KR102264569B1 - Apparatuses and methods for classifying heart condition based on class probability output network - Google Patents

Abstract

The present invention relates to a method and apparatus for classifying a heart condition based on a class probability output network, and the method for classifying a heart condition according to an embodiment of the present invention includes the steps of estimating a parameter of a gamma distribution representing a statistical characteristic of an interval of an electrocardiogram signal. , calculating the distance between the center point of the parameter of the estimated gamma distribution and the parameter of the pre-learned gamma distribution, inputting the distance between the calculated center points to the class probability output network to determine the significance probability of the normal class and the abnormal class determining, and classifying the collected electrocardiogram signal as a normal wave or an abnormal wave by comparing significant probabilities of the determined normal class and the abnormal class with each other. In the heart condition classification method according to an embodiment of the present invention, when the ECG signal is classified as a normal wave, an interval of the normal ECG signal is additionally collected to calculate a correlation coefficient between a current interval and a next interval, and the calculated and determining the heart state as a normal state or an abnormal state using the correlation coefficient value.

Description

Method and apparatus for classifying heart condition based on class probability output network {APPARATUSES AND METHODS FOR CLASSIFYING HEART CONDITION BASED ON CLASS PROBABILITY OUTPUT NETWORK}

The present invention relates to a method and apparatus for classifying a heart condition based on a class probability output network.

The electrocardiogram (ECG) signal is an important signal used for judging diseases of the heart system. It is possible to determine the presence or absence of a disease by measuring the electrical signal generated from the heart and checking for abnormalities in the conduction system from the heart to the electrodes. It is a signal to be

The heartbeat, which is the cause of the electrocardiogram signal, is an impulse that starts at the sinus node located in the right atrium, first depolarizes the right and left atrium, and is delayed for a while at the atrioventricular node. then activate the ventricles. The septum is the fastest and the thin-walled right ventricle is activated before the thick-walled left ventricle. The depolarization wave transmitted to the Purkinje Fiber spreads from the endocardium to the epicardium like a wave in the myocardium, causing ventricular contraction. Because electrical impulses are normally conducted through the heart, the heart contracts about 60 to 100 times per minute. Each contraction is represented by one heart rate.

A normal electrocardiogram signal consists of PQRST waves, P wave represents atrial depolarization and contraction, QRS represents ventricular depolarization and contraction, and T wave represents ventricular depolarization. Doctors can infer heart conditions from the pattern of PQRST waves. When these waves have a standard shape, the electrical activity of the heart can be considered normal. In order to understand the normal state of the heart, it is necessary to determine whether characteristics such as the duration of each wave, the interval between each wave, the amplitude of each wave, and the kurtosis are within the normal range.

On the other hand, determination of the presence or absence of abnormality in heart health using characteristics used in the conventional heart rate variability (HRV) analysis takes a lot of time and does not provide a probability-based determination measure.

In addition, the conventional technique for determining whether there is an abnormality in heart health makes it difficult to determine the potential risk of the heart when the heartbeat is normal.

Korean Patent Registration No. 10-1689401 (Registration Date: 2016.12.19.)

Embodiments of the present invention provide a statistical-based characteristic effective for heart health and provide a probabilistic measure for heart health based on this, thereby providing more reliable diagnostic information, a class probability output network-based heart It is intended to provide a state classification method and apparatus.

According to an embodiment of the present invention, there is provided a method for classifying a heart condition performed by an apparatus for classifying a heart condition, the method comprising: estimating a parameter of a gamma distribution representing a statistical characteristic of an interval between electrocardiogram signals; calculating a distance between the parameter of the estimated gamma distribution and a center point of the parameter of the previously learned gamma distribution; determining the significance probability of a normal class and an abnormal class by inputting the calculated distance between the center points to a class probability output network; and classifying the ECG signal as a normal wave or an abnormal wave by comparing the determined significance probabilities of the normal class and the abnormal class with each other.

The interval between the ECG signals may be an R-R interval of the ECG signal or a beat-to-beat interval.

The calculating of the distance between the center points may include calculating a distance between the center points of the estimated gamma distribution parameter and the pre-learned gamma distribution parameter on a logarithmic scale.

The calculated distance between the center points may be output by a kernel function and input to a class probability output network.

In the determining of the significance probability, a Gaussian kernel function is assigned to the distance between the calculated center points, and a weight value is determined according to the assigned Gaussian kernel function to determine a normal class and an abnormal class for heart state classification. function can be calculated.

The determining of the significance probability may include estimating the distribution of the determined function for the calculated normal class and the abnormal class as a parameter of a beta distribution.

In the determining of the significance probability, the significance probability for the hypotheses for the normal class and the abnormal class may be determined by using the beta cumulative distribution function to which the parameter of the estimated beta distribution is reflected.

In the method, when the ECG signal is classified as a normal wave, the interval of the normal ECG signal is additionally collected to calculate a correlation coefficient between the current interval and the next interval, and the heart condition is determined to be normal using the calculated correlation coefficient value. The method may further include determining the state or the abnormal state.

The determining of the heart state as a steady state or an abnormal state includes calculating a significance probability for the steady state with a class probability output network by applying the calculated correlation coefficient value to the pre-learned beta distribution, and the calculated significance probability may be used to determine the heart state as a normal state or an abnormal state.

The determining of the heart state as the normal state or the abnormal state may include determining the heart state as the normal state when the calculated significance probability is equal to or greater than a threshold value.

On the other hand, according to another embodiment of the present invention, a memory for storing at least one program; and a processor connected to the memory, wherein the processor estimates, by executing the at least one program, a parameter of a gamma distribution representing a statistical characteristic for an interval of an electrocardiogram signal, and Calculate the distance between the center points of the parameters of the learned gamma distribution, input the distance between the calculated center points to the class probability output network to determine the significance probability of the normal class and the abnormal class, and By comparing significant probabilities with each other, ECG waves can be classified as standing waves or abnormal waves.

The interval between the ECG signals may be an R-R interval of the ECG signal or a beat-to-beat interval.

The processor may calculate a distance between the parameter of the estimated gamma distribution and a center point of the parameter of the previously learned gamma distribution on a log scale.

The calculated distance between the center points may be output by a kernel function and input to a class probability output network.

The processor assigns a Gaussian kernel function to the distance between the calculated center points, and determines a weight value according to the assigned Gaussian kernel function to calculate a decision function for a normal class and an abnormal class for heart state classification. have.

The processor may estimate the calculated distribution of the decision function for the normal class and the abnormal class as a parameter of the beta distribution.

The processor may determine the significance probability for the hypotheses for the normal class and the abnormal class by using the beta cumulative distribution function to which the parameter of the estimated beta distribution is reflected.

When the ECG signal is classified as a normal wave, the processor additionally collects intervals of the normal ECG signal to calculate a correlation coefficient between the current interval and the next interval, and uses the calculated correlation coefficient value to determine a normal heart condition. It can be determined as a state or abnormal state.

