KR101510522B1 - System for classification electrocardiogram signals - Google Patents

Abstract

An electrocardiogram signal classification system is provided. The electrocardiogram signal classification system according to an embodiment of the present invention includes a preprocessor for converting a first electrocardiogram signal to a second electrocardiogram signal, a characteristic extractor for extracting a characteristic vector from the second electrocardiogram signal, Wherein the classifier classifies the first electrocardiogram signal as one of a plurality of diseases using the characteristic vector.

Description

SYSTEM FOR CLASSIFICATION ELECTROCARDIOGRAM SIGNALS

The present invention relates to an electrocardiogram signal classification system, and more particularly, to an electrocardiogram signal classification system that improves the accuracy of signal classification using morphological characteristics.

Electrocardiogram (ECG) signal is an important signal used to judge the cardiovascular disease. By measuring the electrical signal generated in the heart, it is possible to judge the presence or absence of the conduction system from the heart to the electrode. .

The heart rhythm, which is the cause of ECG signal, is caused by the impulse that starts from the sinus node located in the right atrium, first depolarizes the left atrium and left atrium, and temporarily loses from the atrioventricular node Activate the posterior chamber. The right ventricle with the fastest Septum and thin wall is activated before the wall thickened left ventricle. The depolarization waves delivered to the Purkinje Fiber are spread from the endocardium to the external pericardium, such as the wavefront, in the myocardium, causing ventricular contraction. Because the electrical stimulus normally conducts through the heart, the heart contracts about 60 to 100 times per minute. Each contraction is represented by one heart rate.

Normal electrocardiogram signals are composed of PQRST waves, P wave is depolarization and contraction of the atrium, QRS is depolarization and contraction of the ventricle, and T wave represents ventricular depolarization. Physicians can infer cardiac status with a pattern of PQRST waves. These waves must have a standard form and the cardiac electrical activity is normal. In order to understand the standard form, it is necessary to determine whether the characteristics such as the time that each wave is held, the interval of each wave, the amplitude of each wave, and the kurtosis belong to the normal range. Recently, studies for classifying ECG signals automatically in real time have been carried out. In this regard, Korean Patent Laid-Open Publication No. 10-2008-0059369 discloses an electrocardiogram signal analyzing method for detecting an abnormal electrocardiogram.

The conventional electrocardiogram analysis algorithm has to analyze a large amount of data for a long time, and it is difficult to analyze the electrocardiogram signal in real time due to a large calculation process.

SUMMARY OF THE INVENTION The present invention has been made in view of the above problems, and it is an object of the present invention to provide an electrocardiogram signal classification system which improves the accuracy of analysis by classifying electrocardiogram signals using morphological characteristics.

It is another object of the present invention to provide an electrocardiogram signal classification system capable of shortening the time for analyzing an electrocardiogram signal and classifying it in real time.

The technical objects of the present invention are not limited to the above-mentioned technical problems, and other technical subjects not mentioned can be clearly understood by those skilled in the art from the following description.

According to an aspect of the present invention, there is provided an electrocardiogram signal classification system including a preprocessor for converting a first electrocardiogram signal into a second electrocardiogram signal, a processor for extracting a characteristic vector from the second electrocardiogram signal, And a classifier for classifying the characteristic vector using a characteristic extractor and a classifier, wherein the classifier classifies the first electrocardiogram signal as one of a plurality of diseases using the characteristic vector.

The feature vector may include one of fractal dimension (FD), skewness, kurtosis, spectral entropy (SE), power spectral density (PSD), QRS amplitude, QRS duration .

The fractal dimension may be a fractal dimension (HFD) of Higuchi.

The characteristic extracting unit may detect a QRS interval in the second electrocardiogram signal, and calculate the QRS amplitude from the QRS interval.

The QRS interval can be detected using the Pan and Tomkins algorithm.

The pre-processor may include a filter module, and may perform a filtering process to remove a baseline variation of the first electrocardiogram signal using the filter module.

