion Features of extraction are trained sorted classification using 5 classifiers in the embodiment of the present invention
Effect picture.
Specific embodiment
Present invention will now be described in further detail with reference to the embodiments and the accompanying drawings, but embodiments of the present invention are unlimited
In this.
Embodiment
As shown in Figure 1, be the monocardiogram classification method based on Nonlinear Dynamical Characteristics of the embodiment of the present invention
Flow chart, comprising the following steps:
Step 1 obtains monocardiogram: the present invention is to keep result of study true and reliable, verifies its preferable effect, data
Collection is the PTB Diagnostic ECG provided by (https: //www.physionet.org/cgi-bin/atm/ATM)
Database database.Its monocardiogram is obtained by tri- lead system of Frank, is a certain cardiac muscle as shown in Figure 2 a
Tri- lead electrocardiogram of Frank of infarct victims, if Fig. 2 b is the monocardiogram three-dimensional visualization display schematic diagram of the patient, such as
It is tri- lead electrocardiogram of Frank of a certain normal person shown in Fig. 2 c, the monocardiogram three-dimensional visualization if Fig. 2 d is the people is shown
Show schematic diagram.
Step 2 data prediction: median filtering removes dryness: making an uproar to remove baseline drift, muscle noise, power supply disturbance etc.
Electrocardiosignal is carried out median filter process by sound.Median filtering algorithm is input monocardiogram data sequence xk={ xk(n)|n
=1,2 ..., N }, k=1,2,3.Window ranges M is defined, to effect sample Xk(n-M),...,Xk(n)
,...,Xk(n+M) intermediate value, i.e. Y are takenk(n)=med [Xk(n-M),...,Xk(n),...,Xk(n+M)], wherein med
After [] indicates that all numbers are by sequence sequence from small to large in window, median is taken.
Step 3 Nonlinear Dynamical Characteristics are extracted: when 10 Nonlinear Dynamical Characteristics have Embedded dimensions, delay respectively
Between, Kolmogorov entropy, correlation dimension, Lyapunov maximal index spectrum, approximate entropy, Sample Entropy, fuzzy entropy, LC complexity and C0
Complexity.
1, Embedded dimensions and delay time are acquired by improved C_C method.Steps are as follows:
(1) monocardiogram data sequence x is inputtedk={ xk(n) | n=1,2 ..., N }, k=1,2,3, N be the length of sequence
Degree.The section of delay time t and Embedded dimensions m are set, optimal value is found in section.With t, m phase space reconstruction Xk={ Xki
(n) }, XkiIt (n) is the point in phase space.
(2) correlation integral of the electrocardio time series is defined:
Wherein, M=N- (m-1) t, dkij=| | Xki-Xkj||(∞), θ is Heaviside function:R is
Phase space suprasphere radius.
(3) by electrocardio time series xk={ xki| i=1,2 ..., N } resolve into t mutually nonoverlapping subsequences.According to
Correlation integral defines two test statistics:
(4)Sk1(m, r, t)~t reflects the autocorrelation performance of electrocardio time series.Select minimum and maximum two and half
Diameter r defines residual quantity:
ΔSk1(m, t)=max { Sk1(m,rj,t)}-min{Sk1(m,rj,t)}
ΔSk1(m, t) has measured Sk1Maximum deviation of (m, r, the t)~t to all radius r.Sk1The first of (m, r, t)~t
A local minimum point's Δ Sk1Optimal delay, τ corresponding to (m, t)dFor the delay time t of the electrocardio time series.
(5) the electrocardio time series for being T for the period, as fixed m, r, N → ∞, t=aT (a is the integer greater than zero)
It is both Sk1The Local modulus maxima of (m, N, r, t) is S againk2The zero point of (m, N, r, t), therefore find | Sk1(t)-Sk2(t) | week
Phase point is as optimal embedding window τw.The Embedded dimensions M of the electrocardio time seriesk:
2, G_P algorithm asks Kolmogorov entropy and correlation dimension.Steps are as follows:
(1) it can determine the monocardiogram data sequence x of input according to improved C_C methodk={ xk(n) | n=1,2 ...,
N } delay time t, the section Embedded dimensions m and spacing value s are set, with t, m phase space reconstruction Xk={ Xki(n) }, Xki(n) it is
Point in phase space.