The processor applies the calculated correlation coefficient value to the pre-learned beta distribution to calculate a significance probability for a steady state with a class probability output network, and uses the calculated significance probability to determine the heart state as a steady state or an abnormal state. state can be determined.

The processor may determine the heart state as a normal state when the calculated significance probability is equal to or greater than a threshold value.

Meanwhile, according to another embodiment of the present invention, there is provided a non-transitory computer-readable storage medium including at least one program executable by a processor, wherein the at least one program, when executed by the processor, causes the processor to: Estimate the parameters of the gamma distribution representing the statistical characteristics of the ECG signal interval, calculate the distance between the estimated gamma distribution parameter and the center point of the pre-learned gamma distribution parameter, and calculate the distance between the calculated center points Including instructions for determining the significance probability of the normal class and the abnormal class by input to the class probability output network, and classifying the ECG wave as a normal wave or an abnormal wave by comparing the determined significance probability of the normal class and the abnormal class with each other, A non-transitory computer-readable storage medium may be provided.

Embodiments of the present invention may provide information on whether or not the heart is healthy using a small number of parameters using heartbeat interval information in heart health diagnosis.

Embodiments of the present invention may provide information on the presence or absence of abnormality in heart health in real time in an environment in which a portable heartbeat detector is provided.

Embodiments of the present invention may provide a measure of the potential risk of the heart even when the heartbeat is normal. That is, in the embodiments of the present invention, since it is already too late to detect an abnormality in the heartbeat pattern, even when the heartbeat pattern is found to be normal, how normal the heartbeat pattern is, and a symptom of a heart abnormality can be diagnosed in advance.

1 is a configuration diagram for explaining the configuration of a class probability output network-based heart state classification apparatus and a heartbeat detector according to an embodiment of the present invention.
2 is a flowchart of a method for classifying a heart state based on a class probability output network according to an embodiment of the present invention.
3 is a flowchart illustrating an operation of classifying a normal wave and an abnormal wave in a method for classifying a heart state based on a class probability output network according to an embodiment of the present invention.
4 is a flowchart illustrating an operation of classifying a normal state and an abnormal state in a method for classifying a heart state based on a class probability output network according to an embodiment of the present invention.
5 is a diagram showing the presence or absence of abnormality in an ECG waveform according to a parameter of a log-scale gamma distribution according to an embodiment of the present invention.
6 and 7 are diagrams showing the presence or absence of abnormalities in ECG waveforms according to SDNN characteristics and RMSSD characteristics in a method according to an embodiment of the present invention.
8 is a view illustrating an LDA output histogram and beta variable estimation according to a class of an ECG waveform used in an embodiment of the present invention.
9 to 12 are diagrams showing the simultaneous occurrence of the current and the next RR section according to the presence or absence of a heart abnormality according to an embodiment of the present invention.
13 is a diagram illustrating the distribution of correlation coefficients in current and next RR sections according to the presence or absence of heart abnormalities according to an embodiment of the present invention.
14 is a diagram illustrating a distribution of a significance probability (p-value) according to a class of the presence or absence of a heart abnormality according to an embodiment of the present invention.

Since the present invention can have various changes and can have various embodiments, specific embodiments are illustrated in the drawings and described in detail.

However, this is not intended to limit the present invention to a specific embodiment, it should be understood to include all modifications, equivalents and substitutes included in the spirit and scope of the present invention.

Terms such as first, second, etc. may be used to describe various elements, but the elements should not be limited by the terms. The above terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, a first component may be referred to as a second component, and similarly, a second component may also be referred to as a first component. and/or includes a combination of a plurality of related listed items or any of a plurality of related listed items.

When a component is referred to as being “connected” or “connected” to another component, it is understood that the other component may be directly connected or connected to the other component, but other components may exist in between. it should be On the other hand, when it is said that a certain element is "directly connected" or "directly connected" to another element, it should be understood that no other element is present in the middle.

The terms used in the present application are only used to describe specific embodiments, and are not intended to limit the present invention. The singular expression includes the plural expression unless the context clearly dictates otherwise. In the present 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 features It should be understood that this does not preclude the existence or addition of numbers, steps, operations, components, parts, or combinations thereof.

Unless defined otherwise, all terms used herein, including technical and scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Terms such as those defined in a commonly used dictionary should be interpreted as having a meaning consistent with the meaning in the context of the related art, and should not be interpreted in an ideal or excessively formal meaning unless explicitly defined in the present application. does not

Hereinafter, preferred embodiments of the present invention will be described in more detail with reference to the accompanying drawings. In describing the present invention, in order to facilitate the overall understanding, the same reference numerals are used for the same components in the drawings, and duplicate descriptions of the same components are omitted.

1 is a configuration diagram for explaining the configuration of a class probability output network-based heart state classification apparatus and a heartbeat detector according to an embodiment of the present invention.

As shown in FIG. 1 , the apparatus 100 for classifying a heart state based on a class probability output network according to an embodiment of the present invention is connected to a heartbeat detector 101 . The apparatus 100 for classifying a heart state based on a class probability output network according to an embodiment of the present invention includes a receiving module 110 , a memory 120 , and a processor 130 . However, not all illustrated components are essential components. The heart condition classification apparatus 100 may be implemented by more components than the illustrated components, or the heart condition classification apparatus 100 may be implemented by fewer components.

First, the heartbeat detector 101 may measure an electrocardiogram (ECG) signal by detecting a heartbeat. As an example, the heartbeat detector 101 may be implemented as a portable heartbeat detector. The electrocardiogram (ECG) signal may include information on an R-R interval of the heart or a heartbeat interval (beat-to-beat interval).

The heart condition classification apparatus 100 according to an embodiment of the present invention diagnoses whether there is an abnormality in heart health using R-R interval or beat-to-beat interval information of the heart.

The heart condition classification apparatus 100 according to an embodiment of the present invention performs a class probability output network (CPON)-based heart condition classification operation for classifying a heart condition from an electrocardiogram (ECG) signal. To this end, the heart state classification apparatus 100 analyzes the R-R interval of the ECG signal as a gamma distribution parameter and classifies it as a normal (NR) or abnormal (AN) ECG wave. In addition, in the case of a normal ECG wave, the cardiac state classification apparatus 100 finally determines whether the heart is in a normal state based on the correlation between the continuous ECG R-R interval and the long-term analysis. In an embodiment of the present invention, heart conditions may be classified through a hypothesis testing model for heart conditions using a class probability output network (CPON), which is a conditional class probability.

Specifically, the heart state classification apparatus 100 extracts a statistical feature from the heartbeat interval information provided from the heartbeat detector 101 as a parameter of a gamma distribution. In addition, the cardiac state classification apparatus 100 may provide a significance probability (p-value) of hypothesis testing for two states by using a probability-based classifier for a normal wave and an abnormal wave of the heart.

