KR20230169577A - Severity detection system of heart failure using single-lead electrocardiogram and method of detecting severity of heart failure using single-lead electrocardiogram - Google Patents

Abstract

The heart failure severity prediction system includes an ECG signal input unit that receives a single ECG signal and generates input data based on the single ECG signal, and pre-learns the cardiac ejection fraction corresponding to the single ECG signal based on the ECG data set. A cardiac ejection fraction learning model that calculates the cardiac ejection fraction based on the spatial and temporal characteristics of input data, and heart failure severity based on the cardiac ejection fraction into Normal, Mild, Moderate, and a heart failure severity determination unit that determines at least one of severe cases.

Description

Heart failure severity prediction system based on a single electrocardiogram and a method for predicting the severity of heart failure based on a single electrocardiogram

The present invention relates to a heart failure severity prediction system, and more specifically, to classify heart failure severity into one of Normal, Mild, Moderate, and Severe from a single electrocardiogram signal through deep learning model learning. It relates to forecasting systems and methods.

Heart failure refers to a condition in which heart function deteriorates due to excessive activity on the heart, preventing it from operating normally. Based on the ejection fraction, heart failure can be classified into heart failure with reduced left ventricular ejection fraction (HFrEF), heart failure with borderline ejection fraction (HFmrEF), and heart failure with preserved ejection fraction (HFpEF). As the world becomes an aging society, the incidence of heart failure is increasing every year. Once heart function declines due to heart failure, complete cure is difficult, so it is important to detect heart failure early and manage it appropriately.

In heart failure diagnosis clinical trials, heart failure is diagnosed based on the H2FPEF and HFA-PEFF algorithms recommended by the European Society of Cardiology.

First, the H2FPEF algorithm evaluates heart failure using four basic factors, including age, body mass index, atrial fibrillation, and hypertension, and two echocardiography items. Additionally, the HFA-PEFF score is evaluated within three areas: functional, morphological, and natural diuretic peptides. However, these methods have errors based on gender/race and differences between algorithm scores and actual results, so new methods are being introduced through many studies.

Recently, many machine learning-based studies have been conducted to detect cardiac ejection fraction. For example, these studies can be classified into clinical information-based studies, echocardiography-based studies, and electrocardiogram-based studies.

In the existing clinical information-based heart failure diagnosis research, up to five data, including patient basic information, symptoms, biometric information, history, and clinical examination, were applied to a machine learning algorithm for learning and evaluation. However, existing clinical information-based heart failure diagnosis studies did not detect heart failure severity. In particular, to date, no study has been introduced to detect the severity of heart failure using only a single electrocardiogram signal.

Korean Patent No. 10-2322234 “Electrocardiogram visualization method and device using deep learning”

One purpose of the present invention is to provide a heart failure severity prediction system that predicts heart failure severity with high accuracy based only on a single electrocardiogram signal through deep learning model learning.

Another object of the present invention is to provide a heart failure severity prediction system that provides quick diagnosis results by predicting the severity of heart failure as one of Normal, Mild, Moderate, and Severe.

However, the problem to be solved by the present invention is not limited to the above-mentioned problem, and may be expanded in various ways without departing from the spirit and scope of the present invention.

In order to achieve an object of the present invention, the heart failure severity prediction system according to embodiments of the present invention includes an ECG signal input unit that receives a single ECG signal and generates input data based on the single ECG signal, and an ECG data set. A cardiac ejection fraction learning model that pre-learns the cardiac ejection fraction corresponding to the single electrocardiogram signal as a basis and calculates the cardiac ejection fraction based on the spatial and temporal characteristics of the input data, and based on the cardiac ejection fraction It may include a heart failure severity determination unit that determines the heart failure severity to at least one of Normal, Mild, Severe, and Severe.

In one embodiment, the ejection fraction learning model may sequentially use a CNN model for extracting the spatial characteristics of the single ECG signal and an RNN model for extracting the temporal characteristics of the single ECG signal.