Define correlation integral Wherein θ is Heaviside function:L is scale.
(2) known in l → 0, correlation integralThere are following relationships with l:
Wherein DkFor correlation dimension;
(3) the K entropy of the electrocardio time series is defined asWherein
(4) Embedded dimensions by s is continuously increased at equal intervals in the case where, make the equal slopes line of above formula in dimensionless interzone
Property return, can simultaneously obtain correlation dimension D and the Kolmogorov entropy of electrocardio time series stablize estimation.
?In the dimensionless interzone of relationship, enableThere is yij=axij-bi。
Using least square method, a and b are askediOptimal estimation?Wherein
The correlation dimension D of electrocardio time series kth dimensionk:Kolmogorov entropy:
3, the calculating of Lyapunov maximal index spectrum.Steps are as follows:
(1) according to improved C_C method, the phase space orbit evolution of the electrocardio time series can be considered as xk→XkMapping
Fk(x), the differential equation of the electrocardio time series:Wherein xk∈Rn, and cut space midpoint xk(t) place's tangent vector ek
EVOLUTION EQUATION are as follows:In formula, T is the Jacobi matrix of F, and solution can indicate are as follows: ek(t)=
U(t,ek(0)), wherein U:ek(0)→ekIt (t) is linear operator mapping, the progressive behavior of this mapping U can be portrayed with index
Are as follows:
(2) the Lyapunov index of electrocardio time series can be defined as the average of above-mentioned repetitive process:
(3) the Lyapunov maximal index spectrum of electrocardio time series kth dimension is defined as:
4, the calculating of approximate entropy.Algorithm is expressed as follows:
(1) monocardiogram data sequence x is inputtedk={ xk(n) | n=1,2 ..., N }, k=1,2,3;Reconstruct m tie up to
Amount:
Xk(1), Xk(2),....,Xk(N-m+1)
(2) any vector X in monocardiogram data sequence is calculatedk,The distance between d [Xk,X* k]:
Wherein, ukIt (a) is vector XkElement;
(3) given threshold r is counted in the monocardiogram data sequence of reconstruct and is met d [Xk(i),Xk(j)]≤r condition
Xk(j) vector number S;
DefinitionWherein the value range of j is [1, N-m+1], including j=i;
(4) rememberThen the monocardiogram data sequence kth is tieed up
Approximate entropy (ApEn) is defined as:
5, the calculating of Sample Entropy.Algorithm is expressed as follows:
(1) it according to the 1 of approximate entropy and 2, given threshold r, counts and meets in the electrocardio time series of reconstruct
d[Xk(i),Xk(j)]≤r condition Xk(j) vector number S.
DefinitionWherein the value range of j is [1, N-m+1], including j ≠ i.
(2) average value is worth to the i that electrocardio time series is all, be denoted as
(3) (1) and (2) is repeated, given threshold r is counted in the electrocardio time series of reconstruct and met d [Xk(i),Xk(j)]≤
The X of r conditionk(j) vector number S.
Note
4, the Sample Entropy of electrocardio time series kth dimension:
6, the calculating of fuzzy entropy.Algorithm is expressed as follows:
(1) monocardiogram data sequence x is inputtedk={ xk(n) | n=1,2 ..., N }, k=1,2,3.It reconstructs m and ties up electrocardio
The phase space of time series:
Xk(i)=[uk(i),uk(i+1),...,uk(i+m-1)]-uk0(i), i=1,2 ..., N-m+1
Wherein,
(2) fuzzy membership function is introducedFor i=1,2 ..., N-m
+ 1, it calculates
Wherein
For window to
Measure Xk(i) and Xk(j) the maximum absolute distance between.
(3) average value is worth to the i that electrocardio time series is all
(4) it definesThe fuzzy entropy estimation of electrocardio time series kth dimension are as follows:
7, the calculating of LZ complexity.Steps are as follows:
(1) monocardiogram data sequence x is inputtedk={ xk(n) | n=1,2 ..., N }, k=1,2,3.
(2) defining c is the number of mode newly occur in the electrocardio time series.
(3) the electrocardio time series kth ties up Lempel-Ziv complexity:
Wherein l is coarse number of segment (when traditional binaryzation, l=2), and n is the length of the electrocardio time series of input.
