KR20120094811A - A shockable signal detection method of automated external defibrillator using neural network with weighted fuzzy membership function - Google Patents

Abstract

PURPOSE: A shockable signal detection method of an automated external defibrillator using a neural network with a weight fuzzy membership function is provided to increase revive rate of heart-stop patients by accurately and quickly detecting shockable heart signals through neural network with a weighted fuzzy membership function. CONSTITUTION: A shockable signal detection of an automated external defibrillator using a neural network with a weight fuzzy membership function comprises: a step(S100) of collecting electrocardiograph signal; a step(S200) of detecting beats from collected electrocardiograph signal; a step(S300) of detecting shockable signals by quickly judging ventricular tachycardia, if pulse from the beats is over predetermined value; a step(S400) of pre-processing beat signals through signal filtering; a step(S600) of extracting characteristics from the pre-processed signals; and a step(S700) of detecting a shockable signal and non-shockable signal by using the characteristics as input characteristics of a neural network with a weighted fuzzy membership function.

Description

A SHOCKABLE SIGNAL DETECTION METHOD OF AUTOMATED EXTERNAL DEFIBRILLATOR USING NEURAL NETWORK WITH WEIGHTED FUZZY MEMBERSHIP FUNCTION}

The present invention relates to a shock signal detection method of a defibrillator for sudden stop resuscitation, and more particularly to a shock signal detection method of a defibrillator using a weighted fuzzy membership function based neural network.

In Korea, about 30,000 people die each year from Sudden Cardiac Arrest (SCA). Most of them are caused by life-threatening cardiac arrhythmias. Survival probability of patients with life-threatening arrhythmias is known to be closely related to rapid first aid by Automated External Defibrillator (AED). An automatic defibrillator is a device that senses and delivers a shockable signal capable of applying an electric shock in a non-hospital place without an ECG diagnosis by a medical professional and increases the survival rate of cardiac arrest patients.

The survival rate of sudden sudden stop patients in Korea is 4.6%, which is much lower than that of other countries in 15-40%. In order to solve this problem, there is a need for a method of rapidly performing early defibrillation using an automatic defibrillator and moving it to a hospital by an emergency rescuer. Arrhythmia in patients with sudden cardiac arrest is most commonly observed in coarse ventricular fibrillation (coarse VF) or rapid ventricular tachycardia (rapid VT), and the only treatment in sudden stop patients with coarse VF or rapid VT Rapid defibrillation. Survival without coarse VF or rapid VT without defibrillation within 10 minutes has little chance of survival, and the survival rate decreases by 7-10% per minute over time. Therefore, the earlier the defibrillation of these patients using AED, the greater the chance of survival. It is necessary to develop an algorithm that can accurately detect the shock signal of a sudden stop patient using such an AED.

Such a shock signal (coarse VF, rapid VT) detection algorithm for AED has been published continuously, and aims to increase recognition rate, shorten detection time, and simplify logic for small devices. The Time-Delay algorithm, released in 2007, detects VF through an 8-second ECG signal, but has a problem of using a CPU load, a slightly lower detection rate, and an 8-second signal. In addition, the Parameter Set algorithm detects Coarse VF / rapid VT, NSR and N (other arrhythmia) through a 10 second ECG signal, but has a problem of using a 10 second signal. The RBF detection algorithm, released in 2008, showed good results by detecting VF / VT and NSR / N. However, the RBF detection algorithm lacks generality by preselecting an experimental ECG signal. Therefore, there is a continuous demand for the development of an algorithm that reduces unnecessary defibrillation, detects defibrillation precisely when a shock signal is generated, and detects it in a short time.

The present invention has been proposed to solve the above problems of the conventionally proposed methods, using a weighted fuzzy membership function-based neural network to accurately detect shockable cardiac signals that can increase the survival rate of patients with heart disease in a short time. It is an object of the present invention to provide a shock signal detection method of a defibrillator using a weighted fuzzy belonging function based neural network, which can contribute to increasing the survival rate of acute stop patients.