In addition, the cardiac state classification apparatus 100 calculates a correlation coefficient between RR intervals in order to estimate the potential risk of the heart when the heart wave is normal, and uses a probability-based hypothesis verification model to estimate the potential risk of the heart. A measure of risk can be provided.

Hereinafter, a detailed configuration and operation of each component of the heart condition classification apparatus 100 of FIG. 1 will be described.

The receiving module 110 receives an electrocardiogram signal that is heartbeat information detected by the heartbeat detector 101 .

The memory 120 stores at least one program.

The processor 130 is connected to the receiving module 110 and the memory 120 . The processor 130, by executing the at least one program, estimates a parameter of a gamma distribution representing a statistical characteristic of an interval of an electrocardiogram signal, and a center point between the estimated parameter of the gamma distribution and the parameter of the previously learned gamma distribution. Calculate the distance between the two, input the calculated distance between the center points to the class probability output network to determine the significance probability of the normal class and the abnormal class, and compare the determined significance probability of the normal class and the abnormal class with each other to generate an electrocardiogram wave is classified as a standing wave or an abnormal wave.

According to embodiments, the interval between the collected ECG signals may be an R-R interval of the collected ECG signals or a beat-to-beat interval.

According to embodiments, the processor 130 may calculate the distance between the parameter of the estimated gamma distribution and the center point of the parameter of the pre-learned gamma distribution on a logarithmic scale.

According to embodiments, the calculated distance between center points may be output by a kernel function and input to a class probability output network.

According to embodiments, the processor 130 allocates a Gaussian kernel function to the calculated distance between the center points, and determines a weight value according to the assigned Gaussian kernel function to determine a normal class and an abnormal class for cardiac state classification. A decision function can be calculated.

According to embodiments, the processor 130 may estimate the distribution of the calculated decision function for the normal class and the abnormal class as a parameter of the beta distribution.

According to embodiments, the processor 130 may determine the significance probability for the hypotheses for the normal class and the abnormal class by using the beta cumulative distribution function to which the parameter of the estimated beta distribution is reflected.

According to embodiments, when the collected ECG signal is classified as a normal wave, the processor 130 additionally collects the interval of the normal ECG signal to calculate a correlation coefficient between the current interval and the next interval, and the calculated correlation coefficient The value can be used to determine the heart condition as normal or abnormal.

According to embodiments, the processor 130 calculates a significance probability for a steady state with a class probability output network by applying the calculated correlation coefficient value to the pre-learned beta distribution, and uses the calculated significance probability to calculate the significance probability of the heart. The state can be determined to be a steady state or an abnormal state.

According to embodiments, the processor 130 may determine the heart state as a normal state when the calculated significance probability is equal to or greater than a threshold value.

2 is a flowchart of a method for classifying a heart state based on a class probability output network according to an embodiment of the present invention.

In this problem of classifying cardiac status, the R-R intervals detected from the ECG signal data are collected on a window of a finite size and used to obtain statistically meaningful data distribution information and to determine cardiac status. The heart condition classification process is repeated at every sampling time. Here, if the ECG wave of cardiac activity is completely normal, it is determined as normal (NR, Normal), otherwise it is considered as abnormal (AN, Abnormal).

An overall process of the method for classifying a heart condition according to an embodiment of the present invention is shown in FIG. 2 . Here, the automatic classification process of heart condition is as follows.

In step S101, the cardiac condition classification apparatus 100 detects the R-R interval from the ECG signal data. The detected R-R intervals are collected within a window of a fixed size (eg, 30 sec) every sampling time (eg, 10 sec).

In step S102, the cardiac state classification apparatus 100 estimates the parameter of the gamma distribution from the distribution of R-R intervals within the window.

In steps S103 and S104, the cardiac state classification apparatus 100 calculates, on a logarithmic scale, the distance between the parameter of the gamma distribution estimated for ECG waves of the NR and AN classes and the center value of the parameter of the previously learned gamma distribution to analyze the distance from the NR or AN center point.

In step S105, the cardiac state classification apparatus 100 outputs the given distance by the kernel function. The calculated distance is transmitted as an input of a class probability output network (CPON).

In steps S106 and S107, the cardiac state classification apparatus 100 determines p-value values for testing hypotheses of NR and AN classes.

In step S108, the cardiac state classification apparatus 100 compares the significance probability (p-value) values of the NR and AN classes to determine a normal or abnormal wave for the ECG pattern. If the significance probability of the NR class is greater than or equal to the significance probability of the AR class, the cardiac state classification apparatus 100 determines that the wave is a normal wave.

On the other hand, in step S109 , if the significance probability of the NR class is not equal to or greater than the significance probability of the AR class, the cardiac state classification apparatus 100 determines that the wave is an abnormal wave.

Here, if the R-R interval is classified as an NR class representing a normal ECG wave, the following procedure is applied to determine whether the heart is in a normal state as follows.

In step S110, the cardiac condition classification apparatus 100 collects normal R-R intervals for a fixed time window (eg, a 30-minute window).

In step S111, the cardiac state classification apparatus 100 calculates a correlation coefficient between the current R-R interval and the next R-R interval.

In step S112 , the cardiac state classification apparatus 100 applies the correlation coefficient value to the previously learned beta distribution to obtain a significance probability value for the normal heart state using the class probability output network (CPON).

The cardiac state classification apparatus 100 then determines whether there is an abnormality in the final cardiac state using the significance probability value. In step S113 , the cardiac state classification apparatus 100 checks whether the significance probability value is equal to or greater than a preset threshold value δ.

In step S114 , the cardiac condition classification apparatus 100 accepts the hypothesis for a normal heart condition when the significance probability value is greater than or equal to the preset threshold value δ. Here, as an example, a routine threshold is given as 0.05.

On the other hand, in step S114 , the cardiac state classification apparatus 100 classifies the heart state as an abnormal heart state when the significance probability value is less than the preset threshold value δ.

An embodiment of the present invention provides an accurate estimation of the conditional class probability because the kernel parameter as well as the beta distribution parameter are adjusted so that the output distribution of the classifier is closer to the ideal beta distribution in the training of the classifier. In addition, the class probability output network (CPON) according to an embodiment of the present invention may provide a measure of uncertainty for the final decision of classification by estimating a confidence interval for the conditional class probability.

3 is a flowchart illustrating an operation of classifying a normal wave and an abnormal wave in a method for classifying a heart state based on a class probability output network according to an embodiment of the present invention.

In step S201, the cardiac state classification apparatus 100 estimates a parameter of a gamma distribution representing a statistical characteristic of an interval of an electrocardiogram signal.

In step S202 , the cardiac state classification apparatus 100 calculates a distance between the parameter of the estimated gamma distribution and the center point of the parameter of the previously learned gamma distribution.

In step S203, the cardiac state classification apparatus 100 inputs the calculated distance between the center points to the class probability output network to determine the significance probability of the normal class and the abnormal class.

In step S204 , the cardiac state classification apparatus 100 classifies the collected ECG signal as a normal wave or an abnormal wave by comparing the determined significance probabilities of the normal class and the abnormal class with each other.