The pre-processing unit may further include a sampling module, and may convert the filtered first electrocardiogram signal into the second electrocardiogram signal having a specific sampling frequency.

The sampling module may perform an interpolation process based on Fast Fourier Transform (FFT).

The classifier may comprise a K-proximity neighbor (K-NN) or a Gaussian mixture model (GMM).

According to the present invention, it is possible to classify the electrocardiogram signal into various morphological characteristics such as fractal dimension, thereby improving the accuracy of classification, and reduce the computational process required for the analysis using the sampling process and various algorithms.

1 is a diagram showing a schematic configuration of an ECG signal classification system according to an embodiment of the present invention.
2 is a diagram illustrating a specific configuration of an electrocardiogram signal classification system according to an embodiment of the present invention.
3 is a diagram illustrating a schematic processing procedure of an electrocardiogram signal in an electrocardiogram signal classification system according to an embodiment of the present invention.
4 is a diagram illustrating a detailed classification process of an electrocardiogram signal in the electrocardiogram signal classification system according to an embodiment of the present invention.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS The advantages and features of the present invention and the manner of achieving them will become apparent with reference to the embodiments described in detail below with reference to the accompanying drawings. The present invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art. To fully disclose the scope of the invention to those skilled in the art, and the invention is only defined by the scope of the claims. Like reference numerals refer to like elements throughout the specification.

Unless defined otherwise, all terms (including technical and scientific terms) used herein may be used in a sense commonly understood by one of ordinary skill in the art to which this invention belongs. Also, commonly used predefined terms are not ideally or excessively interpreted unless explicitly defined otherwise.

The terminology used herein is for the purpose of illustrating embodiments and is not intended to be limiting of the present invention. In the present specification, the singular form includes plural forms unless otherwise specified in the specification. The terms " comprises "and / or" comprising "used in the specification do not exclude the presence or addition of one or more other elements in addition to the stated element.

Hereinafter, an electrocardiogram signal classification system according to an embodiment of the present invention will be described with reference to the drawings.

1 and 2, a basic configuration of an electrocardiogram signal classification system according to an embodiment of the present invention is disclosed. FIG. 1 is a diagram showing a schematic configuration of an electrocardiogram signal classification system according to an embodiment of the present invention. FIG. 2 is a diagram showing a specific configuration of an electrocardiogram signal classification system according to an embodiment of the present invention. FIG. 4 is a diagram illustrating a detailed classification process of an electrocardiogram signal in the electrocardiogram signal classification system according to an embodiment of the present invention. Referring to FIG.

The electrocardiogram signal classification system 10 according to the present embodiment may include a preprocessing unit 100, a characteristic extracting unit 200, and a classifying unit 300.

Specifically, the electrocardiogram signal classification system 10 according to the present embodiment includes a preprocessor 100 for converting a first electrocardiogram signal into a second electrocardiogram signal, a characteristic extraction unit (not shown) for extracting a characteristic vector from the second electrocardiogram signal 200) and a classifier (300) for classifying the characteristic vector using a classifier, wherein the classifier classifies the first electrocardiogram signal as one of a plurality of diseases using the characteristic vector.

The preprocessing unit 100 may perform a preparation process for morphologically analyzing the received electrocardiogram signal. In this embodiment, this is referred to as a preprocessing process and may include a filter module 110 and a sampling module 120 to perform a preprocessing process.

Generally, electrocardiogram signals have a frequency band of about 0.05 to 100 Hz, and include various noise components due to internal and external factors. Electrical noise signals, which are especially important for ECG signal processing, include 60 Hz band-modulated and broadband white noise generated from the power supply line, peripheral noise due to peripheral devices or patient motion, and baseline And fluctuations. In the measurement and interpretation of electrocardiogram signals, there are many cases where the errors of such electrical noise signals are fatal.