In one embodiment, the CNN model included in the cardiac ejection fraction learning model may be a 1-D CNN model that learns morphological characteristics of a QRS complex from the single ECG signal.

In one embodiment, the RNN model included in the cardiac ejection fraction learning model may be a GRU (Gated Recurrent Unit) model that learns changes in the single ECG signal over time.

In one embodiment, the single electrocardiogram signal is measured using an electrocardiogram measuring device and may be a primary parameter for which no separate data processing or data conversion is performed.

In one embodiment, the heart failure severity determination unit determines the severity of heart failure: normal when the ejection fraction is 30% to 35%, mild when the ejection fraction is 35% to 40%, and severe when the ejection fraction is 40%. If the ejection fraction is between 50% and 50%, it can be determined as severe, and if the ejection fraction is greater than 50%, it can be determined as perjury.

In order to achieve another object of the present invention, the method for predicting the severity of heart failure according to embodiments of the present invention includes receiving a single electrocardiogram signal, and pre-learning the ejection fraction corresponding to the single electrocardiogram signal based on the electrocardiogram data set. generating input data based on the single electrocardiogram signal, calculating the cardiac ejection fraction based on spatial and temporal characteristics of the input data, and determining the severity of heart failure based on the ejection fraction. It may include determining at least one of (Normal), Mild (Mild), Severe (Moderate), and Perjury (Severe).

In one embodiment, the step of calculating the cardiac ejection fraction is performed by sequentially using a CNN model for extracting the spatial characteristics of the single ECG signal and an RNN model for extracting the temporal characteristics of the single ECG signal. can be calculated.

According to the heart failure severity prediction system and heart failure severity prediction method according to embodiments of the present invention, heart failure severity can be predicted with high accuracy based only on a single electrocardiogram signal through deep learning model learning.

In addition, according to the heart failure severity prediction system and heart failure severity prediction method according to embodiments of the present invention, the heart failure severity is predicted to be one of Normal, Mild, Moderate, and Severe and quickly Diagnosis results can be provided.

However, the effects of the present invention are not limited to the effects described above, and may be expanded in various ways without departing from the spirit and scope of the present invention.

1 is a conceptual diagram showing a heart failure severity prediction system according to embodiments of the present invention.
FIG. 2 is a block diagram showing the configuration of the heart failure severity prediction system of FIG. 1.
FIG. 3 is a flowchart showing the operation of the heart failure severity prediction system of FIG. 1.
Figure 4 is a diagram showing an example of a single ECG signal and ECG data set input to the ECG signal input unit.
Figure 5 is a diagram showing an example of a cardiac ejection fraction learning model.
Figure 6 is a diagram showing the CNN model of the cardiac ejection fraction learning model.
Figure 7 is a diagram showing the RNN model of the cardiac ejection fraction learning model.
Figure 8 is a diagram showing performance evaluation indicators and prediction results of a heart failure severity prediction system.

Specific structural or functional descriptions of the embodiments according to the concept of the present invention disclosed in this specification are merely illustrative for the purpose of explaining the embodiments according to the concept of the present invention. They may be implemented in various forms and are not limited to the embodiments described herein.

Since the embodiments according to the concept of the present invention can make various changes and have various forms, the embodiments will be illustrated in the drawings and described in detail in this specification. However, this is not intended to limit the embodiments according to the concept of the present invention to specific disclosed forms, and includes changes, equivalents, or substitutes included in the spirit and technical scope of the present invention.

Terms such as first or second may be used to describe various components, but the components should not be limited by the terms. The terms are used solely for the purpose of distinguishing one component from another, for example, a first component may be named a second component, and similar Thus, the second component may also be referred to as the first component.

When a component is said to be "connected" or "connected" to another component, it is understood that it may be directly connected to or connected to the other component, but that other components may exist in between. It should be. On the other hand, when it is mentioned that a component is “directly connected” or “directly connected” to another component, it should be understood that there are no other components in between. Expressions that describe the relationship between components, such as “between”, “immediately between” or “directly adjacent to”, should be interpreted similarly.