4 is a flowchart illustrating an operation of classifying a normal state and an abnormal state in a method for classifying a heart state based on a class probability output network according to an embodiment of the present invention.

It is performed when classified as a standing wave in the classification operation of the standing wave and the abnormal wave according to FIG. 3 .

In step S301, the heart state classification apparatus 100 classifies the collected electrocardiogram signal as a normal wave.

In step S302, the heart condition classification apparatus 100 further collects intervals of normal electrocardiogram signals.

In step S303, the cardiac state classification apparatus 100 calculates a correlation coefficient between the current interval and the next interval.

In step S304 , the heart state classification apparatus 100 determines the heart state as a normal state or an abnormal state by using the correlation coefficient value.

Meanwhile, in an embodiment of the present invention, a feature extraction operation using an interval between heartbeats will be described in detail.

5 is a diagram showing the presence or absence of abnormality in an ECG waveform according to a parameter of a log-scale gamma distribution according to an embodiment of the present invention.

)에 의하여 분석된다. 여기서, (

)는 각 각 감마 분포의 모양과 스케일을 결정한다. 이러한 분석은 2 개의 임의의 이벤트 사이의 간격이 감마 분포의 적절한 파라미터 값으로 설명될 수 있기 때문에 가능하다. 여기서, R-R 간격의 분포에 대한 감마 분포의 파라미터는 최우도 추정(maximum likelihood estimation) 방법에 의해 결정된다. R-R 간격은 ECG 파가 들어있는 유한 크기 창에서 추출되고, 감마 분포의 파라미터는 우도 함수를 최대화하는 방식으로 결정된다. 그 결과

개 의 R-R 간격들

에 대하여 파라미터 추정치

는 하기의 [수학식 1]로 계산된다. In an embodiment of the present invention, the method of classifying the heart condition is based on the RR interval of the heart as a parameter of the Gamma distribution (

) is analyzed by here, (

) determines the shape and scale of each gamma distribution. Such an analysis is possible because the interval between two arbitrary events can be described as an appropriate parameter value of the gamma distribution. Here, the parameter of the gamma distribution with respect to the distribution of the RR interval is determined by a maximum likelihood estimation method. The RR intervals are extracted from a finite size window containing the ECG waves, and the parameters of the gamma distribution are determined in such a way as to maximize the likelihood function. As a result

RR intervals of

parameter estimates for

is calculated by the following [Equation 1].

는 하기의 [수학식 2]로 계산된다.Parameter estimate

is calculated by the following [Equation 2].

는 하기의 [수학식 3]과 같이 계산된다. here

is calculated as in [Equation 3] below.

)의 값이 작아지면 비정상적인 ECG 파동의 가능성이 더 많이 존재함을 알 수 있다.As an example of the estimation of the parameter of the gamma distribution, the result of using the parameter estimation of the gamma distribution for a window of 30 seconds every 10 seconds in the MIT-BIH data set is shown in FIG. 5 . In Fig. 5, 1) normal and abnormal classes are well separated by parameters of gamma distribution, and 2) parameters of gamma distribution (

), it can be seen that there is more possibility of abnormal ECG wave.

Time equal to Standard deviation of all NN intervals (SDNN), Standard Deviation of the 5 minute Average NN intervals (SDANN), and Root mean square of the successive differences (RMSSD) in conventional heart rate variability (HRV) analysis Geometric methods including domain features and HRV trigonometric exponents are frequently used. However, here, SDNN and RMSSD, which are easy to analyze for a short time, are compared with the method according to an embodiment of the present invention.

SDNN is given by the following [Equation 5].

는

번째 R-R 간격을 의미하고,

은 R-R 간격에 대한 평균을 의미한다.

is

means the second RR interval,

is the mean for the RR interval.

RMSSD is given by the following [Equation 6].

6 and 7 are diagrams showing the presence or absence of abnormalities in ECG waveforms according to SDNN characteristics and RMSSD characteristics in a method according to an embodiment of the present invention.

Results calculated using SDNN and RMSSD in the same manner as in the experiment of FIG. 5 are shown in FIGS. 6 and 7 . These results show that the parameter of the gamma 7 distribution is more effective in discriminating abnormal ECG waves.

Meanwhile, the short-term analysis operation of the ECG wave will be described.

에 가우시안 커널 함수가 할당된다. 여기서 로그 스케일된 감마 분포의 파라미터의 공간에서 주어진 입력 패턴

에 대하여 가우시안 커널 함수는 하기의 [수학식 7과 같이 주어진다. After estimating the parameters of the gamma distribution in an embodiment of the present invention, two distributions representing normal and nonstationary classes are determined in the log-scaled space of the parameters of the gamma distribution. The center of each distribution to represent the two distributions

A Gaussian kernel function is assigned to . where the given input pattern in the space of parameters of the log scaled gamma distribution

A Gaussian kernel function is given as in Equation 7 below.

는 공분산 행렬을 나타낸다. 파라미터 c 값은 분류 출력 (이 경우에는 베타 분포)의 적절한 분포를 찾는 방식으로 결정된다.where c is the constant related to the variance of the Gaussian function,

denotes the covariance matrix. The value of parameter c is determined by finding an appropriate distribution of the classification output (in this case, the beta distribution).

The decision function for classifying the heart state is expressed as [Equation 8] below by a linear combination of Gaussian kernel functions.

는

번째 커널과 연결된 가중치를 의미한다.here

is

It means the weight associated with the second kernel.

를 결정하여야 한다. 이를 위하여 선형 판별 분석(linear discriminant analysis, LDA) 을 적용할 수 있다. 먼저

개의 로그 스케일된 감마 분포의 파라미터의 공간에서 주어진 입력 패턴

에 대하여 커널에 대한 출력 패턴

을 결정한다. The next step is weighting so that normal and abnormal ECG waves can be identified.

should be decided For this purpose, linear discriminant analysis (LDA) may be applied. first

A given input pattern in the space of parameters of the log-scaled gamma distribution of

About the output pattern for the kernel

to decide

및 비정상 패턴의 개수

의 합으로 볼 수 있다; 즉,

. 그러면 가중치 벡터

는 [수학식 9]와 같이 추정된다.Here, the total number of patterns is the number of normal patterns.

and the number of abnormal patterns.

can be seen as the sum of; In other words,

. Then the weight vector

is estimated as in [Equation 9].

,

,

은 각 각 정상과 비정상 패턴의 분포를 나타낸다.here

represents the distribution of normal and abnormal patterns, respectively.