QRS waves, P waves, and T waves are the most discriminating factors in heart disease judgment, and these signal components have similar characteristics in amplitude and time interval with general noise signals. That is, the distortion of the T wave and the ST portion provides a fatal adverse effect in terms of reliable heart disease judgment. Therefore, an algorithm capable of minimizing the distortion of these characteristic signals and effectively removing the noise signal is required.

In particular, the baseline fluctuation is a noise of a low frequency component of less than 1 Hz and has a frequency band similar to that of the ST portion which is small in size and low in frequency. Since the ST segment is a signal component that should be measured very accurately for the diagnosis of heart disease such as myocardial infarction or myocardial ischemia, minimizing distortion of the ST segment in removing baseline fluctuation noise must be considered.

For this purpose, the filter module 110 can effectively analyze the electrocardiogram signal by removing the baseline variation of the electrocardiogram signal. The filter module 110 may be a notch filter, but is not limited thereto. The filter module 110 may include various filter modules including a band reject filter (BRF) for rapidly attenuating a specific frequency. The first electrocardiogram signal that has undergone the filtering process can be transmitted to the sampling module 120 after the baseline variation is eliminated.

The sampling module 120 performs a sampling process so that the electrocardiogram signal has a specific frequency.

The sampling process can use Fast Fourier Transform (FFT). In general, the Fourier transform (FT) is used to observe the signal appearing on the time axis in the frequency domain. This can be used to analyze the frequency characteristics of signals measured over time or to analyze temporal characteristics of frequency data. In order to shorten the time for computing the Fourier transform, a signal is sampled at a specific time interval and digitized, which is referred to as fast Fourier transform.

The MIT / BIH database is a clinical data widely used in ECG signal processing and has a sampling frequency of 360 Hz. For the analysis of an effective electrocardiogram signal, various frequencies may be used for analyzing a signal, such as sampling the electrocardiogram signal at a sampling frequency of 360 Hz, but not limited thereto, sampling at a sampling frequency of 250 Hz of the ST-T database.

Although the electrocardiogram signal passes through the preprocessing unit 100 and removes baseline variation, it may further include various noise as described above. It is necessary to improve the signal-to-noise ratio by reducing the influence of such noise components. Generally, the slope of the R wave in an electrocardiogram signal is a signal characteristic used to find a QRS wave in an electrocardiogram signal analysis. Real-time derivative algorithms that provide analog circuitry or slope information are relatively easy to implement. However, the derivative can amplify unnecessary high frequency noise components. Also, in the pure derivative scheme, the abnormal QRS wave with large amplitude and long cycle disappears due to the relatively low R wave slope. Therefore, only the slope of the R wave has a limitation on the detection of an appropriate QRS wave.

For this, the characteristic extraction unit 200 can remove the interference of the noise from the electrocardiogram signal and detect the QRS wave using the Pan and Tompkins algorithm (Pan and Tompkins algorithm). The pan and Tompkins algorithms use band pass filters (BPFs), differential coefficients, and moving window integrators.

The electrocardiogram signal passes through a band pass filter (BPF) including a low pass filter (LPF) and a high pass filter (HPF) to eliminate various interference noises. The signal passed through the filter is differentiated to provide the slope information of the QRS wave through the derivative process. After differentiation, the signal is subjected to different calculation steps such as square, moving window integration, etc. to set the appropriate threshold value.

The characteristic extracting unit 200 can extract the feature vector of the electrocardiogram signal after the preprocessing process. It is possible to effectively classify the electrocardiogram signal using the characteristic vector. Since the electrocardiogram signal is a nonlinear signal, various features can be considered for signal classification. Most of the feature vectors may be grouped into one or several types in the time or frequency domain. In an exemplary embodiment of the present invention, a fractal dimension (FD), a QRS duration, a QRS amplitude, a spectral entropy, a kurtosis, And a power spectral density (PSD).