8, C0 complicated dynamic behaviour.Steps are as follows:
(1) monocardiogram data sequence x is inputtedk={ xk(n) | n=1,2 ..., N }, k=1,2,3.
(2) the Fast Fourier Transform (FFT) D of monocardiogram data sequence kth dimension is calculatedk(z):
(3) gained Fast Fourier Transform (FFT) item D is calculatedk(z) mean-square value Gk:
And obtain a new sequence Yk(z):
(4) Y is calculatedk(z) inverse fast Fourier transform of sequence:
The C0 complexity of monocardiogram data sequence kth dimension:
Step 4 normalized and Fusion Features: the numerical value difference in order to eliminate electrocardio dynamic characteristic data, to obtaining
The electrocardio dynamic characteristic data obtained are normalized, and used method is min-max method.Its calculation method is such as
Under: assuming that for three-dimensional electrocardial vector graphic sequence x1,x2,···,xnIt is converted:It is then new
Sequences y1,y2,···,yn∈[0,1]。
Step 5 Classification and Identification: support vector machines, KNN, naive Bayesian, random forest and integrated learning approach classifier
It is trained and classifies.Following table is total classifying quality.
Table 1
The above embodiment is a preferred embodiment of the present invention, but embodiments of the present invention are not by above-described embodiment
Limitation, other any changes, modifications, substitutions, combinations, simplifications made without departing from the spirit and principles of the present invention,
It should be equivalent substitute mode, be included within the scope of the present invention.
Claims (8)
1. a kind of monocardiogram classification method based on Nonlinear Dynamical Characteristics, which is characterized in that comprise the following steps:
Step 1 obtains three lead monocardiogram signals;
Three-dimensional monocardiogram signal is extracted, is stored in the matrix form, one group of monocardiogram signal variable is constituted;
Step 2, data prediction;
Median filter process is carried out to the monocardiogram signal obtained in step 1, removes baseline drift, muscle noise, power supply
Interference;
Step 3, Nonlinear Dynamical Characteristics are extracted;
The pretreated monocardiogram signal of step 2 is calculated separately into ten nonlinear kinetics indexs: Embedded dimensions, delay
Time, Kolmogorov entropy, correlation dimension, Lyapunov maximal index spectrum, approximate entropy, Sample Entropy, fuzzy entropy, LZ complexity and
C0 complexity;
Step 4: normalized and Fusion Features;
Calculate in step 3 ten nonlinear kinetics indexs are normalized respectively, feature are carried out partially and complete
Portion's fusion;
Step 5: Classification and Identification;
Feature fused in step 4 is exercised supervision the training of Study strategies and methods, according between training mode and test pattern
About the difference between nonlinear kinetics index, the classification of monocardiogram is realized;
Monocardiogram signal acquisition described in step 1 refers to using tri- lead system of Frank and obtains three-dimensional electrocardial vector
Figure, and stored in a manner of matrix, it indicates are as follows: xi(n), i=1,2,3;N=1,2 ..., N, wherein i indicates which is tieed up
Data, N indicate the sequence length of monocardiogram data sequence;
Data prediction described in step 2 refers to and carries out median filter process to the data obtained in step 1, inputs electrocardio
Vectogram data sequence xk={ xk(n) | n=1,2 ..., N }, k=1,2,3;Window ranges M is defined, to data sequence xk(n-
M),...,xk(n),...,xk(n+M) intermediate value is taken to substitute xk(n), i.e.,
yk(n)=med [xk(n-M),...,xk(n),...,xk(n+M)],
After wherein med [] indicates that all numbers are by sequence sequence from small to large in window, median is taken;
Embedded dimensions described in step 3 and delay time are referred to and are extracted using improved C_C method;Extraction process is as follows:
4-1. inputs monocardiogram data sequence xk={ xk(n) | n=1,2 ..., N }, k=1,2,3, N be the length of sequence;
The section of delay time t and Embedded dimensions m are set, optimal value is found in section;With t, m phase space reconstruction Xk={ Xki(n) },