를 결정하면 심장 상태를 분류하는 결정 함수

가 결정된다. 그러면 결정 함수 값의 분포를 베타분포를 이용하여 파악하게 된다. 실제로 분류 모델의 분포는 분류 출력이 한정된 범위에 존재하고 단일 분포(unimodal distribution)를 갖는다는 가정 하에 베타 분포에 의해 근사될 수 있다. 이러한 가정 하에서, 출력 값을 0과 1 사이의 값으로 정규화 하면 임의의 무작위 변수(random variable)

에 대하여 베타 확률 밀도 함수(Beta probability density function)는 [수학식 10]과 같이 정의된다.weight vector

A decision function that classifies heart conditions if we determine

is decided Then, the distribution of the decision function values is identified using the beta distribution. In fact, the distribution of the classification model can be approximated by the beta distribution under the assumption that the classification output exists in a finite range and has a unimodal distribution. Under this assumption, if the output value is normalized to a value between 0 and 1, it becomes a random variable.

For , the beta probability density function is defined as [Equation 10].

는 베타 확률밀도 함수의 파라미터를 나타내고,

는 베타 함수로 [수학식 11]과 같이 정의된다.here

represents the parameters of the beta probability density function,

is defined as [Equation 11] as a beta function.

And the beta cumulative distribution function is defined as [Equation 12].

는 모멘트 부합(moment matching) 방법으로 [수학식 13]과 같이 추정할 수 있다.Then the parameters of the beta probability density function

can be estimated as in [Equation 13] by the moment matching method.

는 각 각 정규 화된 결정 함수 값의 평균 및 분산을 의미한다. 이 방법은 샘플의 수가 작으면 부정확할 수도 있는데, 이런 경우 최우도 추정(maximum likelihood estimation) 방법으로 베타 확률밀도 함수의 파라미터를 결정할 수 있다.here

is the mean and variance of each normalized decision function value. This method may be inaccurate when the number of samples is small. In this case, the parameter of the beta probability density function can be determined using the maximum likelihood estimation method.

8 is a view illustrating an LDA output histogram and beta variable estimation according to a class of an ECG waveform used in an embodiment of the present invention.

As an example, the distribution of output values for normal and abnormal ECG waves and the estimation of the beta probability density function accordingly are shown in FIG. 8 .

을 하기의 [수학식 14]와 같이 결정하여 판단한다.On the other hand, after estimating the parameter of the beta distribution, the cardiac state classification apparatus 100 according to an embodiment of the present invention determines the normal or abnormal ECG wave by using the beta cumulative distribution function according to the hypotheses for normal and abnormal. Probability of significance (p value) for

is determined as in [Equation 14] below.

은 각 각 정상 및 비정상에 클래스에 대한 무작위 변수를 의미한다.here

denotes random variables for classes in normal and abnormal, respectively.

을 하기의 [수학식 15]와 같이 구한다.Then, for a given RR interval, a decision about a normal or abnormal ECG wave is made by comparing the significance probabilities for the hypotheses for normal and abnormal. That is, first the probability for the normal class

is obtained as in [Equation 15] below.

Then, the normal and abnormal classes are determined as in [Equation 16] below.

Here, a Kolmogorov-Smirnov (K-S) test can be performed to check whether the parameters of the estimated gamma or beta distribution fit well. Using the MIT-BIT data set as an example, the significance probability was obtained for the validation of each distribution by the K-S test (see Table 1). [Table 1] shows the significance probabilities for verification of gamma and beta distributions. As a result, a value much larger than the typical value (=0.05) of the significance level for the validation was obtained. That is, it was found that the estimation of the parameters of the gamma and beta distributions fit the given data distribution well. These results contribute to more accurate class posterior probability estimation for a given pattern in each classifier.

normal abnormal Gamma distribution 0.6083 0.5959 beta distribution 0.3281 0.3410

For validation of the method, the validity of an embodiment of the present invention was demonstrated using the MIT-BIH data set. As a result, it was shown that the proposed method is not only effective in classifying cardiac conditions, but also more accurate classification than existing classifiers such as k-NN and SVM. FIGS. 9 to 12 show cardiac abnormalities according to an embodiment of the present invention. It is a diagram showing the simultaneous occurrence of the current and the next RR section according to the presence or absence.

9 and 10 show simultaneous diagrams of current and next R-R intervals in the case of a steady-state heart.

11 and 12 show simultaneous diagrams of current and next R-R intervals in the case of an abnormal heart.

Meanwhile, a long-term analysis operation for heart health will be described.

가

개의 정상 ECG 파에 해당하는 R-R 간격의 시계열이라고 하면 현재 R-R 간격과 다음 R-R 간격 사이의 상관 계수

은 하기의 [수학식 17]과 같이 계산된다.A long-term analysis is necessary because a normal ECG waveform can be temporarily generated even in an abnormal heart condition. To this end, the dynamics of cardiac activity are investigated. One of these methods is to investigate the correlation coefficient between the current RR interval and the next RR interval. For example

end

If it is a time series of RR intervals corresponding to two normal ECG waves, the correlation coefficient between the current RR interval and the next RR interval is

is calculated as in [Equation 17] below.

이다. here,

to be.

9 to 12 are examples of scatter plots for the current and next R-R intervals for normal and abnormal heart conditions in order to confirm whether such a correlation coefficient is effective in distinguishing between normal and abnormal hearts. These examples show that in the case of a normal heart, two consecutive R-R intervals are highly correlated, unlike in the case of an abnormal heart.

13 is a diagram illustrating the distribution of correlation coefficients between current and next R-R sections according to the presence or absence of heart abnormalities according to an embodiment of the present invention.

In order to statistically indicate this, correlation coefficients for normal and abnormal heart conditions were collected, and their distribution was displayed as a histogram as shown in FIG. 13 . In addition, the parameters of the beta probability density function were estimated from the correlation data of normal cardiac conditions. As a result, it can be seen that the parameter values of the beta distribution are given as a=41.8859, b=2.1833, and the significance probability value of the K-S test for the beta distribution fit is given as 0.4430, which is much higher than the standard significance level of 0.05. Through this analysis, the determination of the heart state for the R-R interval corresponding to the normal ECG wave of a given sequence is determined as follows.

1. Using the parameters of the estimated beta distribution, the significance probability is obtained as in [Equation 18] below.

(통상적인 값은 0.05)를 이용하여

이면 정상 심장 상태에 대한 가설을 수락하고, 그렇지 않으면, 정상 심장 상태의 가설을 기각한다.2. Baseline significance level

(Typical value is 0.05)

If , the hypothesis of normal heart condition is accepted; otherwise, the hypothesis of normal heart condition is rejected.

The class probability output network using this beta distribution provides a significance probability for each class, so that more accurate and consistent decisions can be made for short-term and long-term analysis of heart conditions.

On the other hand, an abnormality test of the ECG wave will be described.

For the experiment of the method according to an embodiment of the present invention, ECG signal data from the MIT-BIH arrhythmia database was selected. The MIT-BIH arrhythmia database was selected from clinically significant arrhythmias by two independent cardiologists. Comments on and more bits are marked. For this experiment, a 30 sec window for ECG signal data was selected, the R-R interval within the window was used as the input pattern, and input pattern sequences sampled every 10 sec were collected as data sets for NR and AN class classification. As a result, 3,198 NR and 1,565 AN pattern sets were used as data sets for NR and AN class classification.