The terminology used herein is only used to describe specific embodiments and is not intended to limit the invention. Singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, terms such as "include" or "have" are intended to designate the presence of a described feature, number, step, operation, component, part, or combination thereof, and one or more other features or numbers, It should be understood that this does not exclude in advance the possibility of the presence or addition of steps, operations, components, parts, or combinations thereof.

Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the technical field to which the present invention pertains. Terms as defined in commonly used dictionaries should be interpreted as having meanings consistent with the meanings they have in the context of the related technology, and unless clearly defined in this specification, should not be interpreted in an idealized or overly formal sense. No.

Hereinafter, embodiments will be described in detail with reference to the attached drawings. However, the scope of the patent application is not limited or limited by these examples. The same reference numerals in each drawing indicate the same members.

1 is a conceptual diagram showing a heart failure severity prediction system 10 according to embodiments of the present invention.

Referring to FIG. 1, the heart failure severity prediction system 10 of the present invention predicts heart failure severity based on only a single electrocardiogram signal through deep learning model learning into Normal, Mild, Moderate, and False ( Severe) can be predicted as one of the following.

That is, the heart failure severity prediction system 10 of the present invention can provide quick diagnosis results by predicting heart failure severity as one of normal, mild, severe, and false.

FIG. 2 is a block diagram showing the configuration of the heart failure severity prediction system 10 of FIG. 1 , and FIG. 3 is a flowchart showing the operation of the heart failure severity prediction system 10 of FIG. 1 .

Referring to FIG. 2 , the heart failure severity prediction system 10 may include an electrocardiogram signal input unit 100, a cardiac ejection fraction learning model 200, and a heart failure severity determination unit 300.

For example, the heart failure severity prediction system 10 may include an ECG signal input unit 100 that receives a single ECG signal and generates input data based on the single ECG signal.

For example, the heart failure severity prediction system 10 pre-learns the cardiac ejection fraction corresponding to the single electrocardiogram signal based on an electrocardiogram data set, and calculates the ejection fraction based on the spatial and temporal characteristics of the input data. It may include a learning model 200 for calculating cardiac ejection fraction.

The ejection fraction learning model 200 may sequentially use a CNN model for extracting the spatial characteristics of the single ECG signal and an RNN model for extracting the temporal characteristics of the single ECG signal.

For example, the heart failure severity prediction system 10 determines the heart failure severity to be at least one of Normal, Mild, Moderate, and Severe based on the cardiac ejection fraction. It may include unit 300.

As shown in FIG. 3, the heart failure severity prediction system 10 receives a single electrocardiogram signal as input (S100), pre-learns the cardiac ejection fraction corresponding to the single electrocardiogram signal based on the electrocardiogram data set (S200), and calculates the single electrocardiogram signal. Generate input data based on (S300), calculate the ejection fraction (S400) based on the spatial and temporal characteristics of the input data, and classify the severity of heart failure as normal, mild, severe, and false based on the ejection fraction. You can decide (S500) on at least one of the following.

In one embodiment, the heart failure severity prediction system 10 receives a single electrocardiogram signal (S100), pre-learns the cardiac ejection fraction corresponding to the single electrocardiogram signal based on the electrocardiogram data set (S200), and prepares the single electrocardiogram signal. Based on this, input data can be generated (S300).

FIG. 4 is a diagram illustrating an example of a single ECG signal and ECG data set input to the ECG signal input unit 100.

As shown in FIG. 4(a), the ECG signal input unit 100 can receive a single ECG signal measured using an ECG measurement device.

For example, the ECG signal input unit 100 may receive a single ECG signal and generate input data based on the single ECG signal.

The single ECG signal may be a primary parameter for which no separate data processing or data conversion, such as image conversion, has been performed.