In addition, the present invention, by providing a minimum fuzzy rule that can be implanted in a small mobile device, enabling the emergency treatment of a sudden sudden stop patients through real-time monitoring in conjunction with a mobile phone or PDA in the future, the ubiquitous environment Another object of the present invention is to provide a method for detecting shock signals of a defibrillator using a weighted fuzzy belonging function based neural network, which enables the remote management of the health status of a patient with heart disease through the convergence with a portable defibrillator required in a healthcare system. It is done.

Shock signal detection method of the defibrillator using a weighted fuzzy belonging function based neural network according to a feature of the present invention for achieving the above object,

(1) collecting an electrocardiograph signal;

(2) detecting bits from the electrocardiograph signal collected in step (1);

(3) detecting a shock signal by judging as a fast ventricular tachycardia if the pulse rate is greater than a predetermined value from the bit;

(4) preprocessing the bit signal through signal filtering;

(5) extracting features from the preprocessed signal in step (4); And

(6) detecting the shock signal and the non-shock signal by using the extracted feature as an input feature of the weighted fuzzy membership function based neural network;

Preferably, in step (4),

The signal filtering may be performed by wavelet transform.

Preferably, the step (5) is,

Detecting a cardiac contraction signal through a maximum peak mean value from the signal preprocessed in step (4),

The feature can be extracted if it is not a cardiac contraction signal.

Preferably, in step (5),

The feature may be extracted using at least one of a time delay method, a peak number extraction method, a specific coefficient extraction method before and after the tip, and a standard deviation feature extraction method of coefficients of a specific region.

Preferably, in step (6),

By using a non-overlapping area measurement method, a predetermined number of features may be selected as the input feature of the weighted fuzzy membership function based neural network.

More preferably,

The predetermined number of features may be six.

According to the present invention, a shock signal detection method of a defibrillator using a weighted fuzzy belonging function based neural network, using a weighted fuzzy belonging function based neural network, a shockable heart signal that can increase the survival rate of patients with heart disease for a short time Accurate detection within can contribute to increasing the survival rate of acute stop patients.

In addition, according to the present invention, by providing a minimum fuzzy rule that can be implanted in a small mobile device, in the future through the real-time monitoring in conjunction with a mobile phone or PDA, it is possible to more quickly emergency response of emergency stop patients, ubiquitous environment The fusion of portable defibrillators required by the healthcare system enables the remote management of the health of patients with heart disease.

1 is a flow diagram illustrating a shock signal detection method of a defibrillator using a neural network based on a weighted fuzzy membership function according to an embodiment of the present invention.
2 is a diagram illustrating a result of performing step S200 in a method for detecting a shock signal of a defibrillator using a neural network based on a weighted fuzzy membership function according to an embodiment of the present invention.
3 is a diagram illustrating a rapid VT signal detected by step S300 in a shock signal detection method of a defibrillator using a weighted fuzzy membership function based neural network according to an embodiment of the present invention.
4 is a diagram illustrating a filter bank for implementing dichotomous discontinuous wavelet separation that may be used in step S400 of a shock signal detection method of a defibrillator using a weighted fuzzy membership function based neural network according to an embodiment of the present invention.
5 is a diagram illustrating an application of a time delay method for feature extraction at step S600 of a method for detecting a shock signal of a defibrillator using a neural network based on a weighted fuzzy membership function according to an embodiment of the present invention.
6 is a diagram showing the number extraction of the tips for feature extraction in step S600 of the shock signal detection method of the defibrillator using a weighted fuzzy membership function based neural network in accordance with an embodiment of the present invention.
FIG. 7 is a diagram illustrating a specific coefficient extraction before and after a tip for feature extraction in step S600 of a shock signal detection method of a defibrillator using a weighted fuzzy membership function based neural network according to an embodiment of the present invention.
8 is a view showing the distribution of coefficients of a specific region for feature extraction in step S600 of the method for detecting a shock signal of a defibrillator using a weighted fuzzy membership function based neural network according to an embodiment of the present invention.
9 is a diagram illustrating a structure of a weighted fuzzy membership function based neural network used in a shock signal detection method of a defibrillator using a weighted fuzzy membership function based neural network according to an embodiment of the present invention.
10 is a diagram illustrating a non-overlapping area variance measuring method used in a shock signal detection method of a defibrillator using a neural network based on a weighted fuzzy membership function according to an embodiment of the present invention.
FIG. 11 is a diagram illustrating a change in an average value of detection performance according to the number of feature inputs in a shock signal detection method of a defibrillator using a weighted fuzzy membership function based neural network according to an embodiment of the present invention. FIG.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings, in order that those skilled in the art can easily carry out the present invention. In the following detailed description of the preferred embodiments of the present invention, a detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear. In the drawings, like reference numerals are used throughout the drawings.