로 추정하고, 로그 스케일된 감마 분포의 파라미터의 공간

에서 NR 및 AN 클래스에 대한 자료를 수집하였다. 그 후에 각 자료에 대하여 2차원 가우시안 함수의 변수를 구하고(여기서 c값은 0.3으로 결정), 각 가우시안 함수에 따른 가중치 값은 LDA 방법으로 결정하여 심장상태 분률 위한 결정 함수

를 구하였다. 그런 다음 각 NR 및 AN 클래스에 대한 결정 함수의 분포를 베타 파라미터

로 추정하여 NR 및 AN 클래스에 대한 분류기(CPON)를 구현하였다. 본 발명의 일 실시예에 따른 방법에서 감마 분포의 파라미터의 효과를 비교하기 위하여 HRV(heart rate variability) 분석에서 흔히 쓰이는 SDNN 및 RMSSD 특징을 사용하여 같은 방법으로 분류기(CPON)를 구현하였다. 또한 본 발명의 일 실시예에 따른 CPON 분류기는 로그 스케일 감마 분포의 파라미터의 동일한 입력 자료와 NR 및 AN 클래스 레이블의 출력 데이터를 사용하여 kNN 및 SVM 분류기와 같은 널리 사용되는 분류 모델과 비교하였다. 여기서 kNN 및 SVM 분류기는 Scikit-learn Python 라이브러리를 사용하였는데 kNN 분류기의 경우 k=1, 즉 가장 가까운 이웃을 사용하였고 SVM 분류기의 경우 선형 커널을 사용하여 지원 벡터를 선택하였다. In the method according to an embodiment of the present invention, the distribution for the RR interval in each 30-second window is a parameter of the gamma distribution.

Estimated as , the space of parameters of the log-scaled gamma distribution

data on NR and AN classes were collected in After that, the variables of the two-dimensional Gaussian function are obtained for each data (here, the c value is determined to be 0.3), and the weight value according to each Gaussian function is determined by the LDA method to determine the heart condition fraction

was saved. Then, the distribution of the decision function for each NR and AN class was calculated as a beta parameter.

We implemented a classifier (CPON) for NR and AN classes by estimating . In order to compare the effects of parameters of gamma distribution in the method according to an embodiment of the present invention, a classifier (CPON) was implemented in the same way using SDNN and RMSSD features commonly used in heart rate variability (HRV) analysis. In addition, the CPON classifier according to an embodiment of the present invention was compared with widely used classification models such as kNN and SVM classifiers using the same input data of parameters of log scale gamma distribution and output data of NR and AN class labels. Here, the kNN and SVM classifiers used the Scikit-learn Python library. For the kNN classifier, k=1, that is, the nearest neighbor was used, and for the SVM classifier, the support vector was selected using a linear kernel.

In this experiment, the evaluation of each classifier was evaluated by a 10-fold evaluation method for a given data set. That is, the NR and AN patterns were equally divided into 10 parts, and one part out of the 10 parts was used as the test data and the remaining 9 parts were used as the training data to obtain the classification performance of the test data. This process was repeated 10 times in total, with 10 parts being used exactly once as test data. And the average value of classification performance for 10 data sets was determined as a result of the simulation.

측도의 네 가지 표준 성능 측정을 사용하였다.As classification performance, accuracy, precision, recall and

Four standard performance measures of measure were used.

,

,

은 각각 진정한 양수, 진정한 음수, 거짓 양수 및 거짓 음수를 나타낸다. 이에 대한 시뮬레이션 결과는 각각 [표 2]와 [표 3]에 요약되어 있다. 이러한 결과는 1) 본 발명의 일 실시예에 따른 감마 분포의 파라미터가 다른 특징들(SDNN 및 RMSSD) 보다 높은 정확도와

측도에서 개선된 성능을 보여준다. 그리고 2) 본 발명의 일 실시예에 따른 클래스 확률 출력망(CPON)은 이러한 분류에 있어서 다른 분류기(k-NN 및 SVM)와 비교하여 매우 효과적임을 보여준다. 이와 더불어 본 발명의 일 실시예에 따른 클래스 확률 출력망은 NR 및 AN 클래스 테스트에 대하여 조건부 확률 값(즉, 유의확률)을 제공하는 반면에 k-NN 또는 SVM과 같은 판별 함수 기반 분류기는 정규 화된 측정값이 아닌 임의의 범위 안에서 존재하는 판별 값만 제공한다.here

represents a true positive number, a true negative number, a false positive number, and a false negative number, respectively. The simulation results are summarized in [Table 2] and [Table 3], respectively. These results show that 1) the parameter of the gamma distribution according to an embodiment of the present invention has higher accuracy and higher accuracy than other features (SDNN and RMSSD).

The measure shows improved performance. And 2) it shows that the class probability output network (CPON) according to an embodiment of the present invention is very effective compared to other classifiers (k-NN and SVM) in this classification. In addition, the class probability output network according to an embodiment of the present invention provides conditional probability values (ie, significant probabilities) for NR and AN class tests, whereas discriminant function-based classifiers such as k-NN or SVM are normalized Only discriminant values that exist within an arbitrary range, not measured values, are provided.

[Table 2] shows the classification performance of the CPON classifier according to each feature variable.

feature variable Gamma parameter SDNN RMSSD accuracy 0.9733 0.9217 0.9221 + class
(normal) precision 0.9936 0.9944 0.9839 recall 0.9665 0.8884 0.8987

0.9799 0.9384 0.9394 + class
(abnormal) precision 0.9352 0.8127 0.8241 recall 0.9872 0.9898 0.9700

0.9605 0.8925 0.8911

[Table 3] shows the classification performance of each classifier using the parameters of the gamma distribution.

classifier CPON kNN SVM accuracy 0.9733 0.9660 0.9704 + class
(normal) precision 0.9936 0.9798 0.9913 recall 0.9665 0.9694 0.9644

0.9799 0.9745 0.9777 + class
(abnormal) precision 0.9352 0.9387 0.9310 recall 0.9872 0.9591 0.9827

0.9605 0.9488 0.9562

Meanwhile, a description will be given of the test for abnormalities in heart health. Experiments for long-term analysis of cardiac status were performed on ECG signal data from the MIT-BIH Normal Sinus Rhythm Database (NSRDB), the MIT-BIH Malignant Ventricular Arrhythmia Database (VFDB), and the MIT-BIH Atrial Fibrous Data. The NSRDB contains 18 long-term ECG waveform recordings from subjects without significant arrhythmias. The VFDB contains 22 hour and 30 minute ECG waveform recordings of subjects experiencing episodes of persistent ventricular tachycardia (VT), ventricular fibrillation (VF). AFDB contains 25 long-term ECG recordings from subjects of persistent atrial fibrillation (AF) or atrial flutter.

