KR102448860B1 - Prediction of Ventricular Fibrillation by Machine Learning Based on QRS Complex Shape Features - Google Patents

Abstract

The present invention relates to a method for predicting ventricular fibrillation through machine learning based on the QRS complex shape. A method for predicting ventricular fibrillation through machine learning based on the QRS complex form, in which the prediction accuracy of the occurrence of ventricular fibrillation is significantly improved using 4 morphological features and 11 heart rate variability features.
To this end, the present invention provides a method for predicting ventricular fibrillation using artificial neural network learning, the method comprising: selecting a database of subjects for learning through the artificial neural network; specifying an electrocardiogram signal in the database; extracting data according to a ventricular fibrillation prediction parameter from the electrocardiogram signal; and training the extracted data data with an artificial neural network (ANN), wherein the ventricular fibrillation prediction parameter is a feature of a plurality of QRS complexes.

Description

Prediction of Ventricular Fibrillation by Machine Learning Based on QRS Complex Shape Features}

The present invention relates to a method for predicting ventricular fibrillation through machine learning based on QRS complex morphological features. A method for predicting ventricular fibrillation through machine learning based on the QRS complex form, in which the prediction accuracy of the occurrence of ventricular fibrillation is significantly improved using four complex form features and 11 heart rate variability features.

Because measuring and analyzing ECG signals is an efficient method to identify electrical conduction dysfunction of the heart, such as cardiac arrhythmias, existing attempts have been made to detect and predict the occurrence of ventricular fibrillation by examining the ECG. In addition, heart rate variability was used to detect ventricular fibrillation. There is also a case of improving the accuracy of ventricular fibrillation prediction using the Respiratory Rate Variability (RRV) feature and artificial neural networks. The prediction performance using both the respiratory rate variability and heart rate variability features was improved compared to the case where only the heart rate variability features were used.

In the past, ventricular fibrillation was predicted using heart rate variability, respiratory rate variability, other characteristics, and combinations thereof in several cases, but a prediction method considering the QRS complex characteristics has not yet been introduced.

Since the QRS complex is an important feature that can represent the electrical activity of the ventricle well, the performance can be further improved if meaningful features are extracted from the QRS complex and predicting ventricular fibrillation using machine learning techniques along with heart rate variability feature points.

The present invention has been devised to solve the above problems, and it is a QRS complex type that predicts the occurrence of ventricular fibrillation from several seconds to several tens of seconds in advance using a machine learning technique that mainly features the heart rate variability of the electrocardiogram and the shape of the QRS complex. The purpose is to provide a method for predicting ventricular fibrillation through machine learning based on

The present invention has the following features to achieve the above object.

The present invention provides a method for predicting ventricular fibrillation using artificial neural network learning, the method comprising: selecting a database of subjects for learning through an artificial neural network; specifying an electrocardiogram signal in the database; extracting data according to a ventricular fibrillation prediction parameter from the electrocardiogram signal; and training the extracted data with an artificial neural network (ANN), wherein the ventricular fibrillation prediction parameter is a feature of a plurality of QRS complexes.

In the above, the ventricular fibrillation prediction parameter is characterized in that it further comprises a plurality of heart rate variability characteristics extracted from the electrocardiogram signal.

wherein the plurality of heart rate variability features include a plurality of time domain features, a plurality of frequency domain features, and a plurality of Poincare nonlinear analysis features calculated using successive R-peak to R-peak intervals (RR intervals). characterized in that

The plurality of time domain features are four, respectively, the mean of the RR intervals (represented as NN(RR) in MeanNN), the standard deviation (SDNN) of the NN(RR) intervals, and the mean square difference of consecutive NN(RR) intervals. It is composed of the interval difference ratio (pNN50) of successive NN(RR) intervals greater than 50 ms by the square root (RMSSD) and total NN(RR) intervals, and is characterized as defined by Equations (1) to (4) below.

(Here, RR(i) means the interval between the i-th R-peak and R-peak, and i and N are natural numbers)

The plurality of frequency domain features are four, each consisting of a very low frequency (VLF) band (0 to 0.04 Hz), a low frequency (LF) band (0.04 to 0.15 Hz), and a high frequency (HF) band (0.15 to 0.4 Hz). The power spectral density (PSD) of each frequency band calculated using Welch's periodogram with a Hanning window (256 points with a window size of 50% overlap), the low frequency (LF) band and the high frequency ( HF) is characterized in that it consists of a ratio of the power spectrum density of the band.