In addition, in the entire specification, when a part is referred to as being 'connected' to another part, it may be referred to as 'indirectly connected' not only with 'directly connected' . In addition, the term 'comprising' of an element means that the element may further include other elements, not to exclude other elements unless specifically stated otherwise.

1 is a diagram illustrating a method of detecting a shock signal of a defibrillator using a neural network based on a weighted fuzzy membership function according to an embodiment of the present invention. As shown in FIG. 1, a method of detecting a shock signal of a defibrillator using a neural network based on a weighted fuzzy membership function according to an embodiment of the present invention includes: collecting an electrocardiograph signal (S100) and detecting a bit from an electrocardiograph signal Step S200, detecting the shock signal when the pulse rate is more than a predetermined value (S300), preprocessing the bit signal through wavelet transform (S400), and characterizing the preprocessed bit signal using a non-overlapping area measurement method. Extracting (S600), and detecting the shock signal and the non-shock signal (S700) by using the extracted feature as an input feature of the weighted fuzzy membership function-based neural network, and using the maximum peak average value. The method may further include detecting a heart contraction signal (S500).

In step S100, the ECG signal may be collected. Electrocardiograph (ECG) signals for shock signal detection can be collected for 8 seconds. According to a method of detecting a shock signal of a defibrillator using a neural network based on a weighted fuzzy membership function according to an embodiment of the present invention, since a shock signal can be detected within a short time, it is necessary to collect an ECG signal of about 10 seconds. Alternatively, the shock signal can be detected using only an ECG signal of about 8 seconds.

In step S200, a bit may be detected from the ECG signal collected in step S100. In step S200, a bit may be detected from the ECG signal collected in step S100 through a bit detection process, and a real time pulse and a QRS may be detected by a bit detection algorithm. The bit detection algorithm used in step S200 is also used for studies of arrhythmia detection, ECG diagnosis, Holter and heart rate variabillity (HRV). The most famous Hamilton algorithm may be used for the bit detection algorithm used in step S200 of the present invention.

The bit detection algorithm can be divided into seven stages. Low pass filtering, High pass filtering, Taking the derivative, Taking the absolute value of the signal, Averaging the absolute value over an 80㎳ window (Absolute value average of 80 ms window abnormal value), Peak Detection, and Rule Detection can detect the bit. Through this bit detection algorithm, the RR interval and the real-time heart rate of the ECG signal can be calculated. FIG. 2 is a diagram illustrating a result of performing step S200 in a shock signal detection method of a defibrillator using a neural network based on a weighted fuzzy membership function according to an embodiment of the present invention. FIG. 2A illustrates the ECG signal collected in step S100, and FIG. 2B illustrates a signal obtained by applying the signal of (a) to a bit detection algorithm.

In step S300, if the pulse rate is greater than or equal to a predetermined value from the bit, it may be determined as a fast ventricular tachycardia and the shock signal may be detected. The collected 8-bit bit signal is used to detect a rapid VT (fast ventricular tachycardia) signal within the shock signal range. According to the AAMI standard, a pulse rate of 180 beats / min results in a rapid VT and a shockable signal.