[Table 4] shows experimental data for long-term analysis of ECG normal waveforms.

material database recording number study material NSRDB 16265, 16272, 16273, 16410, 16483, 16786, 16795, 17052, 17453, 18177 test material NSRDB 18184, 19088, 19090, 19093, 19140
VFDB 418, 419, 420, 421, 422, 423, 424, 425, 428, 429, 602, 605, 607, 612, 614 AFDB 04015, 04048, 04746, 04908, 05121, 06426, 07162, 07879. 07910

측도의 측면에서의 성능이 평가되었다.For the time series set in NSRDB, RR intervals were used within the 30-minute window. Here, the first 4 hours of each recording were used as data, as indicated in Table 4. Then, the correlation coefficients between the current and next RR intervals were collected, and the distribution of the correlation coefficients was estimated by the beta distribution parameter. As a result, a beta distribution with estimated parameters (a=41.8859, b=2.1833) was obtained. The corresponding p-value for the KS test for beta distribution fit was calculated to be 0.4430. That is, it was found that the given data were in good agreement with the beta distribution. In this regard, the correlation coefficient distribution and the estimated beta distribution of the RR time series are plotted in FIG. 13 . On the other hand, for long-term analysis tests of cardiac status, RR intervals corresponding to normal ECG waveforms within a 30-minute window were collected and the first 4 hours of each recording were used. Here, VFDB used the first 30 minutes, and in the case of VFDB, only 30-minute recordings were given. For the given data, the correlation coefficient between the current and the next RR interval was calculated, and the significance probability for the steady-state test was calculated using the beta distribution. Then, for three levels of significance (0.05, 0.1, 0.2), the classification accuracy, precision, recall and

Performance in terms of measures was evaluated.

14 is a diagram illustrating a distribution of a significance probability (p-value) according to a class of the presence or absence of a heart abnormality according to an embodiment of the present invention.

[Table 5] shows the performance of determining the presence or absence of cardiac abnormalities using long-term analysis of ECG normal waveforms.

significance level

accuracy 0.9677 0.9892 0.9677 precision 0.9512 1.0000 1.0000 recall 0.9750 0.9750 0.9250

0.9630 0.9873 0.9610

From these experimental results, it was found that the optimal classification performance was given at the significance level of 0.1. Specifically, the precision and recall measures were determined to be 1.00 and 0.975, respectively. This means that at a significance level of 0.1, the classifier's decision is 100% accurate for a normal heart condition, meaning that only 2.5% of the normal heart condition is not captured. That is, when the significance value is greater than or equal to 0.1, it means that the reliability of the normal heart condition is very high. Histograms of the log-scaled p-values of correlation coefficients for the NDRDB, AFDB and VFDB test sets for comparison of the significance values of normal and abnormal heart conditions are shown in FIG. 14 . In this figure, it shows that the significance probability value is very small even when the ECG waveform is normal when the heart is abnormal. In fact, the mean significance values of NR, AF, and VT/VF in this test set were calculated to be 0.5375, 0.045 and 0.0015, respectively. In conclusion, an embodiment of the present invention analyzes the RR interval of ECG signal data by a parameter of gamma distribution for the classification of cardiac conditions, and uses a class probability output network (CPON) to determine whether the ECG wave is normal (NR) or abnormal. (AN) for classification. In one embodiment of the present invention, when the ECG wave is determined to be normal, the R-R interval is collected and the heart condition is further analyzed and confirmed. An embodiment of the present invention determines the heart state by calculating a correlation coefficient between the current and the next R-R interval for analysis thereof. Here, an embodiment of the present invention uses a conditional class probability value (ie, p-value) by using a class probability output network (CPON) as a classifier for heart conditions. As a result, it was confirmed that the condition of the heart or the potential risk of the heart was classified in a consistent and reliable way even when the ECG wave was normal. In addition, the method according to an embodiment of the present invention is a handheld or wearable device capable of detecting the RR (or between the heartbeat and the heartbeat) interval of the ECG signal data because it can be determined with a small number of parameters. easy to apply to

The class probability output network-based cardiac state classification method according to the above-described embodiments of the present invention may be implemented as computer-readable codes on a computer-readable recording medium. The class probability output network-based cardiac state classification method according to the embodiments of the present invention may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable recording medium.

A non-transitory computer-readable storage medium comprising at least one program executable by a processor, wherein the at least one program, when executed by the processor, causes the processor to: Gamma representing a statistical characteristic of an interval of an electrocardiogram signal Estimate the distribution parameter, calculate the distance between the estimated gamma distribution parameter and the center point of the pre-learned gamma distribution parameter, and input the distance between the calculated center points to the class probability output network to determine normal class and abnormal A non-transitory computer-readable storage medium comprising instructions for determining the significance probability of a class and comparing the determined significance probability of the normal class and the abnormal class with each other to classify the electrocardiogram wave as a normal wave or an abnormal wave may be provided. have.

The above-described method according to the present invention may be implemented as computer-readable codes on a computer-readable recording medium. The computer-readable recording medium includes any type of recording medium in which data that can be read by a computer system is stored. For example, there may be a read only memory (ROM), a random access memory (RAM), a magnetic tape, a magnetic disk, a flash memory, an optical data storage device, and the like. In addition, the computer-readable recording medium may be stored and executed as code readable in a distributed manner by being distributed in a computer system connected through a computer communication network.

Although described above with reference to the drawings and embodiments, it does not mean that the protection scope of the present invention is limited by the drawings or embodiments, and those skilled in the art will It will be understood that various modifications and variations of the present invention can be made without departing from the spirit and scope thereof.

Specifically, the described features may be implemented in digital electronic circuitry, or computer hardware, firmware, or combinations thereof. The features may be executed in a computer program product embodied in storage in a machine readable storage device, for example, for execution by a programmable processor. and features may be performed by a programmable processor executing a program of instructions for performing functions of the described embodiments by operating on input data and generating output. The described features include at least one programmable processor, at least one input device, and at least one output coupled to receive data and instructions from, and transmit data and instructions to, a data storage system. may be executed within one or more computer programs that may be executed on a programmable system comprising the device. A computer program includes a set of directives that can be used directly or indirectly within a computer to perform a particular action on a given result. A computer program is written in any form of programming language, including compiled or interpreted languages, and included as a module, element, subroutine, or other unit suitable for use in another computer environment, or as a standalone operable program. It can be used in any form.

Suitable processors for execution of a program of instructions include, for example, both general and special purpose microprocessors, and either a single processor or multiple processors of a different kind of computer. Storage devices suitable for implementing computer program instructions and data embodying the described features also include, for example, semiconductor memory devices such as EPROM, EEPROM, and flash memory devices, magnetic devices such as internal hard disks and removable disks. devices, magneto-optical disks and all forms of non-volatile memory including CD-ROM and DVD-ROM disks. The processor and memory may be integrated within or added by ASICs (application-specific integrated circuits).

Although the present invention described above has been described based on a series of functional blocks, it is not limited by the above-described embodiments and the accompanying drawings, and various substitutions, modifications and changes within the scope without departing from the technical spirit of the present invention It will be apparent to those of ordinary skill in the art to which the present invention pertains.