The plurality of Poencare nonlinear analysis features are three, the i-th RR interval is the point of the vertical line and the variance of the point along the axis of the same line, the standard deviation of consecutive RR intervals (SD1), the standard deviation of the point along the axis of the same line (SD2), and the ratio of SD1 to SD2 (SD1/SD2), and is characterized by being defined by Equations (5) to (6) below.

(where RR(i) is the interval between the i-th R-peak and the R-peak, and i is a natural number)

The plurality of QRS complex features are four QRS complex features, extracted from a QRS complex shape, and the QRS complex form includes Q, R and S waves from the calculated QRS complex display region and R-peak. characterized by being

The four QRS complex features are composed of the QRS complex display area, the averages ((QRSaM, RPampM)) of the R-peak amplitudes, and the standard deviations (QRSaSD, RPampSD) of the means, as shown in Equations (7) to (10) below. characterized by being defined.

(where N is a natural number)

According to the present invention, there is an effect that can significantly improve the accuracy of early prediction of ventricular fibrillation compared to the existing machine learning technique based on the heart rate variability characteristic.

1 is a flowchart illustrating a ventricular fibrillation prediction process according to an embodiment of the present invention.
2 is a block diagram schematically illustrating a ventricular fibrillation prediction process according to an embodiment of the present invention.
3 is a block diagram schematically illustrating a ventricular fibrillation prediction process according to another embodiment of the present invention.
4 is a block diagram of performing a ventricular fibrillation prediction process using only 11 heart rate variability characteristics according to a comparative example.
5 is a diagram for explaining an electrocardiogram signal.
6 is a diagram illustrating an example of an electrocardiogram (a) and a heart rate variability signal (b) before ventricular fibrillation occurs.
7 is a diagram showing the structure of an artificial neural network according to an embodiment of the present invention.
8 is a view showing ROC (receiver operating characteristic) curves for two models according to an embodiment and another embodiment of the present invention and one model according to a comparative example.
9 is a diagram illustrating the average and standard deviation of prediction accuracy for two models according to an embodiment and another embodiment of the present invention and one model according to a comparative example.

In order to explain the present invention, the operational advantages of the present invention, and the objects achieved by the practice of the present invention, preferred embodiments of the present invention will be exemplified below and will be described with reference to them.

First, the terms used in the present application are only used to describe specific embodiments, and are not intended to limit the present invention, and the singular expression may include a plural expression unless the context clearly indicates otherwise. In addition, in this application, terms such as "comprise" or "have" are intended to designate that a feature, number, step, operation, component, part, or combination thereof described in the specification exists, but one or more other It is to be understood that this does not preclude the possibility of addition or presence of features or numbers, steps, operations, components, parts, or combinations thereof.

In describing the present invention, if it is determined that a detailed description of a related known configuration or function may obscure the gist of the present invention, the detailed description thereof will be omitted.

1 is a flowchart illustrating a ventricular fibrillation prediction process according to an embodiment of the present invention, FIG. 2 is a block diagram schematically illustrating a ventricular fibrillation prediction process according to an embodiment of the present invention, and FIG. It is a block diagram schematically illustrating a ventricular fibrillation prediction process according to another embodiment.

4 is a block diagram of performing a ventricular fibrillation prediction process only with 11 heart rate variability characteristics according to a comparative example, FIG. 5 is a diagram illustrating an electrocardiogram signal, and FIG. 6 is an electrocardiogram (a) and It is a diagram showing an example of a heart rate variability signal (b), and FIG. 7 is a diagram showing the structure of an artificial neural network according to an embodiment of the present invention.

Referring to the drawings, the method for predicting ventricular fibrillation according to an embodiment of the present invention includes a step (S10) of selecting a database of subjects for learning with an artificial neural network (S10), and a step (S20) of specifying an electrocardiogram signal in the database; Extracting data according to the ventricular fibrillation prediction parameter from the electrocardiogram signal (S30) and training the extracted data with an artificial neural network (ANN) (S40).

Here, in the step (S10) of selecting a database of a subject for learning with the artificial neural network, a database that can be used for free in PhysioNet (xRRID: SCR_007345) was used according to an embodiment of the present invention.

Normal data sets from, for example, the Creighton University (CU) Ventricular Tachyarrhythmia Database (CUDB), Paroxysmal Atrial Fibrillation (PAF) Predictive Challenge Database (PAFDB), and MIT-BIH Normal Sinus Rhythm Database (NSRDB) are used.

