CN102697491A - Identification method and system of characteristic waveform of electrocardiogram - Google Patents

Abstract

The invention discloses identification method and system of a characteristic waveform of an electrocardiogram. The identification system comprises at least two neural network modules and a fuzzy logic mode identification module, wherein a characteristic waveform identification algorithm of each neural network module is different from all characteristic waveform identification algorithms of other neural network modules, the neural network modules operate an input electrocardiogram signal according to the characteristic waveform identification algorithms and output by identifying the characteristic waveforms from operated signals; the fuzzy logic mode identification module is used for receiving the characteristic waveforms output by each neural network module and setting medical history weight values according to the characteristic waveforms, and the fuzzy logic mode identification module carries out weighted operation on the priority levels of the characteristic waveforms and then correspondingly outputs the characteristic waveforms and the weighted priority levels of the characteristic waveforms. In the invention, neural networks with different characteristic waveform identification algorithms are adopted to identify the characteristic waveforms, thereby extending the characteristic waveforms capable of being identified by a single neural network and achieving the purpose of more accurate analysis and classification of the electrocardiogram signals.

Description

Characteristics of electrocardiogram method for waveform identification and system

Technical field

The present invention relates to medical instruments field, relate in particular to the method and system of the Electrocardiographic signature waveform of a kind of aid identification.

Background technology

Mode identification technology is widely used in medical field, such as, ECG signal is carried out pattern recognition, therefrom analyze, sort out typical signature waveform, so that the doctor judges patient according to analyzing, sort out the signature waveform that.

The main method of pattern recognition has: 1. statistical model identification; 2. syntactic pattern identification; 3. Fuzzy Pattern Recognition; 4. logical reasoning method; 5. neutral net method.Wherein, statistical model identification can not reflect the architectural feature of pattern to baroque pattern extraction characteristic difficulty, is difficult to the character of description scheme, is difficult to consider identification problem from whole angle.Syntactic pattern recognition method if having interference and noise, then extracts characteristic primitive difficulty when utilization, and is prone to error.The accurately rational membership function of Fuzzy Pattern Recognition often is difficult to set up, and has limited its application.The logical reasoning method has damaged at sample, background is unintelligible, and rule is indeterminate even when ambiguity is arranged, and effect is bad.And the neutral net rule can be handled very complicacy of some environmental informations, and background knowledge is unclear, and therefore the indefinite problem of rule of inference, is more prone to use the neutral net method when ECG signal being analyzed and sort out.But; In the prior art, the neutral net method still has some problems, and its major defect is that model is few; Ability recognized patterns (being signature waveform) is insufficient; Therefore, the neutral net method of prior art might occur identifying some signature waveform when ECG signal being analyzed classification, and causes the accuracy of ECG signal analysis low.

Summary of the invention

The embodiment of the invention provides a kind of Characteristics of electrocardiogram method for waveform identification and system, sorts out in order to ECG signal is realized analyzing more accurately.

According to an aspect of the present invention, a kind of Characteristics of electrocardiogram waveform recognition system is provided, has comprised:

At least two neural network modules; The signature waveform recognizer of each neural network module and the signature waveform recognizer of other arbitrary neural network module are all inequality; Said neural network module is used for the ECG signal of input is carried out computing based on its signature waveform recognizer, from the signal that calculates, identifies signature waveform and exports;

The fuzzy logic pattern recognition module is used to receive the signature waveform that each neural network module is exported, and according to the error size between each signature waveform and the said ECG signal, calculates the priority of each signature waveform respectively; And, after the priority of each signature waveform computed weighted, the priority after signature waveform and its weighting is carried out corresponding output according to the medical history weighted value that is provided with for each signature waveform.

Wherein, said neural network module is specially two; Signature waveform recognizer in two neural network modules is respectively: based on the signature waveform recognizer of incremental learning, based on the signature waveform recognizer of batch learning.

Said medical history weighted value is provided with according to the medical history situation.

Said fuzzy logic pattern recognition module is according to the medical history weighted value for each signature waveform setting, and the priority of each signature waveform is computed weighted to be specially:

Said fuzzy logic pattern recognition module is based upon the medical history weighted value that each signature waveform is provided with; Confirm fuzzy matrix: respectively corresponding each signature waveform of each row in the said fuzzy matrix, the value of all elements in every row is the medical history weighted value of this row characteristic of correspondence waveform; Use said fuzzy matrix that the priority of each signature waveform is computed weighted.

According to another aspect of the present invention, a kind of Characteristics of electrocardiogram method for waveform identification is provided also, has comprised:

Neural network module carries out computing to the ECG signal of input according to its signature waveform recognizer, from the signal that calculates, identifies signature waveform and exports; Wherein, said neural network module is at least two, and the signature waveform recognizer of each neural network module and the signature waveform recognizer of other arbitrary neural network module are all inequality;

The fuzzy logic pattern recognition module receives the signature waveform of each neural network module output, according to the error size between each signature waveform and the said ECG signal, calculates the priority of each signature waveform respectively; And, after the priority of each signature waveform computed weighted, the priority after signature waveform and its weighting is carried out corresponding output according to the medical history weighted value that is provided with for each signature waveform.

Wherein, said neural network module is specially two; Signature waveform recognizer in two neural network modules is respectively: based on the signature waveform recognizer of incremental learning, based on the signature waveform algorithm of batch learning.