XkiIt (n) is the point in phase space;
4-2. defines the correlation integral of the monocardiogram data sequence:
Wherein, M=N- (m-1) t, dkij=| | Xki-Xkj||(∞), θ is Heaviside function:R is mutually empty
Between suprasphere radius;
4-3. is by monocardiogram data sequence xk={ xki| i=1,2 ..., N } resolve into t mutually nonoverlapping subsequences;Root
According to correlation integral, two test statistics are defined:
4-4.Sk1(m, r, t)~t reflects the autocorrelation performance of monocardiogram data sequence;Select minimum and maximum two
Radius r defines residual quantity:
ΔSk1(m, t)=max { Sk1(m,rj,t)}-min{Sk1(m,rj,t)}
ΔSk1(m, t) has measured Sk1Maximum deviation of (m, r, the t)~t to all radius r;Sk1First office of (m, r, t)~t
Portion minimal point Δ Sk1Optimal delay, τ corresponding to (m, t)dFor the delay time t of the monocardiogram data sequence;
The monocardiogram data sequence that 4-5 is T for the period, as fixed m, r, N → ∞, t=aT is both Sk1(m,N,r,
T) Local modulus maxima is S againk2The zero point of (m, N, r, t), a are the integer greater than zero;Therefore find | Sk1(t)-Sk2(t)|
Periodic point as optimal embedding window τω;The Embedded dimensions M of the monocardiogram data sequencek:
2. a kind of monocardiogram classification method based on Nonlinear Dynamical Characteristics according to claim 1, feature
It is, Kolmogorov entropy and correlation dimension described in step 3 are referred to and extracted using G_P algorithm;Extraction process is such as
Under:
5-1 determines the monocardiogram data sequence x of input according to step 4k={ xk(n) | n=1,2 ..., N } delay time
The section Embedded dimensions m and spacing value s is arranged, with t, m phase space reconstruction X in tk={ Xki(n) }, Xki(n) in phase space
Point;
Define correlation integral
Wherein, l is scale, and θ is Heaviside function:
5-2 is known in l → 0, correlation integralThere are following relationships with l:
Wherein DkFor correlation dimension;
5-3 defines the K entropy of monocardiogram data sequence are as follows:
Wherein
5-4 Embedded dimensions by s is continuously increased at equal intervals in the case where, the equal slopes for making above formula in dimensionless interzone linearly return
Return, can simultaneously obtain correlation dimension D and the Kolmogorov entropy of monocardiogram data sequence stablizes estimation;
?In the dimensionless interzone of relationship, enableThere is yij=axij-bi;Using most
Small square law, asks a and biOptimal estimation
?Wherein
The correlation dimension D of monocardiogram data sequence kth dimensionk:
Kolmogorov entropy:
3. a kind of monocardiogram classification method based on Nonlinear Dynamical Characteristics according to claim 2, feature
It is, the spectrum of Lyapunov maximal index described in step 3 refers to and extracts with the following method:
6-1 can be considered as x according to 4-1, the phase space orbit evolution of the monocardiogram data sequencek→XkMapping Fk(x),
The differential equation of the monocardiogram data sequence:Wherein xk∈Rn, and cut space midpoint xk(t) place's tangent vector
ekEVOLUTION EQUATION are as follows:In formula, T is the Jacobi matrix of F, and solution can indicate are as follows: ek(t)
=U (t, ek(0)), wherein U:ek(0)→ekIt (t) is linear operator mapping, the progressive behavior of this mapping U can be carved with index
It is depicted as:
The Lyapunov Index Definition of 6-2 monocardiogram data sequence is the average of above-mentioned repetitive process:
The Lyapunov maximal index spectrum of 6-3 monocardiogram data sequence kth dimension is defined as:
Approximate entropy described in step 3 refers to and extracts with the following method:
7-1 inputs monocardiogram data sequence xk={ xk(n) | n=1,2 ..., N }, k=1,2,3;Reconstruct m dimensional vector:
Xk(1), Xk(2),....,Xk(N-m+1)
7-2 calculates any vector X in monocardiogram data sequencek,The distance between d [Xk,X* k]:
Wherein, ukIt (a) is vector XkElement;
7-3 given threshold r is counted in the monocardiogram data sequence of reconstruct and is met d [Xk(i),Xk(j)]≤r condition Xk
(j) vector number S;
DefinitionWherein the value range of j is [1, N-m+1], including j=i;