Here, the sampling frequency is 250 Hz for CUDB and 128 Hz for the other two databases, CUDB had 35 data but excluded some because they only contained ventricular tachycardia (VT) events (excluding ventricular fibrillation events), and among the rest, the necessary As 7 data shorter than the data length (150 sec) were also excluded, a total of 27 data were used for data analysis.

The control group consisted of 28 subjects (22 subjects without fibril cases in PAFDB and 6 subjects in NSRDB).

Accordingly, when the database selection is completed (S10), the ECG signal is specified in the corresponding database (S20). The database of PhysioNet (RRID:SCR_007345) includes the raw ECG signal and the corresponding heart rate variability signal.

The interval (RR) between R-peak and R-peak was generated by reading the annotation file of PhysioNet (PhysioNet, RRID:SCR_007345), and FIGS. and examples of heart rate variability signals, respectively.

First, as shown in FIG. 5, the waveform of the electrocardiogram reflecting the electrical activation phase of the heart is basically composed of P, Q, R, S, and T waves as shown in the following figure. The P wave occurs during atrial depolarization, the QRS group reflects the ventricular depolarization period, and the T wave reflects the ventricular repolarization period.

Usually, the P wave is upward and has a slightly rounded shape, and the normal atrial depolarization vector is downward and leftward. It is usually directed upward, but appears downward when an atrial action potential occurs in the ectopic pacemaker located below the atria or atrioventricular junction.

Increased right atrium pressure, right atrium dilatation and hypertrophy (overload), etc. create a large, symmetrical, sharp P-wave (P-pulmonale). It can be found in diseases such as chronic obstructive pulmonary disease, asthma attack, acute pulmonary embolism, and acute pulmonary edema. Abnormal P-wave can also appear in sinus tachycardia.

The QRS complex composed of Q, R, and S waves reflects the depolarization state of the ventricles.

Ventricular depolarization begins in the left portion of the interventricular septum near the atrioventricular junction and proceeds from left to right across the interventricular septum. Therefore, the Q wave represents the depolarization of the interventricular septum, and the rest of the QRS complex represents the simultaneous left and right ventricular depolarization.

Because the left ventricle is larger than the right ventricle and has many muscles, most of the QRS complex exhibits depolarization of the left ventricle. The R wave is defined as the first recorded upwave, the Q wave is defined as the downwave recorded before the R wave, and the S wave is defined as the downwave recorded after the R wave.

The causes of abnormal QRS group include intraventricular conduction disorder (angle block, etc.) and abnormal atrioventricular conduction (premature excitability syndrome).

In addition, as shown in FIG. 6, in the present invention, the ECG signal is divided into a required time and a predicted time, and the required time represents the time used for shape extraction between 150 and 30 seconds before the start time of ventricular fibrillation, and the predicted time is It represents the time between 30 seconds and 0 seconds before the onset of ventricular fibrillation. Demand time data can be used to predict the occurrence of ventricular fibrillation before the predicted time.

When the electrocardiogram signal is specified (S20), a step of extracting data according to the ventricular fibrillation prediction parameter from the electrocardiogram signal (S30) is performed. There are four QRS complex features, and in another embodiment of the present invention, 11 heart rate variability features are further included in the four QRS complex features as shown in FIG. 3 .

[Table 1] shows the 11 heart rate variability characteristics and the 4 QRS complex characteristics, which are ventricular fibrillation prediction parameters according to one embodiment and another embodiment of the present invention.

First, the ventricular fibrillation prediction parameter according to an embodiment of the present invention consists of four QRS complex features.

These four QRS complex features (time domain) extract the desired features from the required time between 150 and 30 seconds before the onset of ventricular fibrillation. The QRS complex display area and R-peak amplitude are calculated using a feature known as electrocardiogram-guided respiratory (EDR) from the PhysioNet (PhysioNet, RRID: SCR_007345) Matlab toolbox.

The ECG-guided respiratory method uses the modulation effect of the R-peak amplitude, which is calculated by processing the display area under the QRS complex in the ECG signal. The display area of the QRS complex is calculated using the weighted sum of the samples between the two boundaries, which is the interval from the PQ junction, which is the point where P and Q waves meet in the ECG signal, to the J point, the starting point of the ST segment.

