KR20230076781A - Apparatus and method of classifying heart disease using mobilenet - Google Patents

Abstract

An apparatus for classifying heart diseases using a mobilenet according to an embodiment of the present invention includes an input unit for receiving an ECG signal in the time domain, and a wavelet transform unit for converting the ECG signal in the time domain into an ECG signal in the frequency domain using wavelet transformation. and a neural network that classifies the electrocardiogram signal in the frequency domain as one of AFIB (Atrial fibrillation), LBBB (Left bundle branch block beat), NSR (Normal sinus rhythm), and PVC (Premature ventricular contraction), wherein the neural network is trained It may be a MobileNet learned using a data set.

Description

Cardiac disease classification apparatus and method using mobile net {APPARATUS AND METHOD OF CLASSIFYING HEART DISEASE USING MOBILENET}

The present application relates to a heart disease classification apparatus and method using a mobile net.

An electrocardiogram shows the rhythmic beating of the heart represented by electrical signals. A healthy person's heart rate is between 60 and 100 beats per minute. When a person is exercising, tense or excited, the heart beats faster. In this case, there is usually no problem. However, if the heart beats irregularly for no reason, there is a problem with heart health, and these symptoms are called heart disease. Early detection of heart disease is very important because heart disease causes symptoms such as dizziness, fainting, chest pain, and shortness of breath when manifested, and can cause a heart attack.

Existing heart disease diagnosis methods cause time and cost inconvenience to patients who need to periodically visit a hospital and check up periodically. In addition, the electrocardiogram measured by the widely available smartwatch is data measured in a short time of less than 1 minute, and cannot contribute to identifying cardiovascular diseases including heart diseases in advance. Cardiac disease detection methods include methods for detecting early ventricular contractions using RR intervals, methods using discrete Fourier transforms, and methods based on Hilbert transforms.

Such a heart disease diagnosis method must be performed by an expert, and a large amount of irregularly generated electrocardiogram data must be secured for early detection of heart disease. However, when measuring ECG for a long time, a large amount of data must be analyzed, and there are many factors to be considered for accurate diagnosis, such as movement of the measurer and signal interference when measuring ECG in daily life. In addition, hardware performance due to the vast size of the deep learning structure can be an issue.

Korean Patent Publication No. 2008-0038512 (“System and method for providing driver stress index using electrocardiogram”, Publication date: May 7, 2008)

According to an embodiment of the present invention, it is possible to improve hardware performance due to a vast deep learning structure, and through this, it is possible to periodically use a smartphone or smart watch in daily life without a PC (Personal Computer) or a large device. Cardiovascular health can be managed and a heart disease classification device and method using a mobile net capable of securing a sufficient amount of training data set are provided.

According to an embodiment of the present invention, an input unit for receiving an electrocardiogram signal in a time domain; a wavelet transform unit converting the electrocardiogram signal in the time domain into an electrocardiogram signal in the frequency domain using wavelet transform; and a neural network that classifies the electrocardiogram signal in the frequency domain as one of AFIB (Atrial fibrillation), LBBB (Left bundle branch block beat), NSR (Normal sinus rhythm), and PVC (Premature ventricular contraction), wherein the neural network, Provided is a heart disease classification device using a mobile net, which is a mobile net learned using a training data set.

According to another embodiment of the present invention, a first step of receiving an electrocardiogram signal in the time domain from an input unit; a second step of converting the electrocardiogram signal in the time domain into an electrocardiogram signal in the frequency domain by using a wavelet transform in a wavelet transform unit; And a third step of classifying the electrocardiogram signal in the frequency domain into one of AFIB (Atrial fibrillation), LBBB (Left bundle branch block beat), NSR (Normal sinus rhythm), and PVC (Premature ventricular contraction) in a neural network, , The neural network provides a heart disease classification method using MobileNet, which is a MobileNet learned using a training data set.

According to another embodiment of the present invention, a computer-readable recording medium on which a program for executing the method on a computer is recorded is provided.

According to an embodiment of the present invention, an electrocardiogram signal in the time domain is converted into an electrocardiogram signal in the frequency domain using wavelet transform, and AFIB (Atrial fibrillation), LBBB (Left bundle branch block beat), NSR (Normal sinus rhythm) and PVC (Premature ventricular contraction), it is possible to improve the hardware performance due to the vast size of the deep learning structure, and through this, it is possible to improve the performance of the hardware without a PC (Personal Computer) or a large device. Cardiovascular health can be managed periodically in daily life using a smart watch.

In addition, according to an embodiment of the present invention, a sufficient amount of training data set can be secured by using a matching pursuit algorithm or by increasing the number of data by rotating each electrocardiogram signal in the time domain. .

