KR101295072B1 - An apparatus of heart sound analysis base on simplicity and the method - Google Patents

Abstract

A heart sound analysis method is disclosed. The heart sound analysis method calculates simplicity of the heart sound from the measured heart sound signal and extracts a simplicity curve of at least one period, and applies a Gaussian curve to the extracted simplicity curve. And extracting a component corresponding to the heart noise of the heart sound signal from the extracted Gaussian fit curve, and detecting a heart sound component using the Gaussian fit curve from which the noise is removed. Thereby, the core sound main component can be detected correctly.

Description

SIMITY-based heart sound analysis device and method thereof {AN APPARATUS OF HEART SOUND ANALYSIS BASE ON SIMPLICITY AND THE METHOD}

The present invention relates to a heart sound analysis device and a method thereof, and more particularly, to a simplicity-based heart sound analysis device and a method for detecting a heart sound principal component using a Gaussian curve fitting method.

Phonocardiography (PCG) testing is widely used in clinical practice as the primary and primary means of early diagnosis of heart disease. However, doctors need a long time to listen to stethoscopes, and subjective judgments are needed when diagnosing heart disease with a stethoscope. Some heart sounds are not easily distinguished by human hearing. Recently, a number of precise test methods such as electrocardiogram (ECG), electrocardiogram (EMG), echocardiography, etc. have been used a lot, but there is a limit to the cost to use as an initial diagnostic means and to be diagnosed by a doctor in a hospital. Normal cardiac disease is normal during normal times, abnormal symptoms only appear when the disease is seizure, and urgent response is required to develop a ubiquitous health care terminal that can be used at home. Recently, with the launch of the electronic stethoscope, a lot of research on the PCG is being conducted, and the automatic heart sound analyzer using the electronic stethoscope can be used as a home health care system and medical assistant diagnostic equipment.

The normal heart sound is composed of the sound of S2 and S2 of closing the tricuspid and tricuspid plates, and S1 and S2 are the main components of the planting sound. Heart disease includes S2 cleavage (if the time difference between the two meniscus is closed), heart discomfort due to S3 or S4, and heart murmuring in the systolic or diastolic phase. Therefore, the presence of abnormal signals and the detection of the location of the occurrence of heart noise are important information for diagnosing heart disease.

Conventional heart sound analysis methods include energy-based envelope extraction, Haghighi-Mood and Torry proposed sub-band energy based envelope extraction, Shannon energy based envelope extraction, Hilbert transform based envelope extraction, wavelet based Methods, moment-based methods, and heart-based models. The method based on envelope extraction assumes that S1 and S2 are dominant and is incorrectly judged when the energy of heart noise is large. The method based on wavelet has the problem that low frequency heart noise is not separated. This decreases the performance. In 2005, Nigam proposed a simplicity-based heart analysis technique. Simplicity of relatively simple S1, S2, S3, and S4 heart signals has a relatively large value, and a few complex noises having a relatively complex shape It has a small value and is characterized by not being greatly influenced by energy. In 2008, Nigam proposed a method to analyze simplicity using fuzzy clustering. By simulating the simplicity into three classes by fuzzy clustering technique, it is determined that the class with the high simplicity is the heart sound component interval, the class with the simplicity value is the middle noise section, and the class with the simplicity value is the silent section. However, at this time, the inter-gap between rising and falling between the high and low values of heart sound simplicity is also divided into the middle class, and short peaks are generated in the middle class, which makes it impossible to accurately determine the heart noise. It is difficult to detect.

SUMMARY OF THE INVENTION The present invention is directed to the above-mentioned necessity, and an object of the present invention is to provide a heart sound analysis apparatus and method capable of accurately detecting a heart sound principal component by detecting a heart sound principal component using a Gaussian curve fitting method.

The heart sound analysis method according to an embodiment of the present invention for achieving the above object, the step of extracting a simplicity curve of at least one period by calculating the simplicity of the heart sound from the previously measured heart sound signal, Extracting a Gaussian fitting curve by applying a Gaussian curve fitting method to the extracted simplicity curve, removing a component corresponding to the heart noise of the heartbeat signal from the extracted Gaussian fitting curve, and removing the Gaussian fitting curve from which the noise is removed Detecting the core sound component using the.