Preferably, the training method of the neural network module of said signature waveform recognizer based on incremental learning is:

Through supposing that primary iteration is n '-T, final iteration is that current iteration n' realizes during the simulation learning, and wherein, n ' natural number, T are positive integer constants, and T+1 is a hypothesis iteration siding-to-siding block length, if < 0, primary iteration is 0 to n '-T so;

When forward direction iteration n', mean square error E _N'Calculate according to following formula 11:

$E_{n^{\&prime;}}=\frac{1}{2}{(d_{n^{\&prime;}}-y_{n^{\&prime;}})}^T(d_{n^{\&prime;}}-y_{n^{\&prime;}})$ (formula 11)

D in the above-mentioned formula 11 _N'Be the expectation vector, y _N'It is output vector;

Back feedback alternative manner through following formula 12 calculates approximate Lagrange's multiplier

${\widehat{\mathrm{\&lambda;}}}_n=\frac{{\&PartialD;}^{+}{E}_{n^{\&prime;}}}{\&PartialD;{\hat{y}}_n}=\frac{{\&PartialD;E}_{n^{\&prime;}}}{\&PartialD;{\hat{y}}_n}+\underset{j=1}{\overset{L}{\mathrm{\&Sigma;}}}{\widehat{\mathrm{\&lambda;}}}_{n+j}\frac{\&PartialD;{\hat{y}}_{n+j}}{\&PartialD;{\hat{y}}_n},$ N=n'-1 ..., n'-T, (formula 12)

In the above-mentioned formula 12; L is the lead-out terminal number,

be approximate output vector;

Formula 13-15 is used to set up boundary condition:

${\widehat{\mathrm{\&lambda;}}}_{n^{\&prime;}}=-(d_{n^{\&prime;}}-y_{n^{\&prime;}})$ (formula 13)

${\widehat{\mathrm{\&lambda;}}}_{n^{\&prime;}+1},...,{\widehat{\mathrm{\&lambda;}}}_{n^{\&prime;}+L}=0$ (formula 14)

$\frac{{\&PartialD;E}_{n^{\&prime;}}}{\&PartialD;{\hat{y}}_{n^{\&prime;}-T}},\frac{{\&PartialD;E}_{n^{\&prime;}}}{\&PartialD;{\hat{y}}_{n^{\&prime;}-T+1}}...,\frac{{\&PartialD;E}_{n^{\&prime;}}}{\&PartialD;{\hat{y}}_{n^{\&prime;}-1}}=0$ (formula 15)

Obtain the error of approximation gradient according to formula 16:

$\frac{\widehat{{\&PartialD;}^{+}{E}_{n^{\&prime;}}}}{\&PartialD;w\left(n^{\&prime;}\right)}=\underset{n=n^{\&prime;}-T}{\overset{n^{\&prime;}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}^{n^{\&prime;}-n}{\widehat{\mathrm{\&lambda;}}}_n\frac{\&PartialD;{\hat{y}}_n}{\&PartialD;w\left(n\right)},$ (formula 16)

In the above-mentioned formula 16 0 < α < the 1st, the exponential form weight coefficient of setting;

Weighted value w (n'+1) when the weighted value according to formula 17 during by the n' time iteration calculates the n'+1 time iteration:

$w(n^{\&prime;}+1)=w\left(n^{\&prime;}\right)-\mathrm{\&eta;}\frac{\widehat{{\&PartialD;}^{+}{E}_{n^{\&prime;}}}}{\&PartialD;w\left(n^{\&prime;}\right)}$ (formula 17)

η is the adjustment positive number of setting in the above-mentioned formula 17;

in the above-mentioned formula 12 confirms according to following formula 18,19,20:

$W_0=0.7\&times;s^{\frac{1}{r}}\&times;\mathrm{Normr}(2\mathrm{Rand}(s,r)-I(s,r))$ (formula 18)

In the above-mentioned formula 18, W ₀The original state of expression weight matrix W, s is the implicit number of plies of network, r is the number of the input sample of neural network module.(s r) is the equally distributed random number matrix of the capable r row of s to rand, and (s is complete 1 matrix of the capable r row of s r) to I, and normr () is the standardization normalizing matrix of matrix;

Calculate latent node output based on following formula 19:

$o_j=f(\underset{i}{\mathrm{\&Sigma;}}{w}_{\mathrm{Ij}}\&times;x_i-q_j)$ (formula 19)

In the above-mentioned formula 19, w _IjBe the weighted value that connects i node and j node in the neutral net of neural network module, i, j are the integer smaller or equal to the total node number of neutral net; q _jBe j the neural unit threshold value of setting; F is predetermined non-linear action function;

K the approximate output

of output node when the n time iteration is:

${\hat{y}}_{\mathrm{Nk}}=f(\underset{j}{\mathrm{\&Sigma;}}{w}_{\mathrm{Jk}}\&times;o_j-q_k)$ (formula 20)

In the formula 20, k is the positive integer smaller or equal to the output layer interstitial content; During the n time iteration, said approximate output vector

is formed in the approximate output of each output node

Said employing is carried out computing to the ECG signal of input according to its signature waveform recognizer based on the neural network module of the signature waveform recognizer of incremental learning, from the signal that calculates, identifies signature waveform and exports and be specially:

Said employing based on the neural network module of the signature waveform recognizer of incremental learning according to the weight matrix W that confirms in its signature waveform recognizer and the said training process; ECG signal to input carries out computing, from the signal that calculates, identifies signature waveform and exports; Element among the said weight matrix W is the weighted value that last iteration obtains.

Said medical history weighted value is provided with according to medical history and health condition.

Said basis is the medical history weighted value that each signature waveform is provided with, and the priority of each signature waveform is computed weighted, and specifically comprises:

Be based upon the medical history weighted value that each signature waveform is provided with, confirm fuzzy matrix: respectively corresponding each signature waveform of each row in the said fuzzy matrix, the value of all elements in every row is the medical history weighted value of this row characteristic of correspondence waveform;

Use said fuzzy matrix that the priority of each signature waveform is computed weighted.

The embodiment of the invention like this, has just been expanded the signature waveform that single neutral net can be discerned owing to adopted the neutral net of multiple different characteristic waveform recognition algorithm to carry out signature waveform identification; That is to say; The Characteristics of electrocardiogram waveform recognition system can different algorithms, method, angle are come the recognition feature waveform; Reach ECG signal is more accurately analyzed the purpose of sorting out; Thereby can Electrocardiographic signature waveform be provided as much as possible for the doctor, judge patient on this basis.

Further, according to the difference between each signature waveform and the said ECG signal, the priority of confirming each signature waveform is known the degree of the reference reliability of each signature waveform for the doctor, for Electrocardiographic analysis provides higher accuracy in the embodiment of the invention.

Further, the fuzzy logic pattern recognition module of the embodiment of the invention shows the probability that pairing disease of each signature waveform or relevant with it situation occur through for each signature waveform is provided with the medical history weighted value in medical history; Thereby after the medical history weighted value that use to be provided with computes weighted to the priority of each signature waveform; Obtain the priority after the weighting of each signature waveform; Further can reflect the probability of this signature waveform appearance now, thereby higher accuracy is provided for Electrocardiographic analysis.The corresponding output of priority after each signature waveform and the weighting thereof is shown to confession doctor reference on the display, and the doctor can preferentially analyze, consider, more help the doctor to Electrocardiographic analysis the higher signature waveform of the priority after the weighting.

Description of drawings

Fig. 1 is the Characteristics of electrocardiogram waveform recognition system structured flowchart of the embodiment of the invention;

Fig. 2 is the Characteristics of electrocardiogram method for waveform identification flow chart of the embodiment of the invention;

Fig. 3 is the Characteristics of electrocardiogram waveform recognition system structured flowchart of the embodiment of the invention.

The specific embodiment

For making the object of the invention, technical scheme and advantage clearer, below with reference to accompanying drawing and enumerate preferred embodiment, to further explain of the present invention.Yet, need to prove that many details of listing in the description only are in order to make the reader to one or more aspects of the present invention a thorough understanding arranged, even if there are not these specific details also can realize these aspects of the present invention.

Terms such as " module " that the application uses, " system " are intended to comprise the entity relevant with computer, such as but not limited to hardware, firmware, combination thereof, software or executory software.For example, module can be, but be not limited in: the thread of the process of moving on the processor, processor, object, executable program, execution, program and/or computer.For instance, the application program that moves on the computing equipment can be a module with this computing equipment.One or more modules can be positioned at an executory process and/or thread.

Inventor of the present invention analyzes the neutral net method of prior art, finds that the method for neural network training is different, and the neutral net that then trains is can recognized patterns also inequality.When the neutral net method of prior art was carried out pattern recognition, can recognized patterns limited, precision of analysis be low when being applied to ECG signal analyzed.To the neutral net method of prior art, main thought of the present invention is when ECG signal is analyzed, and adopts multiple neutral net, and each neutral net obtains based on different training methodes; The pattern that comprehensive these neutral nets are identified, just comprehensive these neutral nets are sorted out the signature waveform that, thereby have expanded the pattern recognition limitation of single neutral net, thereby improve the accuracy that ECG signal is analyzed.

Specify the concrete technical scheme of the embodiment of the invention below in conjunction with accompanying drawing.The Characteristics of electrocardiogram waveform recognition system that the embodiment of the invention provides, as shown in Figure 1, comprising: fuzzy logic pattern recognition module 101 and at least two neural network modules 102.

Wherein, the signature waveform recognizer of the signature waveform recognizer in each neural network module 102 and other arbitrary neural network module 102 is all inequality.That is to say, in the Characteristics of electrocardiogram waveform recognition system, comprise the neural network module of various features waveform recognition algorithm.The Characteristics of electrocardiogram waveform recognition system carries out the method for Characteristics of electrocardiogram waveform recognition, and idiographic flow is as shown in Figure 2, comprises the steps:

S201: trained each neural network module 102 receives the ECG signal of input in the Characteristics of electrocardiogram waveform recognition system.

Each neural network module 102 in the Characteristics of electrocardiogram waveform recognition system all receives ECG signal, and it is carried out computing.

S202: each neural network module 102 carries out computing, the signal that obtains calculating according to its signature waveform recognizer to the ECG signal of importing.

The different neural network modules 102 waveform recognition algorithm that takes on a different character, each neural network module 102 carry out computing according to the signature waveform recognizer that is adopted separately to the ECG signal of input respectively, obtain the signal that calculates respectively.

S203: each neural network module 102 identifies signature waveform and exports from the signal that calculates.

Each neural network module 102 is carrying out computing according to the signature waveform recognizer that it adopted to the ECG signal of importing; Behind the signal that obtains calculating, from the signal that calculates, identify the signal waveform that is complementary with some predetermined signature waveform again and export as the signature waveform that identifies.

Like this, just expanded the signature waveform that single neutral net can be discerned, can different algorithms, method, angle come the recognition feature waveform, for the doctor Electrocardiographic signature waveform is provided as much as possible more, judges patient on this basis.

In fact; ECG signal is cyclical signal often; Neural network module 102 carries out real-time operation according to the signature waveform recognizer that it adopts to ECG signal, and the signal that calculates that obtains also can be periodic, the signal that periodically calculates is analyzed again; Therefrom identify the signal waveform that is complementary with some predetermined signature waveform and export as the signature waveform that identifies, then the signature waveform of output also is periodically.For example, in the cycle, neural network module 102 identifies two signature waveforms and exports at an ECG signal, and so at next ECG signal in the cycle, neural network module 102 also can continue identification and export this two signature waveforms.

S204: fuzzy logic pattern recognition module 101 receives the signature waveform of each neural network module 102 output.

S205: each signature waveform and primary ECG signal that fuzzy logic pattern recognition module 101 will receive compare, and according to the error between each signature waveform and the said ECG signal, calculate the priority of each signature waveform respectively.

Though a plurality of neural network modules 102 can identify signature waveform more; But because trained neural network module receives the restriction of self algorithm; May in the signature waveform identifying, identify the signature waveform that differs greatly with original electrocardiographicdigital figure signal, and the reliability of this signature waveform reflection electrocardiogram truth is just very low.So, if the doctor directly carries out Electrocardiographic judgement according to these signature waveforms erroneous judgement may appear.Therefore, according to the difference between each signature waveform and the said ECG signal, the priority of confirming each signature waveform is known the degree of the reference reliability of each signature waveform for the doctor, for Electrocardiographic analysis provides higher accuracy.How to judge the difference between signature waveform and the ECG signal, and confirm what the priority of each signature waveform was well known to those skilled in the art, repeat no more here according to difference.

Because ECG signal is periodic, therefore, neural network module 102 also can periodicity output characteristic waveform.101 of fuzzy logic pattern recognition modules specifically are to compare to each signature waveform and the primary ECG signal that receive in the one-period of ECG signal; According to the error between each signature waveform and the said ECG signal, calculate the priority of each signature waveform respectively.

S206: fuzzy logic pattern recognition module 101 is according to the medical history weighted value for each signature waveform setting, after the priority of each signature waveform is computed weighted, the priority after signature waveform and its weighting carried out corresponding output.

The signature waveform that identifies for a plurality of neural network modules 102; Though in above-mentioned steps S205,, confirm that the priority of each signature waveform can provide higher accuracy for Electrocardiographic analysis based on the difference between each signature waveform and the said ECG signal; But in actual conditions, current condition often has phase relation with its medical history before, if with reference to the medical history before it, can further for Electrocardiographic analysis higher accuracy be provided.

Therefore, can show the probability that pairing disease of each signature waveform or relevant with it situation occur in medical history through fuzzy logic pattern recognition module 101 for each signature waveform is provided with the medical history weighted value.Particularly, said medical history weighted value can be provided with according to medical history and health condition.After the medical history weighted value that use to be provided with computes weighted to the priority of each signature waveform, obtain the priority after the weighting of each signature waveform, further can reflect the probability that this signature waveform occurs in the medical history.The corresponding output of priority after each signature waveform and the weighting thereof is shown on the display, for doctor's reference.And the priority higher signature waveform of doctor after to weighting can preferentially analyze, consider, thereby for Electrocardiographic analysis higher accuracy is provided.For ease of computational analysis, the medical history weighted value of setting can be the value after the normalization.

For example, a plurality of neural network modules 102 are discerned 4 signature waveform a, b, c, the d that obtains sorting by priority.Signature waveform a, b, c, d are respectively according to the priority that error size calculates: 0.4,0.3,0.25,0.05.Thereby constitute the priority matrix A: [0.4 0.3 0.25 0.05].

If according to patient's health condition and medical history, the probability that b, d signature waveform occur is a, c 4 times, so, to signature waveform a, b, c, d normalized medical history weighted value is set respectively and is: 0.1,0.4,0.1,0.4.According to the medical history weighted value that is provided with for each signature waveform, to the priority of each signature waveform compute weighted specifically can for:

Make up fuzzy matrix B earlier, respectively corresponding each signature waveform of each row among the fuzzy matrix B, the value of all elements in every row is the medical history weighted value of this row characteristic of correspondence waveform.For example, to above-mentioned signature waveform a, b, c, d normalized medical history weighted value is set respectively and is: 0.1,0.4,0.1,0.4, constructed fuzzy matrix B is:

$\left[\begin{array}{cccc}0.1& 0.1& 0.1& 0.1\\ 0.4& 0.4& 0.4& 0.4\\ 0.1& 0.1& 0.1& 0.1\\ 0.4& 0.4& 0.4& 0.4\end{array}\right]$

Use fuzzy matrix B that the priority of each signature waveform is computed weighted; Obtain the priority after the weighting of each signature waveform: C=B*A=[0.16 0.48 0.1 0.08], each element promptly is respectively the priority after the weighting of signature waveform a, b, c, d in the Matrix C.Can find out that from Matrix C the priority after the signature waveform b weighting is the highest, then the doctor can pay the utmost attention to, analyze this signature waveform according to the priority after its weighting.

Preferably; A plurality of neural network modules specifically can be two in the Characteristics of electrocardiogram waveform recognition system; As shown in Figure 3, one of them is the neural network module that adopts based on the signature waveform recognizer of incremental learning, and this paper is referred to as the incremental learning neural network module; Another is the neural network module that adopts based on the signature waveform recognizer of batch learning, and this paper is referred to as the batch learning neural network module.

Incremental learning neural network module and batch learning neural network module be applied to carry out signature waveform identification in the Characteristics of electrocardiogram waveform recognition system before, need carry out learning training earlier.

The training method of batch learning neural network module is:

Be to terminal point (n=n the learning period from initial value (n=0) during the learning training _e) the iterative cycles of discrete series, n _eInteger.The batch learning training finishes back upgrading weight between a learning period, and circulates with this.In following formula 1, is the output valve that the batch learning neural network module goes out according to the input iteration during a learning training;

is the promptly predetermined signature waveform of expectation state amount, and error E can calculate according to following formula 1 so:

$E=\underset{n=0}{\overset{n_e}{\mathrm{\&Sigma;}}}\frac{1}{2}{(d_n-y_n)}^T(d_n-y_n)$ (formula 1)

T in the above-mentioned formula 1 is a matrix transpose operator.Lagrange's multiplier can calculate according to the orderly partial derivative of error E to output valve, promptly calculates lagrangian multiplier according to following formula 2 _n:

${\mathrm{\&lambda;}}_n=\frac{{\&PartialD;}^{+}E}{{\&PartialD;y}_n},$ (formula 2)

${\mathrm{\&lambda;}}_n=\frac{\&PartialD;E}{{\&PartialD;y}_n}+\underset{j=1}{\overset{L}{\mathrm{\&Sigma;}}}{\mathrm{\&lambda;}}_{n+j}\frac{{\&PartialD;y}_{n+j}}{{\&PartialD;y}_n},$ N=n _e-1 ..., 0, (formula 3)

Wherein L is the output number, and its original state is confirmed according to following formula 4:

${\mathrm{\&lambda;}}_{n_e}=\frac{\&PartialD;E}{{\&PartialD;y}_{n_e}},{\mathrm{\&lambda;}}_{n_e+1},...,{\mathrm{\&lambda;}}_{n_e+L}=0$ (formula 4)

Through Lagrange's multiplier, according to the orderly partial derivative of following formula 5 errors of calculation to weight:

$\frac{{\&PartialD;}^{+}E}{\&PartialD;w\left(i\right)}=\underset{n=0}{\overset{n_e}{\mathrm{\&Sigma;}}}{\mathrm{\&lambda;}}_n\frac{\&PartialD;y_n}{\&PartialD;w_n\left(i\right)}$ (formula 5)

The output of project

when being the n time iteration is to the partial derivative of weight.The number of times that on behalf of weight vector, sign i upgraded; The initial value of i is 0.So far, can utilize the gradient descent method upgrading weight shown in the following formula 6:

$w(i+1)=w\left(i\right)-\mathrm{\&eta;}\frac{{\&PartialD;}^{+}E}{\&PartialD;w\left(i\right)}$ (formula 6)

η is the learning coefficient of setting in the formula 6.

The batch learning neural network module through output valve and expectation state amount, calculates error E, Lagrange's multiplier, and then calculates error gradient during learning training, utilizes the weight in the error gradient upgrading batch learning neural network module.After batch learning neural network module learning training finishes, will obtain weight matrix W, the element among the weight matrix W is the weighted value w (i+1) that last iteration obtains.The batch learning neural network module will carry out signature waveform identification according to these weighted values, the signature waveform of output identification after being applied to the Characteristics of electrocardiogram waveform recognition system.After how the batch learning neural network module carries out computing according to the weight matrix W that confirms in its signature waveform recognizer and the above-mentioned training process to the ECG signal of input; Therefrom identify the technology that signature waveform is well known to those skilled in the art, repeat no more here.

In the training method of incremental learning neural network module, the weight of current iteration n' upgrading only depend on system iteration 0 ... N '-1; The state of n'}, and weight all can change in iterative process each time, so Lagrange's multiplier can not be by accurate Calculation; Again because each iteration needs upgrading, thus can not the accurate Calculation error gradient, thus the weighted value in also just can't accurate Calculation incremental learning neural network module.Therefore; Consider that each time weight all can change in the iterative process; The incremental learning neural network module of the embodiment of the invention adopts approximation method to calculate approximate Lagrange's multiplier in the learning training process; The approximation upgrading weight of application error partial derivative, thus the weighted value in the incremental learning neural network module can comparatively fast be obtained, and the weighted value that obtains is also comparatively accurate.The learning training method of the incremental learning neural network module of the embodiment of the invention is following:

The learning training of the incremental learning neural network module of the embodiment of the invention is based on the simulation learning time period concept, and through supposing that primary iteration is n '-T, final iteration is that current iteration n' realizes during the simulation learning.Wherein, n ' is a natural number, and T is a positive integer constant, and T+1 is a hypothesis iteration siding-to-siding block length.If < 0, primary iteration is 0 to n '-T so.

When forward direction iteration n', mean square error E _N'Calculate according to following formula 11:

$E_{n^{\&prime;}}=\frac{1}{2}{(d_{n^{\&prime;}}-y_{n^{\&prime;}})}^T(d_{n^{\&prime;}}-y_{n^{\&prime;}})$ (formula 11)

D in the formula 11 _N'Be the expectation vector, y _N'It is output vector.

Use for real-time learning; Because weight all changes in iteration each time; So Lagrange's multiplier can not be by accurate Calculation; Therefore, can calculate approximate Lagrange's multiplier

through the back feedback alternative manner of following formula 12

${\widehat{\mathrm{\&lambda;}}}_n=\frac{{\&PartialD;}^{+}{E}_{n^{\&prime;}}}{\&PartialD;{\hat{y}}_n}=\frac{{\&PartialD;E}_{n^{\&prime;}}}{\&PartialD;{\hat{y}}_n}+\underset{j=1}{\overset{L}{\mathrm{\&Sigma;}}}{\widehat{\mathrm{\&lambda;}}}_{n+j}\frac{\&PartialD;{\hat{y}}_{n+j}}{\&PartialD;{\hat{y}}_n},$ N=n'-1 ..., n'-T, (formula 12)

In the formula 12; L is the lead-out terminal number,

be approximate output.

Formula 13-15 is used to set up boundary condition:

${\widehat{\mathrm{\&lambda;}}}_{n^{\&prime;}}=-(d_{n^{\&prime;}}-y_{n^{\&prime;}})$ (formula 13)

${\widehat{\mathrm{\&lambda;}}}_{n^{\&prime;}+1},...,{\widehat{\mathrm{\&lambda;}}}_{n^{\&prime;}+L}=0$ (formula 14)

$\frac{{\&PartialD;E}_{n^{\&prime;}}}{\&PartialD;{\hat{y}}_{n^{\&prime;}-T}},\frac{{\&PartialD;E}_{n^{\&prime;}}}{\&PartialD;{\hat{y}}_{n^{\&prime;}-T+1}}...,\frac{{\&PartialD;E}_{n^{\&prime;}}}{\&PartialD;{\hat{y}}_{n^{\&prime;}-1}}=0$ (formula 15)

Because each iteration all will be upgraded, so can not the accurate Calculation error gradient.Alternative method is the approximation upgrading weight of application error partial derivative.To produce the approximation of a true partial derivative with error of calculation partial derivative during the simulation learning of length T+1.So, weight can be used following error of approximation gradient formula 16 upgradings, promptly obtains the error of approximation gradient according to formula 16:

$\frac{\widehat{{\&PartialD;}^{+}{E}_{n^{\&prime;}}}}{\&PartialD;w\left(n^{\&prime;}\right)}=\underset{n=n^{\&prime;}-T}{\overset{n^{\&prime;}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}^{n^{\&prime;}-n}{\widehat{\mathrm{\&lambda;}}}_n\frac{\&PartialD;{\hat{y}}_n}{\&PartialD;w\left(n\right)},$ (formula 16)

In the formula 16 0 < α < the 1st, the exponential form weight coefficient of setting;

expression error of approximation gradient.

So far, can obtain the incremental learning pattern in the upgrading of the weight of iteration n', the weighted value w (n'+1) the when weighted value in the time of promptly can be according to the n' time iteration in the formula 17 calculates n '+1 time iteration according to following formula 17:

$w(n^{\&prime;}+1)=w\left(n^{\&prime;}\right)-\mathrm{\&eta;}\frac{\widehat{{\&PartialD;}^{+}{E}_{n^{\&prime;}}}}{\&PartialD;w\left(n^{\&prime;}\right)}$ (formula 17)

η is the adjustment positive number of setting in the formula 17.

in the above-mentioned formula 12 confirms according to following formula 18,19,20:

$W_0=0.7\&times;s^{\frac{1}{r}}\&times;\mathrm{Normr}(2\mathrm{Rand}(s,r)-I(s,r))$ (formula 18)

In the above-mentioned formula 18, W ₀The original state of expression weight matrix W, s is the implicit number of plies of network, r is the number of the input sample of neural network module.The number of latent layer in the multiple structure that the implicit number of plies of network is a neural network module, network optimum structure (comprising the number of plies) obtains according to structural optimization methods such as genetic algorithms.The computational methods of the implicit number of plies of network are well known to those skilled in the art, and repeat no more here.(s r) is the equally distributed random number matrix of the capable r row of s to rand, and (s is complete 1 matrix of the capable r row of s r) to I, and norm r () is the standardization normalizing matrix of matrix;

Calculate latent node output based on following formula 19:

$o_j=f(\underset{i}{\mathrm{\&Sigma;}}{w}_{\mathrm{Ij}}\&times;x_i-q_j)$ (formula 19)

In the above-mentioned formula 19, w _IjBe the weighted value that connects i node and j node in the neutral net of neural network module, i, j are the integer smaller or equal to the total node number of neutral net.Promptly in iterative process, as in the n' time iterative process, w _IjIt is the weighted value that connects i node and j node in the neutral net among the weight matrix W.q _jBe j the neural unit threshold value of setting; F is predetermined non-linear action function;

K the approximate output

of output node when the n time iteration is:

${\hat{y}}_{\mathrm{Nk}}=f(\underset{j}{\mathrm{\&Sigma;}}{w}_{\mathrm{Jk}}\&times;o_j-q_k)$ (formula 20)

In the formula 20, k is the positive integer smaller or equal to the output layer interstitial content; During the n time iteration, each output node is formed approximate output vector

The incremental learning neural network module obtains the weight matrix W of incremental learning neural network module after through above-mentioned learning training method, and the element among the weight matrix W is w (n'+1) weighted value that last iteration obtains.The incremental learning neural network module will carry out signature waveform identification according to the weighted value among these weight matrix W after being applied to the Characteristics of electrocardiogram waveform recognition system, the signature waveform of output identification.Be the incremental learning neural network module according to the weight matrix W that confirms in its signature waveform recognizer and the above-mentioned training process, the ECG signal of input is carried out computing, from the signal that calculates, identify signature waveform and export.How the incremental learning neural network module carries out computing according to its signature waveform recognizer and weight matrix W thereof to the ECG signal of importing, and from the signal that calculates, identifies the technology that signature waveform is well known to those skilled in the art, and repeats no more here.

The embodiment of the invention has been expanded the signature waveform that single neutral net can be discerned owing to adopted the neutral net of multiple different characteristic waveform recognition algorithm to carry out signature waveform identification.That is to say; The Characteristics of electrocardiogram waveform recognition system can different algorithms, method, angle are come the recognition feature waveform; Reach ECG signal is realized analyzing more accurately the purpose of sorting out; Thereby can Electrocardiographic signature waveform be provided as much as possible for the doctor, judge patient on this basis.

Further, according to the difference between each signature waveform and the said ECG signal, the priority of confirming each signature waveform is known the degree of the reference reliability of each signature waveform for the doctor, for Electrocardiographic analysis provides higher accuracy in the embodiment of the invention.

Further, the fuzzy logic pattern recognition module of the embodiment of the invention shows the probability that pairing disease of each signature waveform or relevant with it situation occur through for each signature waveform is provided with the medical history weighted value in medical history; Thereby after the medical history weighted value that use to be provided with computes weighted to the priority of each signature waveform; Obtain the priority after the weighting of each signature waveform; Further can reflect the probability that this signature waveform occurs in the medical history, thereby higher accuracy is provided for Electrocardiographic analysis.The corresponding output of priority after each signature waveform and the weighting thereof is shown to confession doctor reference on the display, and the doctor can preferentially analyze, consider, more help the doctor to Electrocardiographic analysis the higher signature waveform of the priority after the weighting.

One of ordinary skill in the art will appreciate that all or part of step that realizes in the foregoing description method is to instruct relevant hardware to accomplish through program; This program can be stored in the computer read/write memory medium, as: ROM/RAM, magnetic disc, CD etc.

The above only is a preferred implementation of the present invention; Should be pointed out that for those skilled in the art, under the prerequisite that does not break away from the principle of the invention; Can also make some improvement and retouching, these improvement and retouching also should be regarded as protection scope of the present invention.

Claims (10)

1. Characteristics of electrocardiogram waveform recognition system comprises:

At least two neural network modules; The signature waveform recognizer of each neural network module and the signature waveform recognizer of other arbitrary neural network module are all inequality; Said neural network module is used for the ECG signal of input is carried out computing based on its signature waveform recognizer, from the signal that calculates, identifies signature waveform and exports;

The fuzzy logic pattern recognition module is used to receive the signature waveform that each neural network module is exported, and according to the difference between each signature waveform and the said ECG signal, calculates the priority of each signature waveform respectively; And, after the priority of each signature waveform computed weighted, the priority after signature waveform and its weighting is carried out corresponding output according to the medical history weighted value that is provided with for each signature waveform.

2. the system of claim 1 is characterized in that, said neural network module is specially two; Signature waveform recognizer in two neural network modules is respectively: based on the signature waveform recognizer of incremental learning, based on the signature waveform recognizer of batch learning.

3. system as claimed in claim 2 is characterized in that, said medical history weighted value is provided with according to the medical history situation.

4. system as claimed in claim 3 is characterized in that,

Said fuzzy logic pattern recognition module is according to the medical history weighted value for each signature waveform setting, and the priority of each signature waveform is computed weighted to be specially:

Said fuzzy logic pattern recognition module is based upon the medical history weighted value that each signature waveform is provided with; Confirm fuzzy matrix: respectively corresponding each signature waveform of each row in the said fuzzy matrix, the value of all elements in every row is the medical history weighted value of this row characteristic of correspondence waveform; Use said fuzzy matrix that the priority of each signature waveform is computed weighted.

5. Characteristics of electrocardiogram method for waveform identification comprises:

Neural network module carries out computing to the ECG signal of input according to its signature waveform recognizer, from the signal that calculates, identifies signature waveform and exports; Wherein, said neural network module is at least two, and the signature waveform recognizer of each neural network module and the signature waveform recognizer of other arbitrary neural network module are all inequality;

The fuzzy logic pattern recognition module receives the signature waveform of each neural network module output, according to the difference between each signature waveform and the said ECG signal, calculates the priority of each signature waveform respectively; And, after the priority of each signature waveform computed weighted, the priority after signature waveform and its weighting is carried out corresponding output according to the medical history weighted value that is provided with for each signature waveform.

6. method as claimed in claim 5 is characterized in that, said neural network module is specially two; Signature waveform recognizer in two neural network modules is respectively: based on the signature waveform recognizer of incremental learning, based on the signature waveform algorithm of batch learning.

7. method as claimed in claim 6 is characterized in that, the training method of the neural network module of said signature waveform recognizer based on incremental learning is:

Through supposing that primary iteration is n '-T, final iteration is that current iteration n' realizes during the simulation learning, and wherein, n ' natural number, T are positive integer constants, and T+1 is a hypothesis iteration siding-to-siding block length, if < 0, primary iteration is 0 to n '-T so;

When forward direction iteration n', mean square error E _N'Calculate according to following formula 11:

$E_{n^{\&prime;}}=\frac{1}{2}{(d_{n^{\&prime;}}-y_{n^{\&prime;}})}^T(d_{n^{\&prime;}}-y_{n^{\&prime;}})$ (formula 11)

D in the above-mentioned formula 11 _N'Be the expectation vector, y _N'It is output vector;

Back feedback alternative manner through following formula 12 calculates approximate Lagrange's multiplier

${\widehat{\mathrm{\&lambda;}}}_n=\frac{{\&PartialD;}^{+}{E}_{n^{\&prime;}}}{\&PartialD;{\hat{y}}_n}=\frac{{\&PartialD;E}_{n^{\&prime;}}}{\&PartialD;{\hat{y}}_n}+\underset{j=1}{\overset{L}{\mathrm{\&Sigma;}}}{\widehat{\mathrm{\&lambda;}}}_{n+j}\frac{\&PartialD;{\hat{y}}_{n+j}}{\&PartialD;{\hat{y}}_n},$ N=n'-1 ..., n'-T, (formula 12)

In the above-mentioned formula 12; L is the lead-out terminal number,

be approximate output vector;

Formula 13-15 is used to set up boundary condition:

${\widehat{\mathrm{\&lambda;}}}_{n^{\&prime;}}=-(d_{n^{\&prime;}}-y_{n^{\&prime;}})$ (formula 13)

${\widehat{\mathrm{\&lambda;}}}_{n^{\&prime;}+1},...,{\widehat{\mathrm{\&lambda;}}}_{n^{\&prime;}+L}=0$ (formula 14)

$\frac{{\&PartialD;E}_{n^{\&prime;}}}{\&PartialD;{\hat{y}}_{n^{\&prime;}-T}},\frac{{\&PartialD;E}_{n^{\&prime;}}}{\&PartialD;{\hat{y}}_{n^{\&prime;}-T+1}}...,\frac{{\&PartialD;E}_{n^{\&prime;}}}{\&PartialD;{\hat{y}}_{n^{\&prime;}-1}}=0$ (formula 15)

Obtain the error of approximation gradient according to formula 16:

$\frac{\widehat{{\&PartialD;}^{+}{E}_{n^{\&prime;}}}}{\&PartialD;w\left(n^{\&prime;}\right)}=\underset{n=n^{\&prime;}-T}{\overset{n^{\&prime;}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}^{n^{\&prime;}-n}{\widehat{\mathrm{\&lambda;}}}_n\frac{\&PartialD;{\hat{y}}_n}{\&PartialD;w\left(n\right)},$ (formula 16)

In the above-mentioned formula 16 0 < α < the 1st, the exponential form weight coefficient of setting;

Weighted value w (n'+1) when the weighted value according to formula 17 during by the n' time iteration calculates the n'+1 time iteration:

$w(n^{\&prime;}+1)=w\left(n^{\&prime;}\right)-\mathrm{\&eta;}\frac{\widehat{{\&PartialD;}^{+}{E}_{n^{\&prime;}}}}{\&PartialD;w\left(n^{\&prime;}\right)}$ (formula 17)

η is the adjustment positive number of setting in the above-mentioned formula 17;

in the above-mentioned formula 12 confirms according to following formula 18,19,20:

$W_0=0.7\&times;s^{\frac{1}{r}}\&times;\mathrm{Normr}(2\mathrm{Rand}(s,r)-I(s,r))$ (formula 18)

In the above-mentioned formula 18, W ₀The original state of expression weight matrix W, s is the implicit number of plies of network, r is the number of the input sample of neural network module.(s r) is the equally distributed random number matrix of the capable r row of s to rand, and (s is complete 1 matrix of the capable r row of s r) to I, and normr () is the standardization normalizing matrix of matrix;

Calculate latent node output based on following formula 19:

$o_j=f(\underset{i}{\mathrm{\&Sigma;}}{w}_{\mathrm{Ij}}\&times;x_i-q_j)$ (formula 19)

In the above-mentioned formula 19, w _IjBe the weighted value that connects i node and j node in the neutral net of neural network module, i, j are the integer smaller or equal to the total node number of neutral net; q _jBe j the neural unit threshold value of setting; F is predetermined non-linear action function;

K the approximate output of output node when the n time iteration is:

${\hat{y}}_{\mathrm{Nk}}=f(\underset{j}{\mathrm{\&Sigma;}}{w}_{\mathrm{Jk}}\&times;o_j-q_k)$ (formula 20)

In the formula 20, k is the positive integer smaller or equal to the output layer interstitial content; During the n time iteration, said approximate output vector

is formed in the approximate output of each output node

8. method as claimed in claim 7; It is characterized in that; Said employing is carried out computing to the ECG signal of input according to its signature waveform recognizer based on the neural network module of the signature waveform recognizer of incremental learning, from the signal that calculates, identifies signature waveform and exports and be specially:

Said employing based on the neural network module of the signature waveform recognizer of incremental learning according to the weight matrix W that confirms in its signature waveform recognizer and the said training process; ECG signal to input carries out computing, from the signal that calculates, identifies signature waveform and exports; Element among the said weight matrix W is the weighted value that last iteration obtains.

9. method as claimed in claim 7 is characterized in that, said medical history weighted value is provided with according to medical history and health condition.

10. method as claimed in claim 7 is characterized in that, said basis is the medical history weighted value that each signature waveform is provided with, and the priority of each signature waveform is computed weighted, and specifically comprises:

Be based upon the medical history weighted value that each signature waveform is provided with, confirm fuzzy matrix: respectively corresponding each signature waveform of each row in the said fuzzy matrix, the value of all elements in every row is the medical history weighted value of this row characteristic of correspondence waveform;

Use said fuzzy matrix that the priority of each signature waveform is computed weighted.

Images