1 is a block diagram of a heart disease classification apparatus using a mobile net according to an embodiment of the present invention.
2 is a diagram illustrating increased data using a matching pulse suit according to an embodiment of the present invention.
3 is a diagram illustrating a wavelet-transformed electrocardiogram signal according to an embodiment of the present invention.
4 shows the classification accuracy of the heart disease classification apparatus using mobilenet according to an embodiment of the present invention using 12-fold Cross Validation.
5 is a flowchart illustrating a heart disease classification method using a mobile net according to an embodiment of the present invention.
6 is a block diagram of a computer device capable of fully or partially implementing a heart disease classification device using a mobile net according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. However, the embodiments of the present invention can be modified in many different forms, and the scope of the present invention is not limited only to the embodiments described below. The shapes and sizes of elements in the drawings may be exaggerated for clearer explanation, and elements indicated by the same reference numerals in the drawings are the same elements.

1 is a block diagram of a heart disease classification apparatus using a mobile net according to an embodiment of the present invention.

As shown in FIG. 1, the heart disease classification apparatus 100 using a mobile net according to an embodiment of the present invention includes an input unit 111, a filter unit 112, a wavelet transform unit 113, and a training data generator. (115) and a neural network (114).

Specifically, the input unit 111 may receive an electrocardiogram signal in the time domain and transmit it to the filter unit 112 .

The filter unit 112 may low-pass filter the electrocardiogram signal in the time domain received from the input unit 111 . Here, the cut-off frequency of the low-pass filtering may be 130 Hz. A noise correction effect of 91.79% can be obtained through the filter unit 112 having such a cutoff frequency.

Meanwhile, the wavelet transform unit 113 may convert an ECG signal in the time domain into an ECG signal in the frequency domain using wavelet transform. The converted electrocardiogram signal in the frequency domain may be transmitted to the neural network 114 . 3 shows an exemplary wavelet-transformed electrocardiogram signal according to an embodiment of the present invention.

Thereafter, the neural network 114 may classify the electrocardiogram signal in the frequency domain into one of atrial fibrillation (AFIB), left bundle branch block beat (LBBB), normal sinus rhythm (NSR), and premature ventricular contraction (PVC).

According to one embodiment of the present invention, neural network 114 may be a MobileNet learned using a training data set.

The above-described mobile net is a lightweight deep learning model, and is a Convolution Neural Network (CNN) structure designed to be used where computer performance is limited or battery performance is important. This mobile net consists of well-known layers such as an input layer, a 2D layer, a normalization layer, a max pooling layer, a fully connected layer, and a softmax layer. block beat), NSR (Normal sinus rhythm), and PVC (Premature ventricular contraction). Since the internal structure and functions of these mobilenets are widely known, a detailed description thereof will be omitted.

Meanwhile, according to an embodiment of the present invention, a training data generation unit 115 for generating a training data set for learning of the above-described neural network 114 may be further included.

The training data generation unit 115 increases the number of data by generating a second number of time-domain ECG signals from the first number of time-domain ECG signals, and then uses wavelet transform to generate the second number of time-domain ECG signals. An electrocardiogram signal may be converted into an electrocardiogram signal in a frequency domain. Here, the second number may be greater than the first number.

When learning the neural network 114, accuracy may be improved as the number of data increases. Therefore, according to an embodiment of the present invention, in order to increase the number of ECG signals, the number of data can be increased by using a Matching Pursuit algorithm or by rotating each ECG signal in the time domain. . Of course, both of the above two methods may be used to increase the number of data.

The aforementioned Matching Pursuit algorithm is an algorithm capable of similarly copying ECG data. Electrocardiogram data can be increased 10 times by using the Matching Pursuit algorithm. 2(a) to (f) show increased data using a matching pulse suit according to an embodiment of the present invention.

In addition, each electrocardiogram signal in the time domain may be rotated to increase the number of data. By rotating the electrocardiogram signal in the time domain, the neural network 114 converts the electrocardiogram signal in the time domain before rotation and the electrocardiogram signal in the time domain after rotation. It can be recognized as other data.

The neural network 114 may be learned based on the training data set generated by the training data generation unit 115 described above.

Meanwhile, FIG. 4 shows the classification accuracy of a heart disease classification apparatus using a mobile net according to an embodiment of the present invention using 12-fold cross validation.

K-fold cross validation divides data into K groups, extracts one of the groups and uses it as a test set, and uses the remaining K-1 groups as a training set. use. Repeat K times, each test achieves one classification accuracy, then average K results to get the final performance of classification. In the present invention, as a result of measuring classification accuracy using 12-fold cross-validation, 9-fold showed 92.65%.

As described above, according to an embodiment of the present invention, an electrocardiogram signal in the time domain is converted into an electrocardiogram signal in the frequency domain using wavelet transform, and AFIB (Atrial fibrillation) and LBBB (Left bundle branch block beat), NSR (Normal sinus rhythm), and PVC (Premature ventricular contraction), it is possible to improve hardware performance due to a vast deep learning structure, instead of a PC (Personal Computer) or a large device. By converging with a smartphone or smart watch, you can manage your cardiovascular health periodically in your daily life.

In addition, according to an embodiment of the present invention, a sufficient amount of training data set can be secured by using a matching pursuit algorithm or by increasing the number of data by rotating each electrocardiogram signal in the time domain. .

Finally, FIG. 5 is a flowchart illustrating a heart disease classification method using a mobile net according to an embodiment of the present invention.

Hereinafter, a heart disease classification method using a mobilenet according to an embodiment of the present invention will be described in detail with reference to FIGS. 1 to 5 . However, for simplicity of the invention, descriptions of overlapping parts with those described in FIGS. 1 to 4 will be omitted.

The heart disease classification method using mobilenet according to an embodiment of the present invention may be started by receiving an electrocardiogram signal in the time domain from the input unit 111 (S501).

According to an embodiment of the present invention, as described above, the electrocardiogram signal in the time domain may be low-pass filtered, and the cut-off frequency of the low-pass filtering may be 130 Hz.

Next, the wavelet transform unit 113 may convert the electrocardiogram signal in the time domain into an electrocardiogram signal in the frequency domain using wavelet transform (S502). The converted electrocardiogram signal in the frequency domain may be transmitted to the neural network 114 .

Thereafter, the neural network 114 may classify the electrocardiogram signal in the frequency domain as one of AFIB (Atrial fibrillation), LBBB (Left bundle branch block beat), NSR (Normal sinus rhythm), and PVC (Premature ventricular contraction) (S503 ).

According to one embodiment of the present invention, neural network 114 may be a MobileNet learned using a training data set.

The above-described mobile net is a lightweight deep learning model, and is a Convolution Neural Network (CNN) structure designed to be used where computer performance is limited or battery performance is important. This mobile net consists of well-known layers such as an input layer, a 2D layer, a normalization layer, a max pooling layer, a fully connected layer, and a softmax layer. block beat), NSR (Normal sinus rhythm), and PVC (Premature ventricular contraction) can be classified as described above.

Meanwhile, according to an embodiment of the present invention, a training data generation unit 115 for generating a training data set for learning of the above-described neural network 114 may be further included.

The training data generation unit 115 increases the number of data by generating a second number of time-domain ECG signals from the first number of time-domain ECG signals, and then uses wavelet transform to generate the second number of time-domain ECG signals. An electrocardiogram signal may be converted into an electrocardiogram signal in a frequency domain. Here, the second number may be greater than the first number.

When learning the neural network 114, accuracy may be improved as the number of data increases. Therefore, according to an embodiment of the present invention, in order to increase the number of ECG signals, the number of data can be increased by using a Matching Pursuit algorithm or by rotating each ECG signal in the time domain. . Of course, both of the above two methods may be used to increase the number of data.

The aforementioned Matching Pursuit algorithm is an algorithm capable of similarly copying ECG data. Electrocardiogram data can be increased 10 times by using the Matching Pursuit algorithm. 2(a) to (f) show increased data using a matching pulse suit according to an embodiment of the present invention.

In addition, each electrocardiogram signal in the time domain may be rotated to increase the number of data. By rotating the electrocardiogram signal in the time domain, the neural network 114 converts the electrocardiogram signal in the time domain before rotation and the electrocardiogram signal in the time domain after rotation. It can be recognized as other data.

The neural network 114 may be learned based on the training data set generated by the training data generation unit 115 described above.

As described above, according to an embodiment of the present invention, an electrocardiogram signal in the time domain is converted into an electrocardiogram signal in the frequency domain using wavelet transform, and AFIB (Atrial fibrillation) and LBBB (Left bundle branch block beat), NSR (Normal sinus rhythm), and PVC (Premature ventricular contraction), it is possible to improve hardware performance due to a vast deep learning structure, instead of a PC (Personal Computer) or a large device. By converging with a smartphone or smart watch, you can manage your cardiovascular health periodically in your daily life.

In addition, according to an embodiment of the present invention, a sufficient amount of training data set can be secured by using a matching pursuit algorithm or by increasing the number of data by rotating each electrocardiogram signal in the time domain. .

Meanwhile, FIG. 6 is a block diagram of a computer device capable of fully or partially implementing a heart disease classification device using a mobile net according to an embodiment of the present invention. The heart disease classification device 100 using a mobile net shown in FIG. ) can be applied.

As shown in FIG. 6, the computer device 600 includes an input interface 601, an output interface 602, a processor 604, and a memory 605, and includes an input interface 601 and an output interface 602. , the processor 604 and the memory 605 may be interconnected through a system bus 603.