The method may further include removing an offset of the extracted simplicity curve.

In the extracting of the Gaussian fitting curve, a Gaussian curve fitting method may be applied to the simplicity curve to extract a plurality of Gaussian components and a plurality of Gaussian parameters corresponding to the plurality of Gaussian components.

And, extracting the Gaussian fit curve,

The plurality of Gaussian components and the plurality of Gaussian parameters may be extracted using the following Gaussian curve fitting equation.

Where q is a Gaussian mix number, a, b and c are Gaussian parameters and mean weight, mean and standard deviation, respectively.

Preferably, the Gaussian mix number may be five.

The removing of the murmur noise may include determining whether each Gaussian component is a murmur noise using the extracted Gaussian parameter and removing the Gaussian component determined as the murmur noise among the plurality of Gaussian components. Can be.

The detecting of the heart sound principal component may include detecting a position of the heart sound principal component using a Gaussian parameter corresponding to the plurality of Gaussian components from which the heart noise has been removed, and using the plurality of Gaussian components from which the heart noise has been removed. The method may include detecting a duration of the heart sound main component.

The detecting of the heart sound main component may further include merging the overlapping Gaussian component into one Gaussian component when there is an overlapping Gaussian component among the extracted Gaussian components. On the basis of this, the position and duration of the heart sound main component can be detected.

The detecting of the heart sound main component may further include determining a type of the heart sound main component by using a difference between the interval between the heart systolic and the dilator.

On the other hand, the heart sound analysis apparatus according to an embodiment of the present invention, the simplicity unit for extracting a simplicity curve of at least one period by calculating the simplicity (simplicity) of the heart sound signal from the previously measured heart sound signal, the extracted simplicity An extractor for extracting a Gaussian fit curve by applying a Gaussian curve fitting method to a city curve, a noise canceller for removing a component corresponding to the heartbeat of the heartbeat signal from the extracted Gaussian fit curve, and a Gaussian fit curve from which the noise is removed It comprises a principal component detection unit for detecting the core sound component using.

The apparatus may further include an offset remover configured to remove an offset of the extracted simplicity curve.

Here, the extractor may apply a Gaussian curve fitting method to the simplicity curve to extract a plurality of Gaussian components and a plurality of Gaussian parameters corresponding to the plurality of Gaussian components.

The murmur removal unit may determine whether each Gaussian component is a murmur noise by using the extracted Gaussian parameter, and may remove the Gaussian component determined as the murmur among the plurality of Gaussian components.

On the other hand, the main component detector, the position detection unit for detecting the position of the heart sound main component using a Gaussian parameter corresponding to the plurality of Gaussian components from which the heart noise is removed and the heart sound main component using the plurality of Gaussian components from which the heart noise is removed It may include a duration detection unit for detecting the duration of the.

According to various embodiments of the present disclosure, the core sound component may be accurately detected by applying the Gaussian curve fitting method to the simplicity-based heart sound analysis apparatus and method.

1 is a flowchart illustrating a heart sound analysis method according to an exemplary embodiment.
2 is a block diagram showing the configuration of an apparatus for analyzing heart sounds according to an embodiment of the present invention.
3 shows an exemplary simplicity result of the heart sound signal.
4 and 5 show the results of the Gaussian curve fitting method when there is an offset and when the offset is removed, respectively.
6 to 8 show Gaussian curve fitting results when the Gaussian mix number is 3, 4 and 5, respectively.
9 is a histogram of ac_rate calculated after fitting a Gaussian curve having a mix number of 5 to a heart sound signal.
FIG. 10 illustrates a result of merging overlapping Gaussian components among the Gaussian components shown in FIG. 8.
FIG. 11 shows the results of detecting the position and duration of the core sound component from the Gaussian fit curve of FIG. 10.
12 to 15 show the results of simplicity based heart sound analysis using Gaussian curve fitting.

Hereinafter, the present invention will be described in detail with reference to the accompanying drawings.