7-4 noteThen monocardiogram data sequence kth dimension is close
Like entropy (ApEn) is defined as:
4. a kind of monocardiogram classification method based on Nonlinear Dynamical Characteristics according to claim 3, feature
It is, Sample Entropy described in step 3 refers to and extracts with the following method:,
8-1 is counted in the monocardiogram data sequence of reconstruct according to 7-1 and 7-2, given threshold r and is met d [Xk(i),Xk(j)]
The X of≤r conditionk(j) vector number S;
DefinitionWherein the value range of j is [1, N-m+1], including j ≠ i;
8-2 is worth average value to the i that monocardiogram data sequence is all, is denoted as
8-3 repeats 8-1 and 8-2, given threshold r, counts in the monocardiogram data sequence of reconstruct and meets d [Xk(i),Xk(j)]
The X of≤r conditionk(j) vector number S;
Note
The Sample Entropy of 8-4 monocardiogram data sequence kth dimension:
5. a kind of monocardiogram classification method based on Nonlinear Dynamical Characteristics according to claim 4, feature
It is, fuzzy entropy extracting method described in step 3 refers to following method:
9-1 inputs monocardiogram data sequence xk={ xk(n) | n=1,2 ..., N }, k=1,2,3;It reconstructs m and ties up electrocardial vector
The phase space of diagram data sequence:
Xk(i)=[uk(i),uk(i+1),...,uk(i+m-1)]-uk0(i), i=1,2 ..., N-m+1
Wherein,
9-2 introduces fuzzy membership function:
For i=1,2 ..., N-m+1, calculate
WhereinFor window to
Measure Xk(i) and Xk(j) the maximum absolute distance between;
9-3 is worth average value to the i that monocardiogram data sequence is all
9-4 definitionThe fuzzy entropy estimation of monocardiogram data sequence kth dimension are as follows:
6. a kind of monocardiogram classification method based on Nonlinear Dynamical Characteristics according to claim 1, feature
It is, LZ complexity extracting method described in step 3 refers to following method:
10-1 inputs monocardiogram data sequence xk={ xk(n) | n=1,2 ..., N }, k=1,2,3;
It is the number of mode newly occur in the monocardiogram data sequence that 10-2, which defines c,;
The 10-3 monocardiogram data sequence kth ties up Lempel-Ziv complexity:
Wherein, l is coarse number of segment, and n is the length of the monocardiogram data sequence of input;
C0 complexity extracting method described in step 3, refers to following method:
11-1 inputs monocardiogram data sequence xk={ xk(n) | n=1,2 ..., N }, k=1,2,3;
11-2 calculates the Fast Fourier Transform (FFT) D of monocardiogram data sequence kth dimensionk(z):
11-3 calculates gained Fast Fourier Transform (FFT) item Dk(z) mean-square value Gk:
And obtain a new sequence Yk(z):
11-4 calculates Yk(z) inverse fast Fourier transform of sequence:
The C0 complexity of monocardiogram data sequence kth dimension:
7. a kind of monocardiogram classification method based on Nonlinear Dynamical Characteristics according to claim 1, feature
It is, the processing of feature normalization described in step 4 refers to, characteristic is normalized with min-max method,
Its calculation method is as follows:
Assuming that for sequence a1,a2,···,anIt is converted:Then new sequence b1,
b2,···,bn∈[0,1]。
8. a kind of monocardiogram classification method based on Nonlinear Dynamical Characteristics according to claim 7, feature
It is, Classification and Identification described in step 55 refers to and abnormal electrocardiogram vectogram and normal monocardiogram exercise supervision together
The training of Study strategies and methods is tested with the data for having neither part nor lot in trained.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910364525.6A CN110151165B (en) | 2019-04-30 | 2019-04-30 | Vector cardiogram classification method based on nonlinear dynamic characteristics |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910364525.6A CN110151165B (en) | 2019-04-30 | 2019-04-30 | Vector cardiogram classification method based on nonlinear dynamic characteristics |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110151165A true CN110151165A (en) | 2019-08-23 |
CN110151165B CN110151165B (en) | 2022-02-01 |
Family
ID=67633203