The shock signal can be detected by the pulse rate condition determination method in order to detect the rapid VT signal as soon as possible. In the pulse rate condition determination method, the real-time pulse number is calculated using the 8 second ECG bit signal detected in step 200. If the number of real-time pulses is greater than 24, the corresponding 8 second ECG signal is recognized as a rapid VR (shock signal). 3 is a diagram illustrating a rapid VT signal detected by step S300 in a shock signal detection method of a defibrillator using a weighted fuzzy membership function based neural network according to an embodiment of the present invention. As shown in FIG. 3, the rapid VT signal may appear as an 8 second ECG episode consisting of 26 beats. Since the number of pulses is larger than 24, it can be detected by rapid VT, i.e., a shock signal, by the present invention.

In operation S400, the bit signal may be preprocessed through signal filtering. If the real-time pulse number in step S300 is 24 or less, step S400 is performed as a preprocessing step for performing the next step, and the ECG signal may be filtered through wavelet transform. The wavelet transform compensates for the shortcomings of Fourier analysis, which provides global frequency characteristic information by simultaneously analyzing frequency characteristics at specific local points in time in signal processing. Discontinuous wavelet transform can separate a time-frequency signal into discontinuous signals of various scales.

4 is a diagram illustrating a filter bank for implementing bisected discrete wavelet separation that may be used in step S400 of a method for detecting a shock signal of a defibrillator using a weighted fuzzy membership function based neural network according to an embodiment of the present invention. As shown in FIG. 4, g (n), called detail, is a finite impulse response filters (FIR) high-pass filters coefficient associated with the wavelet coefficient, and h (n), called approximation, is a scale function coefficient. FIR low-pass filters coefficients associated with The signal past each filter is halved and converted back to the h (n) signal at the next scale level. The wavelet coefficient extracted by the wavelet transform is a similarity to the wavelet mother function, which represents a frequency signal over time given by a scale.

In step S500, the cardiac contraction signal ASYS may be detected from the signal preprocessed in step S400 through the maximum peak average value. In other words, in step S500, a maximum peak average (MPA) value may be used for the ASYS detection process.

The ASYS detection process detects cardiac asystole signals, and may use the Max Peaks Average (MPA) detection method to detect asystole signals in the shortest time. The MPA detection method uses wavelet transform in the preprocessing step S400. The five peaks having the largest peak value among the values obtained by the ECG signal through the wavelet transform are obtained to average the peaks. If the average value is smaller than the predetermined value, it is recognized as an asystole signal. If the average value is larger, the next step S600 is continued. In this case, the MPA detection method may perform the scale level 3 frequency band shown in FIG. 4 based on d3, and the predetermined value may be 0.2.

In step S600, the feature may be extracted from the signal preprocessed in the step S400. In this case, in operation S600, a predetermined number of features may be extracted based on d3 and d4 illustrated in FIG. 4. There are various methods for feature extraction, including at least one of a time-delay method, an advanced number extraction method, a front and rear specific coefficient extraction method, and a standard deviation feature extraction method of coefficients in a specific region. Feature can be extracted. Hereinafter, the four feature extraction methods mentioned above will be described in detail.

First, the d value by the time delay method can be extracted as a feature. The time delay method (Time-Delay Method or Phase Space Reconstruction, PSR) is a technique for analyzing dynamic waveforms or random signals based on phase space. When the ECG signal is called x (y + τ), two-dimensional diagrams are generated by mapping x (t) on the x-axis of the phase space and x (t + τ) on the y-axis. τ is 0.5 second as a time constant.

5 is a diagram illustrating an application of a time delay method for feature extraction in step S600 of a shock signal detection method of a defibrillator using a weighted fuzzy membership function based neural network according to an embodiment of the present invention. As shown in FIG. 5, when the normal waveform is applied to the time delay method, the diagram shows a regular shape and occupies a small space, but in the case of ventricular fibrillation, the diagram shows an irregular shape and occupies a large space. In particular, the feature d may be obtained by Equation 1 based on the configured phase space.