1 is a flowchart illustrating a heart sound analysis method according to an exemplary embodiment. According to Figure 1, the heart sound analysis method, by calculating the simplicity of the heart sound from the previously measured heart sound signal to extract a simplicity curve of at least one cycle (S110), by applying a Gaussian curve fitting method to the extracted simplicity curve A Gaussian fit curve is extracted (S120). Then, the component corresponding to the heart noise of the heart sound signal is removed from the extracted Gaussian fit curve (S130). Thereafter, the core sound component is detected using the Gaussian fit curve from which the noise is removed (S140).

First, the step (S110) of extracting the simplicity curve will be described in detail. Simplicity is a measure of the complexity of a signal in a dynamic system. Simplicity based on the eigenvalue spectrum has a relatively good effect on the complexity of various physiological signals.

To calculate the simplicity of the heart sound signal, we first construct a trajectory matrix X. When the length of the frame is N, one subframe is composed of m samples, and a total of p = N-m + 1 subframes are constructed by moving one sample and p × m trajectory matrix as shown in Equation 1 below. Construct X.

The simplicity calculation process will be described with reference to Equation 2 below.

1. Obtain the covariance matrix C of the trajectory matrix as in Equation (1). Where p is the number of subframes to normalize the covariance matrix.

을 구하고 식 (2)과 같이 정규화 한다.2. Eigenvalues of covariance matrix C

Find and normalize as in Equation (2).

3. As shown in equations (3) and (4), the simplicity scale S is calculated using the entropy of the eigenvalues.

Repeat steps 1 to 3 while moving the analysis frame by one sample. In the above process, the frame length N and the subframe length m should be well selected because they affect the simplicity result.

를 계산하여 쉽게 한 주기의 심음 신호와 심플리시티 곡선을 추출할 수 있다. 여기서, 편의상 한 주기의 심플리시티 곡선을 추출하여 설명할 것이나, 한 주기 이상의 심플리시티 곡선을 추출하여 심음 신호를 분석할 수 있음은 물론이다.3 shows an exemplary simplicity result of the heart sound signal. According to FIG. 3, the heart sound main components S1 and S2 have a relatively simple signal and a large simplicity value, and the heart noise has a relatively complex signal, thereby reducing the simplicity value. Here, Gaussian white noise is added with a standard deviation of 0.02 in order to reduce the simplicity value of the signal-free section. The added white noise has a very small standard deviation, which significantly reduces the simplicity value of the signal-free section and has little effect on the signal. Simplicity also comes out periodically because heart sound is almost periodic

We can easily extract the heartbeat signal and simplicity curve of one period. Here, for convenience, the simplicity curve of one cycle will be extracted and described. However, the simplicity curve of one or more cycles can be extracted to analyze the heart sound signal.

Here, after extracting the simplicity curve, the method may further include removing an offset from the extracted simplicity curve. First, the simplicity of the heart sound signal is calculated, and the one-cycle simplicity curve is extracted by obtaining the heart sound period. The simplicity curve has an offset. This results in a Gaussian fit curve with a very large standard deviation when the Gaussian curve is fitted and gives unnecessary information, thus eliminating the offset as shown in Equation 3 below.

A result when the offset is removed will be described with reference to FIGS. 4 and 5. 4 and 5 show the results of the Gaussian curve fitting method when there is an offset and when the offset is removed, respectively. In FIG. 4 and FIG. 5, a thick line is the simplicity curve of a heart sound signal, and a dotted line is each Gaussian component when it is suitable by the Gaussian curve fitting method. In Figure 4, there is a Gaussian component with a very large standard deviation, which is created as an offset and does not fit the murmur correctly. In Figure 5 it can be seen that after offset removal, the murmur was suitably fitted to three Gaussians.

Next, a step (S120) of extracting a Gaussian fitting curve by applying a Gaussian curve fitting method to the simplicity curve will be described in detail. In the extracting of the Gaussian fitting curve, the Gaussian curve fitting method may be applied to the simplicity curve to extract a plurality of Gaussian components and a plurality of Gaussian parameters corresponding to the plurality of Gaussian components.

로 되며, q는 믹스 수가 된다. Curve fitting method is a technique that analyzes and evaluates signals by optimally modeling observation data, extracting model parameters, and is widely used in data analysis in statistics. The simplicity of the core sound component is close to the Gaussian shape, so it is reasonable to apply Gaussian curve fitting. Gaussian curve fitting model is shown in Equation 4 below. Here, a, b and c are Gaussian parameters, where a is the weight, b is the mean, c is the standard deviation,

Q is the number of mixes.