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910364525.6A Active CN110151165B (en) | 2019-04-30 | 2019-04-30 | Vector cardiogram classification method based on nonlinear dynamic characteristics |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110151165B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110693489A (en) * | 2019-10-28 | 2020-01-17 | 杭州电子科技大学 | Myocardial infarction detection method based on multi-classifier reinforcement learning |
CN110807438A (en) * | 2019-11-11 | 2020-02-18 | 电子科技大学 | Physiological signal nonlinear feature extraction method based on null permutation |
CN115813401A (en) * | 2022-09-28 | 2023-03-21 | 山东大学 | Electrocardiogram classification method and system based on dynamic learning |
Citations (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP0433626A2 (en) * | 1989-11-17 | 1991-06-26 | Wayne State University | Physiologic waveform analysis |
WO1992017240A1 (en) * | 1991-04-05 | 1992-10-15 | Medtronic, Inc. | Subcutaneous multi-electrode sensing system |
CN101579235A (en) * | 2009-05-13 | 2009-11-18 | 南京大学 | Remote intelligent ECG monitoring system based on EDGE network |
CN101596108A (en) * | 2009-06-19 | 2009-12-09 | 南京大学 | The nonlinear separation and extract methods of fetal electrocardiogram |
CN102274022A (en) * | 2011-05-10 | 2011-12-14 | 浙江大学 | Sleep state monitoring method based on electroencephalogram signals |
EP2505135A1 (en) * | 2011-03-29 | 2012-10-03 | Nihon Kohden Corporation | TWA measuring electrocardiograph, TWA measuring method, and TWA measurement system |
CN106073765A (en) * | 2016-05-27 | 2016-11-09 | 成都信汇聚源科技有限公司 | Die suddenly in a kind of miniature holter devices the implementation method of risk profile |
CN107092874A (en) * | 2017-04-10 | 2017-08-25 | 山东大学 | Personal identification method, apparatus and system based on electrocardio and fingerprint fusion feature |
CN107260161A (en) * | 2017-07-18 | 2017-10-20 | 上海图灵医疗科技有限公司 | A kind of electrocardio dynamics data quantitative analysis method |
CN108652615A (en) * | 2018-03-24 | 2018-10-16 | 度特斯(大连)实业有限公司 | A kind of electrocardiogram (ECG) data pathological characters quantitative analysis method and device |
CN109077715A (en) * | 2018-09-03 | 2018-12-25 | 北京工业大学 | A kind of electrocardiosignal automatic classification method based on single lead |
-
2019
- 2019-04-30 CN CN201910364525.6A patent/CN110151165B/en active Active
Patent Citations (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP0433626A2 (en) * | 1989-11-17 | 1991-06-26 | Wayne State University | Physiologic waveform analysis |
WO1992017240A1 (en) * | 1991-04-05 | 1992-10-15 | Medtronic, Inc. | Subcutaneous multi-electrode sensing system |
CN101579235A (en) * | 2009-05-13 | 2009-11-18 | 南京大学 | Remote intelligent ECG monitoring system based on EDGE network |
CN101596108A (en) * | 2009-06-19 | 2009-12-09 | 南京大学 | The nonlinear separation and extract methods of fetal electrocardiogram |
EP2505135A1 (en) * | 2011-03-29 | 2012-10-03 | Nihon Kohden Corporation | TWA measuring electrocardiograph, TWA measuring method, and TWA measurement system |
CN102274022A (en) * | 2011-05-10 | 2011-12-14 | 浙江大学 | Sleep state monitoring method based on electroencephalogram signals |
CN106073765A (en) * | 2016-05-27 | 2016-11-09 | 成都信汇聚源科技有限公司 | Die suddenly in a kind of miniature holter devices the implementation method of risk profile |
CN107092874A (en) * | 2017-04-10 | 2017-08-25 | 山东大学 | Personal identification method, apparatus and system based on electrocardio and fingerprint fusion feature |
CN107260161A (en) * | 2017-07-18 | 2017-10-20 | 上海图灵医疗科技有限公司 | A kind of electrocardio dynamics data quantitative analysis method |
CN108652615A (en) * | 2018-03-24 | 2018-10-16 | 度特斯(大连)实业有限公司 | A kind of electrocardiogram (ECG) data pathological characters quantitative analysis method and device |
CN109077715A (en) * | 2018-09-03 | 2018-12-25 | 北京工业大学 | A kind of electrocardiosignal automatic classification method based on single lead |
Non-Patent Citations (4)
Title |
---|
FANG, ZHEN; DENG, MUQING; TANG, MIN; 等.: "Classification of pathological and non-pathological Cardiodynamicsgram (CDG) using nonlinear dynamics indexes", 《2018 13TH WORLD CONGRESS ON INTELLIGENT CONTROL AND AUTOMATION (WCICA) 》 * |