Second, the number of tips can be extracted as a feature. The tip number extraction method may receive the wavelet transformed ECG signal in step S400 and extract the tips from the tip. FIG. 6 is a diagram illustrating the number extraction of tips for feature extraction in step S600 of a shock signal detection method of a defibrillator using a weighted fuzzy membership function based neural network according to an embodiment of the present invention. As shown in FIG. 6, the number of tips of the non-shockable signal a and the number of tips of the shockable signal b are significantly different.

Third, specific coefficients before and after the tip can be extracted as features. Characteristic Features Extraction (CFE) method is the technique of detecting the peripheral coefficient value of the maximum step number based on the wavelet transform of step S400. The ECG signal is obtained by wavelet transform and before (b1, b2, b3, b4) the value of the last four coefficients at each peak having the largest value of the leading edge, and after (a1, a2). , a3, a4), we can calculate the average of each of the three before and after (a1 is the three values of a1) with the largest peak value. FIG. 7 is a diagram illustrating a specific coefficient extraction before and after a tip for feature extraction in step S600 of a shock signal detection method of a defibrillator using a weighted fuzzy membership function based neural network according to an embodiment of the present invention. As shown in FIG. 7, the number of tips of the non-shockable signal a and the value input to the shockable signal b are significantly different.

Fourth, it can be extracted by characterizing the standard deviation of the coefficients of a specific zone. The Standard Deviations of the Specific Intervals (SDSI) extraction method is a technique for detecting the standard deviation of coefficient values in a specific y-axis range from a signal based on wavelet transform.

FIG. 8 is a diagram illustrating a distribution of coefficients of a specific region for feature extraction in step S600 of a shock signal detection method of a defibrillator using a weighted fuzzy membership function based neural network according to an embodiment of the present invention. As shown in FIG. 8, all d3 counting points constituted by the wavelet transformed non-shockable signal (a) and the shockable signal (b) are displayed, and then the Y-axis based on the value obtained by the ECG signal through the wavelet transform. Extracting the values of the coefficients included in the range of values (± 0.05, ± 0.1, ± 0.2, ± 0.3) shows a big difference in the distribution between non-shockable and shockable cases. Thus, the standard deviation can clearly show the difference in distribution characteristics between non-shockable and shockable cases.

In operation S700, the shock signal and the non-shock signal may be detected using the extracted feature as an input feature of the weighted fuzzy membership function based neural network. The Neural Network with Weighted Fuzzy Membership Function (NEWFM) is a supervised fuzzy neural network that classifies classes using a boundary sum of weighted fuzzy membership functions learned from input.

9 is a diagram illustrating a structure of a weighted fuzzy membership function based neural network used in a shock signal detection method of a defibrillator using a weighted fuzzy membership function based neural network according to an embodiment of the present invention. As shown in Fig. 9, the structure of NEWFM is composed of three layers of input, hyperbox, and class. The input layer consists of n input nodes, and each input node receives one feature input. The hyperbox hierarchy consists of m hyperbox nodes, the l th hyperbox node B _l is connected to only one class node and has n fuzzy sets.

In operation S700, a predetermined number of features may be selected as the input feature of the weighted fuzzy membership function based neural network by using a non-overlapping area measurement method. The feature input from the neural network may be a feature extracted in operation S600. It can be a predetermined number of and inputting a good feature is a major factor for increasing the recognition rate of the neural network.

Feature selection is a process of selecting the best feature among the features extracted in step S600. Features can sometimes be detrimental to neural network performance. Even if they are not impaired, reducing the number of features can improve performance and reduce the complexity of the neural network. In the present invention, the recognition rate can be improved by selecting a feature using a non-overlapping area variance method, which can improve neural network performance.

FIG. 10 is a diagram illustrating a non-overlapping area variance measuring method used in a shock signal detection method of a defibrillator using a weighted fuzzy membership function based neural network according to an embodiment of the present invention. As shown in FIG. 10, the non-overlapping area variance method uses a bounded sum of WFM (BSWFM) of a weighted fuzzy membership function (WFM), and class 1 and class 2 are classified. This is a good feature when the two classes have a wider non-overlapping area and an even distribution.