6 to 8, a description will be given of the Gaussian mix number that is appropriate. 6 to 8 show Gaussian curve fitting results when the Gaussian mix number is 3, 4 and 5, respectively. When the Gaussian mix number is 3 and 4, some simplicity signals are incorrectly fitted as shown in FIGS. 6 and 7. In FIG. 6, since the heart noise and the main sound component S2 are fitted as one Gaussian component, the heart noise cannot be detected. In FIG. 7, since the sound sound main components S2 and S3 fit into one Gaussian, S3 cannot be detected. Referring to FIG. 8, it can be seen that the simplicity signal can be accurately fitted when the Gaussian mix number is five.

The step S130 of removing the components corresponding to the heart noise of the heart sound signal from the Gaussian fit curve will be described in detail. The step of removing the murmur noise (S130) is a step of determining whether each Gaussian component is a murmur noise by using the Gaussian parameter extracted in the step of extracting a Gaussian fit curve (S120) and a Gaussian component determined as murmur noise among a plurality of Gaussian components. It may include the step of removing.

In general, simplicity values of the main sound components S1 and S2 are relatively large and the standard deviation is small, and simplicity values of the heart noise are relatively small and flat. Therefore, the Gaussian component that fits the core sound component after Gaussian curve fitting has a relatively large weight a and a relatively small standard deviation c. On the contrary, the Gaussian component that fits the core noise has a relatively small weight a and a large standard deviation c. If ac_rate, that is, the ratio of a and c in the Gaussian parameter is defined as in Equation 5 below, the ac_rate of the Gaussian component suitable for the heart sound component may be relatively large, and the ac_rate of the Gaussian component suitable for deep noise may be relatively small. Here, the value is too small and can be calculated by giving a 1000 times gain for the convenience of calculation.

Table 1 below shows the parameters of each Gaussian fit curve of FIG. 8. The ac_rate of the Gaussian component suitable for the heart sound main component may be relatively large, and the ac_rate of the Gaussian component suitable for the heart noise may have a relatively small value. An appropriate threshold value may be set in ac_rate to determine the main component of the heart sound if the threshold value is higher than or equal to the threshold, and to determine the noise level if the threshold value is lower than the threshold value.

9 is a histogram of ac_rate calculated after fitting a Gaussian curve having a mix number of 5 to a heart sound signal. In order to calculate the appropriate threshold of ac_rate, the histogram was plotted by removing the offset of the exemplary heart sound signal, fitting the Gaussian curve with the number of mixes to 5, and then calculating the ac_rate. As can be seen in Figure 5, the ac_rate of the murmur noise is less than 1.6 and shows a peak value at 1, and the ac_rate of the dominant sound components S1, S2, S3, and S4 has a peak value at 3.6. If ac_rate is less than 1.6 from the histogram, it can be determined as the heart noise, and if it is greater than 1.6, it can be determined as the main sound components S1, S2, S3, and S4, and Gaussian components determined as the heart noise can be removed.

Next, the step (S140) of detecting the heart sound principal component using the Gaussian fitting curve from which the heart noise is removed will be described in detail. Detecting the heart sound principal component (S140) includes detecting a position of the heart sound principal component using a Gaussian parameter corresponding to the plurality of Gaussian components from which the heart noise has been removed, and continuing the heart sound principal component using the plurality of Gaussian components from which the heart noise has been removed. Detecting time.

Here, when there is an overlapping Gaussian component among the extracted Gaussian components, the method may further include merging the overlapping Gaussian components into one Gaussian component, which will be described in detail with reference to FIGS. 8 and 10.

로 판단한다. FIG. 10 illustrates a result of merging overlapping Gaussian components among the Gaussian components shown in FIG. 8. When the Gaussian component suitable for heart noise is removed in FIG. Since the main components of S1 and S2 are spaced apart from each other at regular intervals, if the overlapping Gaussian components are refitted into a single Gaussian, each of the S1 and S2 core sounds can be extracted. Here, whether the Gaussian components overlap

Judging by.