ZHANG, JIANHUA; YIN, ZHONG; WANG, RUBIN: "Nonlinear Dynamic Classification of Momentary Mental Workload Using Physiological Features and NARX-Model-Based Least-Squares Support Vector Machines", 《IEEE TRANSACTIONS ON HUMAN-MACHINE SYSTEMS》 * |
张婷婷: "基于非线性动力学提取心电信号分形特征向量算法研究", 《中国优秀硕士学位论文全文数据库 医药卫生科技辑》 * |
邓木清: "基于确定学习的人体生物信号建模识别及其应用研究", 《中国博士学位论文全文数据库 信息科技辑》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110693489A (en) * | 2019-10-28 | 2020-01-17 | 杭州电子科技大学 | Myocardial infarction detection method based on multi-classifier reinforcement learning |
CN110693489B (en) * | 2019-10-28 | 2022-06-17 | 杭州电子科技大学 | Myocardial infarction detection method based on multi-classifier reinforcement learning |
CN110807438A (en) * | 2019-11-11 | 2020-02-18 | 电子科技大学 | Physiological signal nonlinear feature extraction method based on null permutation |
CN110807438B (en) * | 2019-11-11 | 2020-10-30 | 电子科技大学 | Physiological signal nonlinear feature extraction method based on null permutation |
CN115813401A (en) * | 2022-09-28 | 2023-03-21 | 山东大学 | Electrocardiogram classification method and system based on dynamic learning |
Also Published As
Publication number | Publication date |
---|---|
CN110151165B (en) | 2022-02-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Fang et al. | Human identification by quantifying similarity and dissimilarity in electrocardiogram phase space | |
Acharya et al. | Automated detection and localization of myocardial infarction using electrocardiogram: a comparative study of different leads | |
Park et al. | Hierarchical support vector machine based heartbeat classification using higher order statistics and hermite basis function | |
Ye et al. | Combining general multi-class and specific two-class classifiers for improved customized ECG heartbeat classification | |
Ganguly et al. | Automated detection and classification of arrhythmia from ECG signals using feature-induced long short-term memory network | |
Shi et al. | Inter-patient heartbeat classification based on region feature extraction and ensemble classifier | |
Shin et al. | Decision boundary-based anomaly detection model using improved AnoGAN from ECG data | |
Bulusu et al. | Transient ST-segment episode detection for ECG beat classification | |
EP2170155A2 (en) | Diagnostic and predictive system and methodology using multiple parameter electrocardiography superscores | |
CN110151165A (en) | A kind of monocardiogram classification method based on Nonlinear Dynamical Characteristics | |
Chen et al. | Region aggregation network: improving convolutional neural network for ECG characteristic detection | |
Wang et al. | A novel multi-scale dilated 3D CNN for epileptic seizure prediction | |
Desai et al. | Decision support system for arrhythmia beats using ECG signals with DCT, DWT and EMD methods: A comparative study | |
Yang et al. | Heartbeat classification using discrete wavelet transform and kernel principal component analysis | |
Xiong et al. | A real-time QRS detection algorithm based on energy segmentation for exercise electrocardiogram | |
Rangappa et al. | Classification of cardiac arrhythmia stages using hybrid features extraction with k-nearest neighbour classifier of ecg signals | |
CN112932498A (en) | T wave morphology classification system with strong generalization capability based on deep learning | |
Rajput et al. | On arrhythmia detection by deep learning and multidimensional representation | |
Zarrabi et al. | A system for accurately predicting the risk of myocardial infarction using PCG, ECG and clinical features | |
Mazandarani et al. | Wide complex tachycardia discrimination using dynamic time warping of ECG beats | |
Zeng et al. | Deep arrhythmia classification based on SENet and lightweight context transform | |
Erin et al. | Spectral Analysis of Cardiogenic Vibrations to Distinguish Between Valvular Heart Diseases. | |
Bashir et al. | Highlighting the current issues with pride suggestions for improving the performance of real time cardiac health monitoring | |
Fathail et al. | Ecg paper digitization and r peaks detection using fft | |
Pandeyz et al. | Cardiac arrhythmia detection and classification from ECG signals using XGBoost classifier |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |