KR20220008447A - Apparatus for cardiac diseases diagnoses using electrocardiogram based on deep learning and method thereof - Google Patents

Abstract

The present invention relates to a cardiac disease diagnosis apparatus using an electrocardiogram based on deep learning and a method thereof. According to the present invention, the cardiac disease diagnosis apparatus using an electrocardiogram based on deep learning comprises: an input unit receiving electrocardiogram data; a property extraction unit analyzing the electrocardiogram data to extract n properties for diagnosing the cardiac disease; a determination model generation unit generating identification codes for the n properties for each diagnosis of the cardiac disease, and generating n determination models determining each property by learning the generated identification codes; a diagnosis model generation unit generating a diagnosis model learning values determined from the determination models to derive a diagnosis of the cardiac disease; and a control unit using the determination models and the diagnosis model to output diagnosis results for electrocardiogram data to be diagnosed. According to the present invention, a deep learning algorithm is used to learn electrocardiogram data for each property, and learned models are used to derive a diagnosis of cardiac disease to improve diagnosis accuracy and suggest reasons for the diagnosis to improve diagnosis reliability.

Description

Heart disease diagnosis device and method using deep learning-based electrocardiogram

The present invention relates to an apparatus and method for diagnosing heart disease using a deep learning-based electrocardiogram, and more particularly, by using a deep learning algorithm to learn electrocardiogram data for each characteristic, and to derive a diagnosis name for heart disease using the learned model To a device for diagnosing a heart disease using an electrocardiogram based on deep learning and a method therefor.

According to the World Health Organization, cardiovascular disease is the leading cause of death today. More than 17.7 million people have died from cardiovascular disease, accounting for about 31% of all deaths, and more than 75% of these deaths occur in low- and middle-income countries.

Arrhythmia, a representative type of cardiovascular disease, is a disease that shows irregular changes in normal heart rhythm. This is included. Although a single arrhythmic heartbeat may not seriously affect life, a continuous arrhythmic heartbeat can lead to a fatal situation, so it is important to continuously monitor the heartbeat as a routine management and preventive measure.

Although many tests are used for the diagnosis of arrhythmias and coronary artery disease (cardiovascular disease), among them, the electrocardiogram (ECG) test has many advantages and is the most used test in clinical practice.

The electrocardiogram measures the electrical signal (electrocardiogram signal) generated by the heart and records the potential related to the heartbeat on the surface of the body as a graphic. to determine

In general, arrhythmia is not diagnosed as having a disease (1) or not (0), but when it is judged to be abnormal, that is, when it is judged as an arrhythmia, the type of arrhythmia must also be distinguished. ECG analysis and artificial intelligence models have been developed and are being used to diagnose arrhythmias.

1 is a diagram illustrating a conventional artificial intelligence model for diagnosing arrhythmias.

As shown in Fig. 1, it is an algorithm that classifies various arrhythmias at once when ECG data is input. A method of classifying 3 (asystole), abnormal 4 (ventricular fibrillation, abnormal 5 (ventricular tachycardia), and ??N was developed.

However, this method of separating the entire electrocardiogram data at once has a problem in that the accuracy is lowered.

The technology that is the background of the present invention is disclosed in Korean Patent No. 10-1109738 (published on February 24, 2012).

The technical problem to be achieved by the present invention is to learn electrocardiogram data by characteristics using a deep learning algorithm, and to derive a diagnosis name of a heart disease using the learned model. Provide a heart disease diagnosis apparatus and method using deep learning-based electrocardiogram it is to do

In accordance with an embodiment of the present invention for achieving this technical task, there is provided an apparatus for diagnosing a heart disease using an electrocardiogram based on deep learning, comprising: an input unit for receiving electrocardiogram data; a characteristic extracting unit that analyzes the electrocardiogram data and extracts n characteristics for diagnosing the heart disease; a decision model generation unit generating each identification code for the n characteristics for each diagnosis name of the heart disease, and generating n judgment models for determining each characteristic by learning the generated identification code; a diagnostic model generator for generating a diagnostic model for deriving a diagnosis name of the heart disease by learning values determined from the judgment model; and a controller for outputting a diagnosis result for the electrocardiogram data to be diagnosed using the judgment model and the diagnosis model.

In this case, the judgment model generating unit generates two identification codes divided into binary values for each characteristic so as to have a value corresponding to each diagnosis name, and clusters the same identification code into two groups for each characteristic. And, by learning the binary values of the two groups divided by using a deep learning algorithm, it is possible to generate the judgment model for classifying each characteristic.

In addition, the control unit may store the electrocardiogram data to be diagnosed and a diagnosis result for the data in a storage unit, and the judgment model generator may learn the stored electrocardiogram data and the diagnosis result using a deep learning algorithm. .

In addition, the input unit may further receive patient information including age and gender, and the determination model generator may generate the determination model by further learning patient information of the electrocardiogram data.

Also, the controller may output a diagnosis name and diagnosis reason for the electrocardiogram data to be diagnosed.

Further, according to another embodiment of the present invention, there is provided a method for diagnosing a heart disease using an electrocardiogram based on deep learning, the method comprising: receiving, by an apparatus for diagnosing a heart disease, electrocardiogram data; extracting n characteristics for diagnosing the heart disease by analyzing the electrocardiogram data; generating each identification code for the n characteristics for each diagnosis name of the heart disease; generating n judgment models for judging each characteristic by learning the generated identification code; generating a diagnostic model for deriving a diagnosis name of the heart disease by learning values determined from the judgment model; and outputting a diagnosis result for the electrocardiogram data to be diagnosed using the judgment model and the diagnosis model.

As described above, according to the present invention, by using a deep learning algorithm to learn electrocardiogram data by characteristics, and by deriving a diagnosis name of a heart disease using the learned model, diagnosis accuracy is improved, and the reason for diagnosis can be presented together. reliability can be improved.

In addition, according to the present invention, the diagnosis result can be classified into multivariate such as normal, arrhythmia type 1, arrhythmia type 2, arrhythmia type 3, ??N rather than binary classification like normal and abnormal (arrhythmia). Diagnostic results can be presented.

In addition, according to the present invention, an objective and accurate diagnosis is possible because a learning model is learned by a deep learning method that automatically extracts features by judging only by data, and it can be usefully utilized for diagnosing various diseases by learning various bio-signals. It works.

1 is a diagram illustrating a conventional artificial intelligence model for diagnosing arrhythmias.
2 is a block diagram illustrating an apparatus for diagnosing a heart disease using an electrocardiogram based on deep learning according to an embodiment of the present invention.
3 is a flowchart illustrating an operation flow of a method for diagnosing a heart disease using an electrocardiogram based on deep learning according to an embodiment of the present invention.
FIG. 4 is a diagram illustrating a process of generating an identification code for each characteristic for each diagnosis name in step S300 of FIG. 3 .
FIG. 5 is a diagram illustrating a decision model generated in step S400 of FIG. 3 .
6 is a diagram illustrating a process of outputting a heart disease diagnosis result according to an embodiment of the present invention.

Hereinafter, preferred embodiments according to the present invention will be described in detail with reference to the accompanying drawings. In this process, the thickness of the lines or the size of the components shown in the drawings may be exaggerated for clarity and convenience of explanation.

In addition, the terms to be described later are terms defined in consideration of functions in the present invention, which may vary according to intentions or customs of users and operators. Therefore, definitions of these terms should be made based on the content throughout this specification.

First, an apparatus for diagnosing a heart disease using an electrocardiogram based on deep learning according to an embodiment of the present invention will be described with reference to FIG. 2 .

2 is a block diagram illustrating an apparatus for diagnosing a heart disease using an electrocardiogram based on deep learning according to an embodiment of the present invention.

As shown in FIG. 2 , the apparatus 100 for diagnosing a heart disease using an electrocardiogram based on deep learning according to an embodiment of the present invention includes an input unit 110 , a characteristic extraction unit 120 , a judgment model generation unit 130 , and a diagnosis model generation. It includes a unit 140 , a control unit 150 , and a storage unit 160 .

First, the input unit 110 receives electrocardiogram data.

The ECG based on 12 electrodes is measured for about 10 seconds, and the ECG signal measured for 10 seconds is stored as 5000 numbers for each electrode. It is output as an electrocardiogram picture.

At this time, the 60,000 numbers (5000*12 electrodes) that generate the electrocardiogram are called back data (or source data), and in the embodiment of the present invention, this back data can be input as electrocardiogram data.

The embodiment of the present invention has been described as an example of measuring the ECG signal in the 12-guide method, but in addition to this, the ECG signal is measured using various electrode combinations such as the 6-guide method and the single-guide method such as a wrist watch type or patch. may be It is preferable that the measurement time is also increased or decreased according to the signal to be obtained.

In this case, the input unit 110 may further receive patient information including age and gender.

In addition, the characteristic extraction unit 120 extracts n characteristics for diagnosing a heart disease by analyzing the received electrocardiogram data.

In this case, the characteristic extraction unit 120 may extract characteristics included in the electrocardiogram data using an auto-encoder, but is not limited thereto.

Here, the autoencoder method is a machine learning method using unsupervised learning. To explain the ECG signal as an example, when 12*5000 ECG data is input, it is output to have the same value as the input value. That is, the same electrocardiogram is input and output. Instead, in the middle layer, the middle layer is made with a much smaller number of layers, and machine learning is performed so that the ECG data identical to the input ECG data is output. This method summarizes large data into a few numbers and returns it back to the original data, so that the entire ECG data is summarized into a few characteristic values, and then the original ECG data is reconstructed using the characteristic values again. .

In addition, the decision model generating unit 130 generates identification codes for n characteristics for each diagnosis name of heart disease, and generates n judgment models for determining each characteristic by learning the generated identification codes.

In detail, two identification codes divided by binary values are generated for each characteristic to have a value corresponding to each diagnosis name, and the same identification code is clustered to classify each characteristic into two groups, deep learning By learning the binary values of the divided groups using an algorithm, a judgment model that distinguishes each characteristic is created.

Here, the deep learning algorithm is based on a neural network model such as an artificial neural network (ANN) including a recurrent neural network (RNN) algorithm or a convolutional neural network (CNN) algorithm, and here Various machine learning methods such as random forest, support vector machine (SVM), linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), K-nearest neighbor (K-NN), and K-means can be applied together. .

Also, the decision model generating unit 130 may generate a decision model by further learning patient information of the electrocardiogram data received through the input unit 110 .

In addition, the diagnostic model generator 140 generates a diagnostic model for deriving a diagnosis name of a heart disease by learning values determined from the judgment model, respectively.

Then, the controller 150 outputs the diagnosis result for the electrocardiogram data to be diagnosed using the judgment model and the diagnosis model.

In this case, the controller 150 may filter the noise of the electrocardiogram data to be diagnosed, or may input the signal data into the judgment model without noise filtering.

Also, the controller 150 may output a diagnosis name and a diagnosis reason for the electrocardiogram data to be diagnosed.

Finally, the storage unit 160 stores the diagnosis result diagnosed by the control unit 150 and the corresponding electrocardiogram data.

Accordingly, the decision model generator 130 may further learn the ECG data and the diagnosis result stored in the storage 160 using a deep learning algorithm.

Hereinafter, a method for diagnosing a heart disease using an electrocardiogram based on deep learning according to an embodiment of the present invention will be described with reference to FIGS. 3 to 6 .

3 is a flowchart illustrating an operation flow of a method for diagnosing a heart disease using a deep learning-based electrocardiogram according to an embodiment of the present invention. A detailed operation of the present invention will be described with reference to this.

According to an embodiment of the present invention, first, the input unit 110 of the heart disease diagnosis apparatus 100 receives electrocardiogram data (S100).

In this case, the input unit 110 may further receive patient information including age and gender.

Then, the characteristic extraction unit 120 analyzes the electrocardiogram data received in step S100 to extract n characteristics for diagnosing a heart disease (S200).

Next, the decision model generating unit 130 generates identification codes for n characteristics for each diagnosis name of heart disease (S300).

In this case, the decision model generating unit 130 may generate two identification codes divided into binary values for each characteristic so as to have a value corresponding to each diagnosis name.

FIG. 4 is a diagram illustrating a process of generating an identification code for each characteristic for each diagnosis name in step S300 of FIG. 3 .

4, the judgment model generating unit 130 analyzes the electrocardiogram data received in step S100 to generate and label the diagnosis name of heart disease and an identification code corresponding to each characteristic extracted by the characteristic extraction unit 120. desirable. In this case, the identification codes are indicated by A to Z for convenience of understanding, but the present invention is not limited thereto.

Then, the decision model generating unit 130 generates n judgment models for determining each characteristic by learning the identification code generated in step S300 (S400).

At this time, the decision model generating unit 130 classifies the same identification code into two groups for each characteristic by clustering the same identification code, and a decision model for classifying each characteristic by learning the binary values of the two groups divided using a deep learning algorithm. can create

FIG. 5 is a diagram illustrating a decision model generated in step S400 of FIG. 3 .

As shown in FIG. 5 , to describe characteristic 1 as an example, the identification code of characteristic 1 is divided into two groups, the group in which the identification code is A, and the identification code of characteristic 1 is B, and accordingly, the binary value of each identification code is deep learning A judgment model for characteristic 1 that distinguishes characteristic 1 by learning it using an algorithm is generated. In this case, the generated judgment model is not a model for diagnosing a heart disease, but a model for judging the characteristic 1.

Next, the diagnostic model generating unit 140 generates a diagnostic model for deriving a diagnosis name of a heart disease by learning values determined from the judgment model generated in step S400 ( S500 ).

Finally, the controller 150 outputs a diagnosis result for the electrocardiogram data to be diagnosed using the judgment model generated in step S400 and the diagnostic model generated in step S500 ( S600 ).

In step S600, the controller 150 may output a diagnosis name and diagnosis reason for the electrocardiogram data to be diagnosed.

For example, when the diagnosis result for the electrocardiogram data is arrhythmia due to atrial fibrillation, the control unit 150 does not simply read and output only the diagnosis name, but says "Atrial fibrillation, the reason has irregular characteristics, and Pwave is not visible. and because the pulse rate is between 50 and 100."

Accordingly, it is possible to satisfy the needs of a diagnosing doctor or patient, thereby improving the reliability of the device.

In this case, the controller 150 may store electrocardiogram data to be diagnosed and a diagnosis result for the data in the storage 160 .

The decision model generator 130 may further learn the electrocardiogram data stored in the storage 160 and the diagnosis result of the electrocardiogram data by using a deep learning algorithm.

That is, the determination model generator 130 may improve the accuracy of the determination model by outputting the determination result and simultaneously learning the corresponding electrocardiogram data.

Also, the decision model generating unit 130 may further learn patient information of the electrocardiogram data.

6 is a diagram illustrating a process of outputting a heart disease diagnosis result according to an embodiment of the present invention.

As shown in FIG. 6 , the controller 150 inputs electrocardiogram data to be diagnosed into the judgment model generated by the judgment model generator 130 , and each judgment value determined by each judgment model is converted into the diagnosis model generator 140 . ), a diagnosis result for the corresponding electrocardiogram data can be derived using the output diagnosis result.

As described above, the apparatus and method for diagnosing heart disease using an electrocardiogram based on deep learning according to an embodiment of the present invention learns electrocardiogram data by characteristics using a deep learning algorithm, and uses the learned model to diagnose heart disease Diagnosis accuracy is improved by deriving the

In addition, according to an embodiment of the present invention, the diagnostic results are not classified binaryly, such as normal and abnormal (arrhythmia), but multivariate, such as normal, arrhythmia type 1, arrhythmia type 2, arrhythmia type 3, ??N. Therefore, accurate diagnostic results can be presented.

In addition, according to an embodiment of the present invention, an objective and accurate diagnosis is possible because the learning model is learned by a deep learning method that automatically extracts features by judging only by data, and it is useful for diagnosing various diseases by learning various biosignals can have an effect.

Although the present invention has been described with reference to the embodiment shown in the drawings, this is merely exemplary, and it will be understood by those skilled in the art that various modifications and equivalent other embodiments are possible therefrom. will be. Therefore, the true technical protection scope of the present invention should be determined by the technical spirit of the following claims.

100: heart disease diagnosis device 110: input unit
120: characteristic extraction unit 130: judgment model generation unit
140: diagnostic model generation unit 150: control unit
160: storage

Claims (10)

In a heart disease diagnosis device using deep learning-based electrocardiogram,
an input unit for receiving electrocardiogram data;
a characteristic extracting unit that analyzes the electrocardiogram data and extracts n characteristics for diagnosing the heart disease;
a judgment model generation unit for generating identification codes for the n characteristics for each diagnosis name of the heart disease, and generating n judgment models for judging each characteristic by learning the generated identification codes;
a diagnostic model generator for generating a diagnostic model for deriving a diagnosis name of the heart disease by learning values determined from the judgment model; and
and a controller outputting a diagnosis result for the electrocardiogram data to be diagnosed using the judgment model and the diagnosis model. According to claim 1,
The judgment model generation unit,
To have a value corresponding to each diagnosis name, generate two identification codes that are divided into binary values for each characteristic, cluster the same identification code into two groups for each characteristic, and use a deep learning algorithm to learn the binary values of the two groups and generate the judgment model for classifying each characteristic. According to claim 1,
The control unit is
storing the electrocardiogram data to be diagnosed and a diagnosis result for the data in a storage unit;
The judgment model generation unit,
A heart disease diagnosis apparatus for learning the stored electrocardiogram data and the diagnosis result using a deep learning algorithm. According to claim 1,
The input unit,
receiving more patient information including age and gender;
The judgment model generation unit,
A heart disease diagnosis apparatus for generating the judgment model by further learning patient information of the electrocardiogram data. According to claim 1,
The control unit is
A heart disease diagnosis apparatus for outputting a diagnosis name and a diagnosis reason for the electrocardiogram data to be diagnosed. In a method for diagnosing heart disease using deep learning-based electrocardiogram,
The heart disease diagnosis apparatus may include: receiving electrocardiogram data;
extracting n characteristics for diagnosing the heart disease by analyzing the electrocardiogram data;
generating each identification code for the n characteristics for each diagnosis name of the heart disease;
generating n judgment models for judging each characteristic by learning the generated identification code;
generating a diagnostic model for deriving a diagnosis name of the heart disease by learning values determined from the judgment model; and
and outputting a diagnosis result for the electrocardiogram data to be diagnosed using the judgment model and the diagnosis model. 7. The method of claim 6,
The step of generating the judgment model comprises:
To have a value corresponding to each diagnosis name, generate two identification codes that are divided into binary values for each characteristic, cluster the same identification code into two groups for each characteristic, and use a deep learning algorithm A heart disease diagnosis method for generating the judgment model for classifying each characteristic by learning the binary values of the two groups. 7. The method of claim 6,
Further comprising the step of storing the electrocardiogram data to be diagnosed and a diagnosis result for the data,
The step of generating the judgment model comprises:
A heart disease diagnosis method for learning the stored electrocardiogram data and the diagnosis result using a deep learning algorithm. 7. The method of claim 6,
The step of receiving the input is
receiving more patient information including age and gender;
The step of generating the judgment model comprises:
A heart disease diagnosis method for generating the judgment model by further learning patient information from electrocardiogram data. 7. The method of claim 6,
The step of outputting the diagnosis result comprises:
A heart disease diagnosis method for outputting a diagnosis name and a diagnosis reason for the electrocardiogram data to be diagnosed.

Images