CN111568411B - Heart beat classification method based on continuous heart beat activity sequence feature-SVM model - Google Patents

Abstract

The application relates to a heart beat classification method based on a continuous heart beat activity sequence feature-SVM model, which comprises the following steps: s1, removing noise in electrocardiosignals by adopting continuous wavelet change; s2, segmenting the electrocardiosignals processed in the step S1 to obtain complete heartbeats, extracting features from the segmented heartbeats, and establishing the following data sets according to the types of the extracted features: set a= {470 single heart beat morphology feature }, set b= {21 consecutive RR intervals } set c= {491 consecutive heart beat global sequence feature }; s3, inputting any one set of the data sets in the step S2 into an SVM algorithm model for heart beat classification; the application has good heart beat classification accuracy.

Description

Heart beat classification method based on continuous heart beat activity sequence feature-SVM model

Technical Field

The application belongs to the technical field of arrhythmia classification methods, and relates to a heart beat classification method based on a continuous heart beat activity sequence feature-SVM model.

Background

The normal heart has four physiological functions: autonomy, excitability, conductivity and contractility. Arrhythmia refers to the frequency or arrhythmia at which the heart beats due to the pacing position of the heart impulses, errors or disturbances in conduction. Aiming at the continuous change of the heart beat waveform, the computer can accurately and effectively read the electrocardiogram and gradually give out the diagnosis result. At present, an intelligent classification mode is gradually adopted when the electrocardiographic diagnosis is carried out, for example, the existing Gaussian NB (Gao Sibei Yes) is adopted at present, linearDi scriminantAnalysis (linear discriminant analysis), logistic regressionRegression (decision tree), GBDT (gradient lifting iteration decision tree), rannomforest and AdaBoost (adaptive enhancement) are applied to the intelligent classification of the heart beat, good effect of improving the heart beat classification efficiency and accuracy is achieved in the intelligent electrocardiographic classification, but along with the application of the classifier in practice, the classifier is found to have limitation in the heart beat classification, even the classification efficiency and accuracy are low, and the requirement of daily input electrocardiographic signal rapid and accurate classification cannot be met.

Disclosure of Invention

The application aims to overcome the defects of the prior art and provide a heart beat classification method based on a continuous heart beat activity sequence feature-SVM model.

The technical scheme of the application is as follows:

a heart beat classification method based on a continuous heart beat activity sequence feature-SVM model, comprising the steps of:

s1, removing noise in electrocardiosignals by adopting continuous wavelet transformation;

s2, segmenting the electrocardiosignals processed in the step S1 to obtain complete heartbeats, extracting features from the segmented heartbeats, and establishing the following data sets according to the types of the extracted features:

set a = {470 single heart beat morphology feature },

set b= {21 consecutive RR intervals },

set c= {491 continuous cardiac global sequence feature };

and S3, inputting any one set of the data sets in the step S2 into an SVM algorithm model for heart beat classification.

Further, the 470 single heart beat morphology feature extraction method is to extract 235 sampling points near the R peak through two leads contained in each record of the MIT-BIH arrhythmia database respectively by using the R peak position determined in the annotation file.

Compared with the prior art, the application has the beneficial effects that:

according to the application, the continuous wavelet transformation is used for removing noise, then the feature extraction is carried out, the extracted features are established into three sets according to the types of the features, one of the three sets is input into the SVM model for heart beat classification, the set A, the set B and the set C are adopted as the input of electrocardio classification, and the SVM model is adopted as the classifier for classifying the electrocardiosignals in the sets, so that the accuracy of electrocardio classification can be effectively improved.

Drawings

FIG. 1 is a schematic view of an optimal separation hyperplane.

Detailed Description

The following description of the embodiments of the present application will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.

A heart beat classification method based on a continuous heart beat activity sequence feature-SVM model, comprising the steps of:

s1, removing noise in electrocardiosignals by adopting continuous wavelet transformation;

s2, segmenting the electrocardiosignals processed in the step S1 to obtain complete heartbeats, extracting features from the segmented heartbeats, and establishing the following data sets according to the types of the extracted features:

set a = {470 single heart beat morphology feature },

set b= {21 consecutive RR intervals },

set c= {491 continuous cardiac global sequence feature };

and S3, inputting any one set of the data sets in the step S2 into an SVM algorithm model for heart beat classification.

Further, the 470 single heart beat morphology feature extraction method is to extract 235 sampling points near the R peak through two leads contained in each record of the MIT-BIH arrhythmia database respectively by using the R peak position determined in the annotation file.

The SVM model adopted in the application is a new machine learning method based on the statistical learning theory VC dimension theory and the structural risk minimization theory, the algorithm shows a plurality of special advantages in solving the problems of small sample, nonlinearity and high-dimension pattern recognition, and overcomes the problems of dimension disaster, overlearning and the like to a great extent, the basic idea of the SVM is shown in figure 1, the hollow points and the solid points respectively represent different categories, H is a separation hyperplane, and the like ₁ And H ₂ Is called support vector, H ₁ And H ₂ Known as classification intervals; the optimal separation hyperplane is the one that requires the maximum separation of the classifications on the premise of correctly separating the different classifications.

Assume the form of a linear classification plane is:

g(x)＝w ^T x+b (1)；

wherein w is a classification weight vector, b is a classification threshold, and the discriminant function is normalized to satisfy |g (x) |gtoreq 1 for both types of samples,

y _i (w ^T x _i +b)-1≥0，i＝1，2，…，l (2)；

wherein y is _i Is a class mark of the sample, x _i Is the corresponding sample.

Maximizing the classification interval margin=2/(|w|) is equivalent to solving forMinimum.

Introduction of Lagrangian multiplier a _i According to the Karush-Kuhn-Tucker (KKT) condition, the above problem can be translated into maximizing the floodfunction w (a) under constraint (4), the expression of which is shown in (5).

The quadratic programming can be used for obtaining a _i Will a _i Substituting (6) to obtain w;

selecting a which is not zero _i Substituting (7) to obtain b;

α _i (y _j (w ^T x _i +b)-1)＝0 (7)；

by derivation, the decision function becomes the following formula:

substituting the test sample into equation (8), if f (x) =1, it belongs to the category, otherwise it does not.

Experiment and results:

the following experiments were all performed on the MIT-BIH arrhythmia database, with each heart beat being classified as N (normal or bundle branch block), S (supraventricular abnormal heart beat), V (ventricular abnormal heart beat), F (fused heart beat), Q (unclassified heart beat) according to ANSI/AAMI EC 57.

In this embodiment, four values of TP, FP, TN and FN are calculated to obtain a heart beat classification result, where TP is _N Represents N-type true positive heart beat, FP _N Represents N-type false positive heart beat, TN _N Representing N-type true negative heart beat, FN _N Indicating class N false negative heart beats. The classification results for the other heart beat categories are calculated in the same way. Table 1 shows the confusion matrix of the classification results. In the experiment of the embodiment, N, S, V, F and Q represent the true category of heart beat; n, s, v, f, q represent the prediction result.

Table 1: confusion matrix for classification results

TP _N ＝N _N (9)

FP _N ＝Ns+Nv+Nf+Nq (10)

TN _N ＝Ss+Sv+Sf+Sq+Vs+Vv+Vf+Vq+Fs+Fv+Ff+Fq+Qs+Qv+Qf+Qq (11)

FN _N ＝Sn+Vn+Fn+Qn (12)

The present implementation uses sensitivity, specificity, positive predictive value and accuracy to evaluate classifier performance. Sensitivity (se) refers to the proportion of samples determined to be positive examples to all positive examples. The higher the sensitivity, the more samples that are correctly predicted. Specificity (sp) refers to the proportion of samples judged as negative to all negative. The positive predictive value (+p) is also referred to in the literature as accuracy. Accuracy is the ratio of the number of correctly classified samples to the total number of samples, reflecting the consistency between the test results and the actual results. The calculation formulas of the four evaluation indexes are as follows:

Se＝TP/(TP+FN) (13)

Sp＝TN/(TN+FP) (14)

+p＝TP/(TP+FP) (15)

Acc＝(TP+TN)/(TP+TN+FP+FN) (16)。

for the analysis of the experimental results using the set A, B, C described above, the analysis procedure was as follows:

experiment I

Experiment I uses gaussian nb (Gao Sibei leaf) linear discriminant analysis (linear discriminant analysis), logistic regression (decision tree), GBDT (gradient-lifting iterative decision tree), random forest, adaBoost (adaptive enhancement) as reference classifier, SVM model as experimental classifier for heart beat classification analysis, and classification results of different model classifiers based on different data sets as shown in table 2 below.

Table 2 classification results for different models based on different data sets

Experiment II

Experiment II heart beat classification was performed based on set a and SVM model. Experimental results show that the average classification accuracy of the heart beat classification is 98.96%, but only using single heart beat morphology features has a certain locality. The disadvantage of this experimental approach is that the morphology of the single heart beat is too monolithic. Table 3 shows the classification performance of the SVM model with 470 single heart beat morphology features only.

TABLE 3 classification results and Performance based on set A and SVM models

Experiment III

Experiment III carries out heart beat classification based on set B and SVM model. Experimental results show that the classification accuracy of N, S, V and Q classes is obviously lower. The average classification accuracy reaches 88.46%. Table 4 shows the classification results of the SVM model with only continuous beat-to-beat interval activity features. The disadvantage of this experimental method is that the continuous cardiac interval activity characteristics are not comprehensive.

TABLE 4 classification results and Performance based on set B and SVM models

Experiment IV

Experiment IV carries out heart beat classification based on set C and SVM model. The experimental results show that the average accuracy of classification reaches 99.12%. The overall accuracy of experiment IV was improved by 0.16% compared to experiment II. The overall accuracy was improved by 10.66% compared to experiment III. Table 5 shows the classification results and performance of the SVM algorithm model with 491 continuous heart beat global sequence features only.

TABLE 5 classification results and Performance based on set C and SVM models

In summary, the heart beat classification method based on the continuous heart beat activity sequence feature-SVM model provided by the application adopts the set A= {470 single heart beat morphological feature }, the set B= {21 continuous RR intervals }, the set C= {491 continuous heart beat global sequence feature } as input, and uses the SVM model as a classifier to classify heart beat, so that the classification accuracy can reach 99.12%, and the classification accuracy is higher than that of the traditional classifier on the input heart beat data set.

Although the present application has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that modifications may be made to the embodiments described, or equivalents may be substituted for elements thereof, and any modifications, equivalents, improvements and changes may be made without departing from the spirit and principles of the present application.

Claims (1)

1. A heart beat classification method based on a continuous heart beat activity sequence feature-SVM model, comprising the steps of:

s1, removing noise in electrocardiosignals by adopting continuous wavelet change;

s2, segmenting the electrocardiosignals processed in the step S1 to obtain complete heartbeats, extracting features from the segmented heartbeats, and establishing the following data sets according to the types of the extracted features:

set a = {470 single heart beat morphology feature },

set b= {21 consecutive RR intervals },

set c= {491 continuous cardiac global sequence feature };

s3, inputting any one set of the data sets in the step S2 into an SVM algorithm model for heart beat classification;

the 470 single heart beat morphological feature extraction method comprises the steps of respectively extracting 235 sampling points nearby an R peak through two leads contained in each record of an MIT-BIH arrhythmia database by utilizing the R peak position determined in an annotation file;

assume the form of a linear classification plane is:

（1）；

wherein w is a classification weight vector, b is a classification threshold, and the discriminant function is normalized to satisfy the discriminant function for both types of samplesThat is to say,

（2）；

wherein y is _i Is a class mark of the sample, x _i Is the corresponding sample;

（3）；

maximizing the classification interval margin=2/(|w|) is equivalent to solving forMinimum;

introduction of Lagrangian multiplier a _i According to the Karush-Kuhn-Tucker (KKT) condition, it can be converted to maximize the floodfunction w (a) under constraint (4), the expression of the floodfunction w (a) is shown as (5):

（4）；

（5）；

the quadratic programming can be used for obtaining a _i Will a _i Substituting (6) to obtain w;

（6）；

selecting a which is not zero _i Substitution (7)B, obtaining a product;

（7）；

by derivation, the decision function becomes the following formula:

（8）；

substituting the test sample into equation (8), if f (x) =1, it belongs to the category, otherwise it does not.