The R-peak amplitude is calculated by selecting the sample with the maximum value between the two boundaries, and RR represents the R-peak to R-peak interval. For this purpose, four QRS complex features consisting of the mean, standard deviation of the QRS complex display area, and the mean and standard deviation of the R-peak amplitude are used for early prediction of ventricular fibrillation.

In addition, as shown in FIG. 3 , in another embodiment of the present invention, the characteristics of 11 heart rate variability are further included in the QRS complex characteristics of the 4 places. versus R-peak interval), and consists of 4 time domain features, 4 frequency domain features, and 3 Poencare nonlinear analysis features.

The four features in the dual time domain are (1) mean RR intervals (mean NN(RR)), (2) standard deviation (SDNN) of NN(RR) intervals, and (3) mean squares of consecutive NN(RR) intervals. Equations (1), (2), (3), (1), (2), (3), ( 4) respectively.

In addition, the frequency domain has four characteristics, first of all, three types consisting of the very low frequency (VLF) band (0 to 0.04 Hz), the low frequency (LF) band (0.04 to 0.15 Hz), and the high frequency (HF) band (0.15 to 0.4 Hz). Three frequency bands are calculated as the power spectral density (PSD) of each band calculated using Welch's periodogram with a Hanning window (256 points with a window size of 50% overlap), and the ratio of the low frequency band and the high frequency band is It is calculated as the last feature in the frequency domain.

에 의해 스케일링된 연속 RR 간격의 표준 편차(SD1) 및 동일 선의 축에 따른 포인트의 표준 편차(SD2)는 아래 식 (5) 및 (6)에 의해 산출되며, 나머지 1개의 특징은 SD1과 SD2의 비율(SD1/SD2)이다. A characteristic of the Poincare nonlinear analysis is the distribution of points along a vertical line and a point along the axis of the same line.

The standard deviation of consecutive RR intervals scaled by (SD1) and the standard deviation of points along the axis of the same line (SD2) are calculated by the following equations (5) and (6), and the remaining one characteristic is that of SD1 and SD2. It is the ratio (SD1/SD2).

Meanwhile, the four QRS complex features are extracted from the QRS complex shape, which is composed of Q, R and S waves from the calculated display area and R-peak.

The average (QRSaM, RPampM) of the QRS complex display area and R-peak amplitude of the ECG, two of these features, is calculated using equations (7) and (8) below, and the remaining two features are their standard deviation ( QRSaSD, RPampSD) is calculated using the following equations (9) and (10).

As such, when four QRS complex features are extracted according to an embodiment of the present invention (S30) or 11 heart rate variability features and four QRS complex features are extracted according to another embodiment of the present invention (S30), the extracted A step S40 of learning the features with an artificial neural network (ANN) is performed.

The structure of an artificial neural network (ANN) is, as shown in Fig. 7, an input layer with nodes representing the input variables for the problem, a hidden layer with nodes that help capture the nonlinearity of the data, and nodes representing the dependent variables. It is a fully connected network structure consisting of three layers of an output layer with

The input layer and hidden layer used a rectified linear unit (RELU) activation function, and the output layer used a sigmoid activation function.

The activation function determines which neurons should be activated or deactivated, and the hidden layer consists of 6 neurons selected by trial and error processes.

In the present invention, as shown in FIGS. 2 and 3, the artificial neural network ( ANN) to implement the training model. The standardization and shuffling process may precede the input parameters so that they can be used smoothly in the network before they are learned by the artificial neural network.

Accordingly, an artificial neural network (ANN) model that learns four QRS complex shape feature data according to an embodiment of the present invention as an input parameter and 15 feature data plus 11 heart rate variability feature data as input parameters. At the same time as comparing the artificial neural network (ANN) model to be trained, the performance of prediction accuracy for each model was evaluated by using the model trained using only the 11 heart rate variability features shown in FIG. 4 as an input parameter as a comparative example.

For this performance evaluation, the artificial neural network model was evaluated 10 times by 10-fold cross-validation for each model. In 10-fold cross-validation, the data set was randomly divided into approximately 10 groups. One group was selected as the test data set and the remaining nine groups were used for training. Cross-validation was repeated 10 times to calculate the mean and standard deviation of each model and estimate the prediction accuracy.

To observe statistical differences between the three models, one-way repeated measures analysis of variance was performed with Tukey's post hoc analysis for multiple comparisons.

Table 2 below compares the mean and standard deviation of heart rate variability characteristics and QRS complex morphology characteristics between the control and ventricular fibrillation data sets.