The combination of the above-described embodiments is not limited to the above-described embodiment, and various types of combinations may be provided in addition to the above-described embodiments according to implementation and/or necessity.

In the foregoing embodiments, the methods are described on the basis of a flowchart as a series of steps or blocks, but the present invention is not limited to the order of steps, and some steps may occur in a different order or at the same time as other steps as described above. can In addition, those of ordinary skill in the art will recognize that the steps shown in the flowchart are not exclusive, other steps may be included, or that one or more steps in the flowchart may be deleted without affecting the scope of the present invention. You will understand.

The foregoing embodiments include examples of various aspects. It is not possible to describe every possible combination for representing the various aspects, but one of ordinary skill in the art will recognize that other combinations are possible. Accordingly, it is intended that the present invention cover all other substitutions, modifications and variations falling within the scope of the following claims.

101: heart rate detector
100: heart condition classification device
110: receiving module
120: memory
130: processor

Claims (21)

A heart condition classification method performed by a heart condition classification device, the method comprising:
estimating a parameter of a gamma distribution representing a statistical characteristic of an interval of an electrocardiogram signal;
calculating a distance between the parameter of the estimated gamma distribution and a center point of the parameter of the previously learned gamma distribution;
determining the significance probability of a normal class and an abnormal class by inputting the calculated distance between the center points to a class probability output network; and
and classifying the ECG signal as a normal wave or an abnormal wave by comparing the determined significance probabilities of the normal class and the abnormal class with each other. According to claim 1,
The interval of the electrocardiogram signal is,
A class probability output network-based heart state classification method, which is an RR interval of the electrocardiogram signal or a beat-to-beat interval. According to claim 1,
Calculating the distance between the center points comprises:
A class probability output network-based heart state classification method for calculating a distance between the estimated gamma distribution parameter and a center point of the pre-learned gamma distribution parameter on a log scale. According to claim 1,
The method of classifying a heart state based on a class probability output network, wherein the distance between the calculated center points is output by a kernel function and input to a class probability output network. According to claim 1,
The step of determining the significance probability comprises:
A class probability output network for allocating a Gaussian kernel function to the distance between the calculated center points, and determining a weight value according to the assigned Gaussian kernel function to calculate a decision function for a normal class and an abnormal class for heart state classification based heart condition classification method. 6. The method of claim 5,
The step of determining the significance probability comprises:
A method for classifying a heart state based on a class probability output network, wherein the calculated distribution of the decision function for the normal class and the abnormal class is estimated as a parameter of a beta distribution. 7. The method of claim 6,
The step of determining the significance probability comprises:
A class probability output network-based cardiac state classification method for determining a significance probability for a hypothesis for a normal class and an abnormal class using a beta cumulative distribution function in which the parameter of the estimated beta distribution is reflected. According to claim 1,
When the electrocardiogram signal is classified as a normal wave, an interval of the normal electrocardiogram signal is additionally collected to calculate a correlation coefficient between the current interval and the next interval, and the heart condition is determined as a normal state or an abnormal state using the calculated correlation coefficient value. Further comprising the step of determining as, class probability output network-based heart state classification method. 9. The method of claim 8,
Determining the heart state as a normal state or an abnormal state comprises:
By applying the calculated correlation coefficient value to the pre-learned beta distribution to calculate a significance probability for a steady state with a class probability output network, and using the calculated significance probability to determine the heart state as a steady state or an abnormal state , a class probabilistic output network-based cardiac state classification method. 10. The method of claim 9,
Determining the heart state as a normal state or an abnormal state comprises:
If the calculated significance probability is greater than or equal to a threshold, determining the heart state as a normal state, a class probability output network-based heart state classification method. a memory storing at least one program; and
a processor coupled to the memory;
The processor, by executing the at least one program,
estimating the parameters of the gamma distribution representing the statistical characteristics of the intervals of the electrocardiogram signals,
calculating the distance between the parameter of the estimated gamma distribution and the center point of the parameter of the previously learned gamma distribution;
Determine the significance probability of the normal class and the abnormal class by inputting the calculated distance between the center points to the class probability output network,
A class probability output network-based heart state classification apparatus for classifying an electrocardiogram wave as a normal wave or an abnormal wave by comparing the determined significance probabilities of the normal class and the abnormal class with each other. 12. The method of claim 11,
The interval of the electrocardiogram signal is,
RR interval of the electrocardiogram signal or a heartbeat interval (beat-to-beat interval), a class probability output network-based heart state classification device. 12. The method of claim 11,
The processor is
A class probability output network-based heart state classification apparatus for calculating a distance between the estimated gamma distribution parameter and a center point of the pre-learned gamma distribution parameter on a log scale. 12. The method of claim 11,
The distance between the calculated center points is output by a kernel function and input to a class probability output network based on a class probability output network. 12. The method of claim 11,
The processor is
A class probability output network for allocating a Gaussian kernel function to the distance between the calculated center points, and determining a weight value according to the assigned Gaussian kernel function to calculate a decision function for a normal class and an abnormal class for heart state classification based heart condition classification device. 16. The method of claim 15,
The processor is
A class probability output network-based cardiac state classification apparatus for estimating distributions of decision functions for the calculated normal class and abnormal class as parameters of a beta distribution. 17. The method of claim 16,
The processor is
A class probability output network-based cardiac state classification apparatus for determining a significance probability for a hypothesis for a normal class and an abnormal class by using a beta cumulative distribution function reflecting the parameter of the estimated beta distribution. 12. The method of claim 11,
The processor is
When the electrocardiogram signal is classified as a normal wave, an interval of the normal electrocardiogram signal is additionally collected to calculate a correlation coefficient between the current interval and the next interval, and the heart condition is determined as a normal state or an abnormal state using the calculated correlation coefficient value. A heart condition classification device based on a class probability output network. 19. The method of claim 18,
The processor is
By applying the calculated correlation coefficient value to the pre-learned beta distribution to calculate a significance probability for a steady state with a class probability output network, and using the calculated significance probability to determine the heart state as a steady state or an abnormal state , a class probabilistic output network-based cardiac state classification device. 20. The method of claim 19,
The processor is
If the calculated significance probability is equal to or greater than a threshold value, the heart state classification apparatus based on a class probability output network determines the heart state as a normal state. A non-transitory computer-readable storage medium comprising at least one program executable by a processor, wherein the at least one program, when executed by the processor, causes the processor to:
estimating the parameters of the gamma distribution representing the statistical characteristics of the intervals of the electrocardiogram signals,
calculating the distance between the parameter of the estimated gamma distribution and the center point of the parameter of the previously learned gamma distribution;
Determine the significance probability of the normal class and the abnormal class by inputting the calculated distance between the center points to the class probability output network,
and instructions for classifying the electrocardiogram wave as a normal wave or an abnormal wave by comparing the determined significance probability of the normal class and the abnormal class with each other.

Images