Fractal dimensions can be measured using a variety of methods. In one embodiment of the present invention, a Higuchi fractal dimension (HFD) may be used. The fractal dimension of Higuchi can effectively calculate the fractal dimension in discrete time series, and the influence of noise can be relatively small.

Construct a new time series k for m = 1, 2, ???, k in time series x (1), x (2), ???, x (N)

Where k and m are integers. The length of k and m can be defined as shown in Equation (2).

N is the total number of samples. The time intervals K and L (k) are calculated as the average of the K values. The Higuchi fractal dimension is a slope suitable for the pair of ln [Lk], ln (1 / k).

Spectrum entropy can quantify complex features in the spectrum of an ECG signal. The spectral entropy can be obtained by converting the electrocardiogram signal for fast Fourier transform. After calculating the power spectrum described below, a Shannon function can be applied to map each frequency to a numerical value. Entropy E can be expressed as:

Pn is the power spectral density at frequency n.

The QRS interval can be considered another important feature vector for ECG classification. The typical interval of QRS waves is between 0.06 and 0.12 seconds. The kurtosis can measure whether the electrocardiogram signal is a vertex or plane of the normal distribution. The kurtosis K can be expressed as Equation 4.

σ is the standard deviation of X, and μ is the average of X. Similarly, the asymmetry measures the asymmetry of the data and can be expressed as:

σ is the standard deviation of X, and μ is the average of X.

In one embodiment of the present invention, the power spectral density (PSD) may also be included as a morphological feature. The power spectral density can define the frequency distribution of the time series.

The classifier 300 may classify the feature vector as one of a plurality of diseases using a classifier. The classifier can use K-proximity neighbors (K-NN) and Gaussian mixture model (GMM). The two classification algorithms are generally used to classify time series signals. To classify electrocardiographic signals, these algorithms require a collection of technical features, which can take advantage of the morphological features described above.

K-Near Neighbors (K-NN) is a non-parametric learning algorithm and is a kind of distance-based classifier. The K-proximity neighbors assume that the data corresponds to the feature vector space. In general, the Euclidean distance can be used as a matrix of lengths.

The K-NN algorithm calculates the neighbors closest to the characteristic data of the learned ECG signal. Then, measure the distance between K neighbor and K nearest neighbor. K Nearest neighbors are examined, and the group to which most of them belong is displayed.

The Gaussian Mixture Model (GMM) is useful for classification of data and is generally used for parametric learning algorithms. The Gaussian mixture model can provide each group of data as a linear combination of Gaussian distributions. The parameters are derived from the learned data using a max-min algorithm. The expected value and the maximum value may be referred to as E step and M step, respectively. Step E can calculate the conditional expectation of a hidden variable. In step M, the initial model may be replaced with a new model if the log expectation value is higher than the previous step. The Gaussian mixture model is the sum of the weights of the m element densities and can be expressed as:

X is a D-dimensional property vector. Is the Gaussian time density, and is a mixed weight.

3 to 4, an ECG signal classification process in an ECG signal classification system according to an embodiment of the present invention is shown.

The first electrocardiogram signal may be received by the preprocessing unit 100 (S10).

Next, the received first electrocardiogram signal may pass through the filter module 110 to remove baseline variations, and the sampling module 120 may generate a second electrocardiogram signal sampled at a specific frequency for analysis of the signal S20).

When the second electrocardiogram signal is transmitted to the characteristic extraction unit 200, the QRS wave can be detected using the fan and the Tompkins algorithm (S30). The detected QRS can extract the feature vector using the morphological characteristic (S40). The morphological characteristic may include a fractal dimension of a QRS wave, a QRS interval, a QRS amplitude, a kurtosis, a spectral entropy, a power spectral density, and the like.

The extracted extraction vector may be classified as one of a plurality of diseases through a classifier using a K-nearest neighbor (K-NN) and a Gaussian mixture model (GMM) in the classifier 300 (S50).