로 게이트 하여 각 심음 성분의 지속시간을 검출할 수 있다.Table 2 below shows Gaussian parameters of the two Gaussian components of FIG. 10. Here, since the parameter b is the center position of the heart sound main components S1 and S2, the position of the heart sound main component can be detected. In addition, each Gaussian component

It can be gated to detect the duration of each heart sound component.

로 게이트 함으로써 심음 주성분 지속시간을 정확하게 측정할 수 있음을 알 수 있다.FIG. 11 shows the results of detecting the position and duration of the core sound component from the Gaussian fit curve of FIG. 10. From FIG. 11 to Gaussian component and Gaussian parameter of FIG.

It can be seen that by conducting the gate, the core sound component duration can be accurately measured.

On the other hand, detecting the heart sound main component (S140) may further comprise the step of determining the type of heart sound main component using the difference between the interval between the heart systolic and diastolic. After extracting the heart sound component from the Gaussian fitted curve, S1 and S2 can be easily determined by the heart sound characteristic that the cardiac systolic is shorter than the diastolic. In addition, if S3 and S4 are present, the systolic and diastolic devices can be determined by using the characteristic that the interval between S2 and S3 or S1 and S4 is the shortest, whereby S1, S2, S3 and S4 can be determined. In addition, when the positions S1 and S2 are determined, it is possible to determine whether the systolic heartbeat or the diastolic heartbeat according to the position of the heartbeat.

2 is a block diagram showing the configuration of an apparatus for analyzing heart sounds according to an embodiment of the present invention. According to FIG. 2, the heart sound analyzing apparatus 200 includes a simplicity unit 210, an extractor 220, a heart noise remover 230, and a principal component detector 240.

The simplicity unit 210 extracts a simplicity curve. In detail, the simplicity unit 210 may extract a simplicity curve of at least one period by calculating a silplicity of the heart sound from the previously measured heart sound signal. Here, the pre-measured heart sound signal may be previously stored in a storage unit (not shown), or may be measured using a heart sound measurement device (not shown) connected to the heart sound analysis device 200. The simplicity unit 210 may correspond to the step (S110) of extracting a simplicity curve of the heart sound analysis method according to an exemplary embodiment of the present disclosure, and a description thereof is omitted in detail with reference to FIG. 1.

Meanwhile, the heart sound analysis apparatus 200 may further include an offset remover (not shown). The offset remover removes an offset of the simplicity curve extracted by the simplicity unit 210. The offset remover may correspond to a step (not shown) of removing the offset of the heart sound analysis method according to an exemplary embodiment of the present disclosure, and a description thereof will be omitted in detail with reference to FIG. 1.

The extraction unit 220 extracts a Gaussian fit curve from the simplicity curve. Specifically, the extractor 220 may apply a Gaussian curve fitting method to the simplicity curve extracted by the simplicity unit 210 to extract a plurality of Gaussian components and a plurality of Gaussian parameters corresponding to the plurality of Gaussian components. have. The extraction unit 220 may correspond to the step (S120) of extracting a Gaussian fit curve of the heart sound analysis method according to an exemplary embodiment of the present disclosure, and a description thereof is omitted in detail with reference to FIG. 1.

The heart noise removing unit 230 removes a component corresponding to the heart noise from the Gaussian fitting curve. In detail, the murmur remover 230 may determine whether each Gaussian component is a murmur using a Gaussian parameter extracted by the extractor 220, and may remove the Gaussian component determined as the murmur among a plurality of Gaussian components. . The heart noise removing unit 230 may correspond to the step S130 of removing a component corresponding to the heart noise of the heart sound analysis method according to an embodiment of the present disclosure, and the description thereof is omitted in detail with reference to FIG. 1. do.

The principal component detection unit 240 detects a heart sound principal component using a moment characteristic waveform. To this end, the principal component detector 240 may include a position detector 241 and a duration detector 242. The position detector 241 detects the position of the heart sound main component using a Gaussian parameter corresponding to the plurality of Gaussian components from which the noise is removed, and the duration detector 242 uses the Gaussian component from which the noise is removed. The duration of the principal component can be detected. Principal component detection unit 240 may correspond to the step (S140) of detecting the heart sound principal component of the heart sound analysis method according to an embodiment of the present invention, the description thereof will be omitted in detail with respect to FIG.