Nine out of 15 features of SDNN, RMSSD, pNN50, LF, SD1, SD2, QRSaSD, QRsampM and QRSampSD showed statistically significant differences (two-tailed t-test, p<0.05).

[Table 3] shows the performance of the artificial neural network according to one embodiment and another embodiment of the present invention and the artificial neural network according to the comparative example.

The comparative example used 11 heart rate variability characteristics, and a prediction accuracy of 72% was achieved, and the sensitivity and specificity were 65.68% and 98.44%, respectively.

On the other hand, it can be seen that the model using the four QRS complex shape features according to an embodiment of the present invention has significantly improved prediction performance, and the accuracy, sensitivity, and specificity are 98.6%, 98.4%, and 99.04%, respectively.

In addition, it can be seen that the model combining the 4 features extracted from the QRS complex display region and the R-peak amplitude and 11 heart rate variability features according to another embodiment of the present invention has a fine prediction accuracy of 99.83%, but it can be seen that it is improved.

This suggests that the QRS complex morphological characteristics extracted from the electrocardiogram have a significant impact on predicting ventricular fibrillation before it occurs in terms of predictive performance.

8 is a view showing ROC (receiver operating characteristic) curves for two models according to an embodiment and another embodiment of the present invention and one model according to a comparative example, and FIG. 9 is an embodiment of the present invention and FIG. It is a diagram showing the average and standard deviation of prediction accuracy for two models according to another embodiment and one model according to a comparative example.

Referring to the drawings, the artificial neural network with 11 heart rate variability characteristics in the comparative example has the lowest curve area (AUC) value (0.71), and AUC when using the 4 QRS complex form features according to an embodiment of the present invention reached 0.99. In addition, when the heart rate variability characteristic and the QRS complex shape characteristic according to another embodiment of the present invention were used in combination, the AUC was improved to 1.

In addition, as shown in FIG. 9, the performance of the models according to one embodiment and another embodiment of the present invention showed a statistically significant difference from the comparative example (one-way ANOVA: F (2,297) = 208.34, p <0.001). As a result of the post-test, it can be seen that the comparative example has significantly lower performance than the one embodiment and the other embodiment of the present invention (p <0.001).

As such, the present invention shows that the predictive performance of ventricular fibrillation can be improved by using the features extracted from the QRS complex form (QRS complex display region and R-peak amplitude).

As shown in [Table 3], when using the four features of the QRS complex display area and R-peak amplitude as in one embodiment of the present invention, the prediction accuracy was 98.6%, and the same as in other embodiments of the present invention. A maximum prediction accuracy of 99.83% was achieved when the heart rate variability and QRS complex morphological features were used in combination.

However, as in the comparative example, when only the heart rate variability feature was used, the performance was greatly reduced with a prediction accuracy of 72%, and the ROC curve of FIG. 8 shows the same trend.

The results in [Table 3] show that the QRS shape feature can predict ventricular fibrillation more accurately before the onset than the existing heart rate variability feature. Therefore, it can be seen that the QRS morphology contributes the most to the prediction of ventricular fibrillation.

Looking at the reasons for this, as described above, the QRS complex of the electrocardiogram contains information about the ventricular depolarization process, the time the QRS complex is generated is the time required to complete the ventricular depolarization, and the QRS amplitude is consumed for the ventricular depolarization. proportional to the energy

Therefore, if the electrical characteristics of the ventricular tissue do not change, the QRS morphology does not change within a short period of time (seconds to several hours). Cardiac propagation patterns begin to change (which can affect QRS shape) when the ventricles start making reentrant waves for ventricular focus or other reasons.

In addition, different radio waves change the way the ventricles contract, so the location of the ventricular tissue, the source of the electrocardiogram, changes. ), and abnormalities in electrical activation characteristics can be extracted for early prediction of ventricular fibrillation.

The present invention highlights the importance of QRS complex morphological features for predicting such ventricular fibrillation, and the mean and standard deviation presented in [Table 2] show a comparison between the features of the ventricular fibrillation and control data sets.

It can be seen that the nine features have a statistically significant difference between ventricular fibrillation and the control group (p <0.05), and these features can be used to compare the ventricular fibrillation and control groups. However, prediction of ventricular fibrillation can be achieved by classifying complex patterns of all functions based on machine learning techniques, rather than simply comparing some parameters. Therefore, artificial neural network classifiers are used to classify complex patterns of features, and neural network models can be trained using larger datasets for better prediction accuracy.