Referring to FIGS. 3 to 4, an outline process of an electrocardiogram signal in an electrocardiogram signal classification system according to an embodiment of the present invention is shown.

First, a step of receiving a first electrocardiogram signal is performed (S10). The received first electrocardiogram signal may be subjected to a preprocessing process in the preprocessing unit 100 (S20). The preprocessing process may include a filtering and sampling process, and the preprocessing unit 100 may include a filter module 110 for removing noise and a sampling module 120 for sampling a signal having a specific frequency. After the first electrocardiogram signal is subjected to a preprocessing process, it may be changed to a second electrocardiogram signal and transmitted to the characteristic extraction unit 200.

Next, a process of selecting a characteristic of the second electrocardiogram signal can be performed (S30, S40). The characteristic extracting unit 200 may use a pan and a Tompkins algorithm to detect the most effective QRS wave in analyzing the second electrocardiogram signal (S30). The detected QRS wave can extract the characteristic vector according to the morphological characteristic (S40). Morphological characteristics may include fractal dimension, QRS wave amplitude, QRS wave interval, kurtosis and power spectral density.

Finally, a step of classifying the extracted feature vector as one of a plurality of diseases using a classifier can be performed (S50). The classifier can apply the K-neighborhood neighbors and the Gaussian mixture model algorithm.

While the present invention has been described in connection with what is presently considered to be practical exemplary embodiments, it is to be understood that the invention is not limited to the disclosed embodiments, but, on the contrary, You will understand. It is therefore to be understood that the above-described embodiments are illustrative in all aspects and not restrictive.

10: ECG signal classification system
100: preprocessing section
110: Filter module
120: Sampling module
200:
300:

Claims (9)

A preprocessing unit for converting the first electrocardiogram signal into a second electrocardiogram signal;
A feature extraction unit for extracting a feature vector from the second electrocardiogram signal; And
And a classifier for classifying the feature vector using a classifier,
Wherein the classifier classifies the disease of the first electrocardiogram signal as one of a plurality of diseases using the characteristic vector,
Wherein the characteristic extraction unit extracts a fractal dimension (FD) as a first characteristic vector from the second electrocardiogram signal, extracts a spectral entropy as a second characteristic vector from the second electrocardiogram signal, (QRS duration) as a third characteristic vector,
Wherein the classifier classifies the disease of the first electrocardiogram signal as one of a plurality of diseases through a non-parametric learning algorithm using the first characteristic vector, the second characteristic vector, and the third characteristic vector. Electrocardiogram signal classification system. The method according to claim 1,
The feature vector may include at least one of a fractal dimension, a skewness, a kurtosis, a spectral entropy, a power spectral density, a QRS amplitude, a QRS duration, ),
Wherein the non-parametric learning algorithm is a K-proximity neighbor (K-NN). 3. The method of claim 2,
Wherein said fractal dimension applies a fractal dimension (HFD) of Higuchi. 3. The method of claim 2,
Wherein the characteristic extractor detects the QRS interval in the second electrocardiogram signal and calculates the QRS amplitude of the QRS interval. 5. The method of claim 4,
Wherein the QRS interval is detected using a pan and a Tomkins algorithm. The method according to claim 1,
Wherein the preprocessing unit includes a filter module and performs filtering to remove a baseline variation of the first electrocardiogram signal using the filter module. The method according to claim 6,
Wherein the preprocessing unit further includes a sampling module and converts the filtered first electrocardiogram signal into the second electrocardiogram signal having a specific sampling frequency. 8. The method of claim 7,
Wherein the sampling module performs an interpolation process based on Fast Fourier Transform (FFT). The method according to claim 1,
Wherein the classifier classifies the disease of the first electrocardiogram signal as one of a plurality of diseases through at least one of the non-parametric learning algorithm and the parametric learning algorithm,
Wherein the non-parametric learning algorithm is a K-proximity neighbor (K-NN) and the parametric learning algorithm is a Gaussian mixture model (GMM).

Images