On the other hand, the heart sound analysis apparatus 200 according to an embodiment of the present invention will be described as a separate configuration of the simplicity unit 210, extraction unit 220, heart noise removal unit 230 and the principal component detection unit 240, respectively However, the heart sound analysis apparatus 200 may be implemented in one configuration, such as a processor in which a program code for a heart sound analysis method is stored.

Hereinafter, an experimental result of the heart sound analysis method according to an exemplary embodiment of the present invention will be described with reference to FIGS. 12 to 15.

로 판정하며, 겹치는 가우시안 성분은 하나의 가우시안으로 다시 적합하여 S1, S2, S3, S4 심음 성분을 추출하였다. 다음, 다시 적합된 각 가우시안 곡선을

로 게이트 하여 S1, S2, S3, S4 심음 성분의 지속시간을 계산하였다.In order to evaluate the performance of the heart sound analysis method according to an embodiment of the present invention, 22 heart sound files published on the Internet were downsampled at 8 kHz. When simplicity was calculated, N = 50, m = 10 and white Gaussian noise were added as σ = 0.02. When the simplicity curve is fitted as a Gaussian, the number of mixes is 5, the ac_rate threshold for determining the noise is 1.6, and whether the remaining Gaussians overlap after removing the noise is

It was determined that the overlapping Gaussian components were refitted as one Gaussian to extract S1, S2, S3, and S4 core sound components. Next, we reconstruct each Gaussian curve

The gates were then gated to calculate the duration of the S1, S2, S3, and S4 planting components.

12 to 15 show the results of simplicity based heart sound analysis using Gaussian curve fitting. By the heart sound analysis method according to an embodiment of the present invention it can be seen that the position determination of each heart sound component and the duration of the S1, S2, S3, S4 heart sound component can be found accurately.

Gaussian curve fitting works better for a large number of mixes, but the Gaussian for fitting the noise is smaller, resulting in a higher ac_rate value and more difficult to remove.

When clustering into three classes with large simplicity value, middle class, and small class with fuzzy c-means clustering, the transient process of increasing or decreasing simplicity value is also clustered into intermediate class, eliminating short peaks. If the change in simplicity value is late, it is difficult to determine the murmur. The fuzzy c-means clustering technique was able to accurately gate 13 heart sounds among 22 heart sound files used in the experiment, and the heart sound analysis method using the Gaussian curve fitting method according to an embodiment of the present invention accurately recorded 18 heart sounds. Could gate.

The gate success rate is advantageous in that the method proposed in this paper is high and there is no difficulty in eliminating short peaks in the middle class compared to the fuzzy c-means clustering technique.

Although the above has been illustrated and described with respect to the preferred embodiments of the present invention, the present invention is not limited to the above-described specific embodiments, but in the technical field to which the invention belongs without departing from the spirit of the invention claimed in the claims. Anyone of ordinary skill in the art can make various modifications, and such changes are within the scope of the claims.

200: heart sound analysis device 210: simplicity unit
220: extraction unit 230: heart noise removal unit
240: principal component detection unit 241: position detection unit
242: duration detector

Claims (10)

In the heart sound analysis method,
Extracting a simplicity curve of the heart sound by adding noise to the measured heart sound signal;
Extracting a Gaussian parameter by applying a Gaussian curve fitting method to the extracted simplicity curve;
Determining a main component and a heart noise component of the heart sound signal using the Gaussian parameter to remove the heart noise component; And
Detecting the heart sound component from the heart sound signal removed by determining the heart noise component; Heart sound analysis method comprising a. The method of claim 1,
Extracting the Gaussian parameter,
And a plurality of Gaussian components and a plurality of Gaussian parameters corresponding to the plurality of Gaussian components by applying a Gaussian curve fitting method to the simplicity curve. delete delete The method of claim 2,
Removing the deep noise component,
Determining whether each Gaussian component is a deep noise component using the extracted Gaussian parameter; And
Removing a Gaussian component determined as the deep noise component among the plurality of Gaussian components. The method of claim 5,
Detecting the heart sound principal component,
And a position of the heart sound principal component through a Gaussian parameter analysis corresponding to a plurality of Gaussian components from which the heart noise component has been removed. delete delete delete delete

Images