Shockable signal 1442 of the total data using CUDB, MIT-BIHDB, and MIT-BIH-MVADB to evaluate the performance of a defibrillator shock signal detection method using a weighted fuzzy membership function based neural network according to an embodiment of the present invention. Dogs and 15811 non-shockable signals were sampled with an 8-second ECG signal.

FIG. 11 is a diagram illustrating a change in an average value of detection performance according to the number of feature inputs in a shock signal detection method of a defibrillator using a weighted fuzzy membership function based neural network according to an embodiment of the present invention. As shown in FIG. 11, the average value was measured highest when using six feature inputs. Therefore, in step S700, six features among the extracted features may be selected using the non-overlapping area measurement method to be input features of the fuzzy membership function based neural network.

Irena Jekova, “Shock advisory tool: Detection of life-threatening cardiac arrhythmias and shock success prediction by means of a common parameter set”, Biomedical Signal Processing and Control Volume 2, Issue 1, Pages 25-33, January 2007) and the shock signal detection method of the defibrillator using a weighted fuzzy membership function based neural network according to an embodiment of the present invention, the performance of the shock signal detection method is shown in Table 1 below. As can be seen in Table 1, the detection time of the existing study was 10 seconds and the sampling time of the present invention was detected as fast as 8 seconds to 2 seconds, the sensitivity of the existing study 91.4%, specificity 93.4% The sensitivity of the present invention was 91%, specificity of 97.1%, and average recognition rate of 95%.

The present invention may be embodied in many other specific forms without departing from the spirit or essential characteristics of the invention.

S100: collecting the ECG signal
S200: detecting bits from the electrocardiograph signal
S300: detecting a shock signal when the pulse rate is more than a predetermined value
S400: preprocessing the bit signal through wavelet transform
S500: detecting the cardiac contraction signal through the maximum peak mean value
S600: extracting features from preprocessed bit signals using non-overlapping area measurement
S700: detecting shock and non-shock signals using the extracted features as input features of the weighted fuzzy membership function based neural network

Claims (6)

A shock signal detection method of a defibrillator for sudden stop resuscitation,
(1) collecting an electrocardiograph signal;
(2) detecting bits from the electrocardiograph signal collected in step (1);
(3) detecting a shock signal by judging as a fast ventricular tachycardia if the pulse rate is greater than a predetermined value from the bit;
(4) preprocessing the bit signal through signal filtering;
(5) extracting features from the preprocessed signal in step (4); And
(6) detecting the shock signal and the non-shock signal by using the extracted feature as an input feature of the weighted fuzzy membership function based neural network, the shock of the defibrillator using the weighted fuzzy membership function based neural network. Signal detection method.
The method of claim 1, wherein in step (4),
The signal filtering is a shock signal detection method of a defibrillator using a weighted fuzzy membership function based neural network, characterized in that by the wavelet transform.
The method of claim 1, wherein step (5)
Detecting a cardiac contraction signal through a maximum peak mean value from the signal preprocessed in step (4),
A method for detecting a shock signal of a defibrillator using a weighted fuzzy belonging function based neural network, characterized in that the feature is extracted when it is not a heart contraction signal.
The method of claim 1, wherein in step (5),
A weighted fuzzy membership function based neural network, characterized in that the feature is extracted using at least one of a time delay method, a peak number extraction method, a front and rear specific coefficient extraction method, and a standard deviation feature extraction method of coefficients in a specific region. Shock signal detection method of the defibrillator using.
The method of claim 1, wherein in step (6),
Detecting shock signal of a defibrillator using a weighted fuzzy belonging function based neural network, by selecting a predetermined number of features from the extracted features using a non-overlapping area measurement method Way.
The method of claim 5,
The predetermined number of features, characterized in that the shock signal detection method of the defibrillator using a weighted fuzzy belonging function based neural network.

Images