In an embodiment of the invention, the memory 605 is used to store programs, instructions or codes, and the processor 604 executes the programs, instructions or codes stored in the memory 605, and the input interface 601 It can control to receive signals and control the output interface 602 to transmit signals. The aforementioned memory 605 may include read-only memory and random access memory, and may provide instructions and data to the processor 604 .

In an embodiment of the present invention, the processor 604 may be a central processing unit (CPU), another general-purpose processor, a digital signal processor (DSP), an application specific integrated circuit (Application Specific Integrated Circuit) , ASIC), Field Programmable Gate Array (FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware component, etc. A general purpose processor may be a microprocessor, or the processor may be any conventional processor or the like.

In an implementation process, the method performed in each device of FIG. 1 may be achieved by an integrated logic circuit of hardware in the processor 604 or an instruction in the form of software. Contents of a method disclosed in relation to an embodiment of the present invention may be implemented to be performed and completed by a hardware processor or a combination of hardware and software modules of a processor to be performed and completed. The software module may be stored in memory, such as random access memory, flash memory, read only memory, programmable read only memory or electrically erasable programmable memory, registers, and the processor 604 reads the information in the memory 605; In combination with hardware, the contents of the above method are implemented.

The present invention is not limited by the above-described embodiments and accompanying drawings. It is intended to limit the scope of rights by the appended claims, and it is to those skilled in the art that various forms of substitution, modification and change can be made without departing from the technical spirit of the present invention described in the claims. It will be self-explanatory.

100: heart disease classification device
111: input unit
112: filter unit
113: wavelet transform unit
114: training data generating unit
115: neural network (mobilenet)

Claims (10)

an input unit for receiving an electrocardiogram signal in the time domain;
a wavelet transform unit converting the electrocardiogram signal in the time domain into an electrocardiogram signal in the frequency domain using wavelet transform; and
A neural network that classifies the electrocardiogram signal in the frequency domain as one of Atrial fibrillation (AFIB), Left bundle branch block beat (LBBB), Normal sinus rhythm (NSR), and Premature ventricular contraction (PVC),
The neural network is a mobile net (MobileNet) learned using a training data set, a heart disease classification device using a mobile net.
According to claim 1,
The heart disease classification device,
Further comprising a filter unit for low-pass filtering the electrocardiogram signal in the time domain;
The cut-off frequency of the low-pass filtering is 130 Hz, a heart disease classification device using a mobile net.
According to claim 1,
The heart disease classification device,
Further comprising a training data set generating unit for generating the training data set,
The mobile net performs learning using the training data set, a heart disease classification apparatus using a mobile net.
According to claim 3,
The training data set generating unit,
Increasing the number of data by generating a second number of electrocardiogram signals in the time domain, the second number being greater than the first number, from the first number of electrocardiogram signals in the time domain;
A heart disease classification apparatus using a mobile net for converting the second number of electrocardiogram signals in the time domain into electrocardiogram signals in the frequency domain using wavelet transform.
According to claim 4,
The training data set generating unit,
A heart disease classification apparatus using a mobile net, which increases the number of data by using a matching pursuit algorithm or by rotating each of the electrocardiogram signals in the first number of time domains.
A first step of receiving an electrocardiogram signal in the time domain from an input unit;
a second step of converting the electrocardiogram signal in the time domain into an electrocardiogram signal in the frequency domain by using a wavelet transform in a wavelet transform unit; and
In a neural network, a third step of classifying the electrocardiogram signal in the frequency domain as one of AFIB (Atrial fibrillation), LBBB (Left bundle branch block beat), NSR (Normal sinus rhythm), and PVC (Premature ventricular contraction),
The neural network is a mobile net learned using a training data set, a heart disease classification method using a mobile net.
According to claim 6,
The heart disease classification method,
A filter unit further comprising low-pass filtering the electrocardiogram signal in the time domain;
The cut-off frequency of the low-pass filtering is 130 Hz, a heart disease classification method using a mobile net.
According to claim 6,
The heart disease classification method,
generating the training data set in a training data set generator; and
A heart disease classification method using the mobile net, comprising the step of performing learning using the training data set in the mobile net.
According to claim 8,
The step of generating the training data set,
increasing the number of data by generating a second number of electrocardiogram signals in the time domain, wherein the second number is greater than the first number, from the first number of electrocardiogram signals in the time domain; and
and converting the second number of electrocardiogram signals in the time domain into electrocardiogram signals in the frequency domain using wavelet transform.
According to claim 9,
In the step of increasing the number of data,
A heart disease classification method using a mobile net, comprising increasing the number of data by using a matching pursuit algorithm or by rotating each of the electrocardiogram signals in the first number of time domains.

Images