Meanwhile, although the present invention has been described as a method for ventricular fibrillation prediction, a patch-type ECG measurement device is provided on the body of a patient who wants to check whether a prediction server and ventricular fibrillation have occurred, respectively, and the ECG signal detected from the ECG measurement device is predicted by the prediction server. , and the prediction server analyzes such an electrocardiogram signal to predict ventricular fibrillation.

In addition, the detailed configuration of the ventricular fibrillation prediction method of the present invention may be implemented as a computer program stored in a recordable medium, and consequently may be implemented in the form of a wearable electronic device.

As described above, the present invention has been described with reference to one embodiment shown in the drawings, which is merely exemplary, and those skilled in the art will appreciate that various modifications and equivalent other embodiments are possible therefrom. will understand

Accordingly, the true technical protection scope of the present invention should be defined by the technical spirit of the appended claims.

Claims (8)

In the ventricular fibrillation prediction method through machine learning based on the QRS complex shape feature,
The prediction server receives the ECG signal of the subject detected from the ECG measurement device to the prediction server, stores the ECG signal of the subject for learning with an artificial neural network in a database, and extracts data according to the ventricular fibrillation prediction parameter from the ECG signal ;
The prediction server includes the step of learning the extracted data with an artificial neural network (ANN),
The ventricular fibrillation prediction parameter further comprises a plurality of QRS complex features and a plurality of heart rate variability features extracted from the electrocardiogram signal,
wherein the plurality of heart rate variability features include a plurality of time domain features, a plurality of frequency domain features, and a plurality of Poincare nonlinear analysis features calculated using successive R-peak to R-peak intervals (RR intervals); ,
The plurality of QRS complex features are four QRS complex features, extracted from the QRS complex shape,
The QRS complex form is composed of the calculated QRS complex display region and Q, R and S waves from the R-peak,
Implementing a model that trains with an artificial neural network (ANN) using the feature data that adds the four QRS complex shape features and the plurality of heart rate variability features extracted as the ventricular fibrillation prediction parameters as input parameters to predict ventricular fibrillation through the learning A method of predicting ventricular fibrillation through machine learning based on the QRS complex morphology, characterized in that
delete delete The method of claim 1,
The plurality of time domain features are four, respectively, the mean of the RR intervals (represented as NN(RR) in MeanNN), the standard deviation (SDNN) of the NN(RR) intervals, and the mean square difference of consecutive NN(RR) intervals. QRS, characterized in that it consists of the interval difference ratio (pNN50) of successive NN(RR) intervals greater than 50 ms by the square root (RMSSD) and total NN(RR) intervals, and is defined by Equations (1) to (4) below. A method for predicting ventricular fibrillation through machine learning based on complex morphological features.

(Here, RR(i) means the interval between the i-th R-peak and R-peak, and i and N are natural numbers) The method of claim 1,
The plurality of frequency domain features are four, each consisting of a very low frequency (VLF) band (0 to 0.04 Hz), a low frequency (LF) band (0.04 to 0.15 Hz), and a high frequency (HF) band (0.15 to 0.4 Hz). The power spectral density (PSD) of each frequency band calculated using Welch's periodogram with a Hanning window (256 points with a window size of 50% overlap), the low frequency (LF) band and the high frequency ( HF) A method for predicting ventricular fibrillation through machine learning based on the QRS complex shape feature, characterized in that it consists of a ratio of the power spectrum density of the band. The method of claim 1,
The plurality of Poencare nonlinear analysis features are three, the i-th RR interval is the point of the vertical line and the variance of the point along the axis of the same line, the standard deviation of consecutive RR intervals (SD1), the standard deviation of the point along the axis of the same line (SD2), and the ratio of SD1 to SD2 (SD1/SD2), and based on the QRS complex form feature, characterized in that it is defined by Equations (5) to (6) below, ventricular fibrillation through machine learning Prediction method.

(where RR(i) is the interval between the i-th R-peak and the R-peak, and i is a natural number) delete The method of claim 1,
The four QRS complex features are composed of the QRS complex display area, the averages ((QRSaM, RPampM)) of the R-peak amplitudes, and the standard deviations (QRSaSD, RPampSD) of the means, as shown in Equations (7) to (10) below. A method for predicting ventricular fibrillation through machine learning based on the QRS complex morphological feature, characterized in that it is defined.

(where N is a natural number)

Images