For example, the single electrocardiogram signal may consist of a 12-lead standard electrocardiogram (12-lead ECG).

The heart failure severity prediction system 10 can pre-learn the cardiac ejection fraction corresponding to a single electrocardiogram signal based on the electrocardiogram data set.

As shown in Figure 4(b), the ECG data set includes a single ECG signal from a random patient (e.g., 4800 people) and data in which heart failure severity (Mild, Moderate, Severe) corresponding to the single ECG signal of the patient is mapped. It can be included.

For example, the ECG data set may include a training data set (Train set) and a test data set (Test set).

Specifically, the cardiac ejection fraction learning model 200 can learn the cardiac ejection fraction for each single ECG signal of the ECG data set. Additionally, the cardiac ejection fraction learning model 200 can learn the severity of heart failure for each single ECG signal in the ECG data set.

In one embodiment, the heart failure severity prediction system 10 may calculate the cardiac ejection fraction based on the spatial and temporal characteristics of the input data (S400).

FIG. 5 is a diagram illustrating an example of the cardiac ejection fraction learning model 200.

The cardiac ejection fraction learning model 200 may include a plurality of learning layers to calculate the cardiac ejection fraction from a single electrocardiogram signal.

For example, the cardiac ejection fraction learning model 200 may include Batch-Norm to normalize input data.

Additionally, the cardiac ejection fraction learning model 200 may include a learning model with a multi-layer structure that learns input data using a one-dimensional convolution method and uses the ReLU function as an activation function.

Additionally, the cardiac ejection fraction learning model 200 may utilize a fully-connected layer and a Softmax function to perform multi-layer classification.

Figure 6 is a diagram showing a CNN model of the cardiac ejection fraction learning model 200.

The cardiac ejection fraction learning model 200 may use a CNN model that extracts the spatial characteristics of the single ECG signal.

The CNN model included in the cardiac ejection fraction learning model 200 may be a 1-D CNN model (200(a)) that learns morphological characteristics of the QRS complex (QRS complex) from the single electrocardiogram signal. there is.

As shown in Figure 6, the CNN model may include an input unit, a feature extraction unit, a data classification unit, and an output unit.

The input unit of the CNN model may receive input data. The input unit of the CNN model may normalize input data by performing batch normalization on the single ECG signal.

The feature extraction unit of the CNN model can extract features of the input data through a convolution layer, ReLU activation, and pooling layer.

The convolution layer may be a 1-D convolution layer.

Additionally, the feature extraction unit of the CNN model can use dropout for data optimization.

For example, the feature extraction unit of a CNN model may be composed of three layers.

The data classification unit of the CNN model can utilize a fully-connected layer and the Softmax function to perform multi-layer classification.

The output unit of the CNN model may output whether input data is normal or abnormal based on data learned from the feature extraction unit and the data classification unit.

For example, the output unit of the CNN model may output cardiac ejection fraction and whether the ejection fraction is normal or abnormal based on a single electrocardiogram signal.

Figure 7 is a diagram showing the RNN model of the cardiac ejection fraction learning model 200.

The cardiac ejection fraction learning model 200 may use an RNN model that extracts the temporal characteristics of the single ECG signal.

In one embodiment, the RNN model included in the cardiac ejection fraction learning model 200 may be a GRU (Gated Recurrent Unit) model that learns changes in the single ECG signal over time.

As shown in FIG. 7, the GRU model 200(b) may be characterized by a fast data operation processing speed due to its simple logical structure.

In particular, the GRU model has minimal parameters, so it can have optimal efficiency when large amounts of data operations are required.

In this way, the cardiac ejection fraction learning model 200 can calculate the cardiac ejection fraction reflecting both the spatial and temporal characteristics of the input data by sequentially using the CNN model and the RNN model.

In one embodiment, the heart failure severity prediction system 10 may determine the heart failure severity to be at least one of normal, mild, severe, and false based on the cardiac ejection fraction (S500).

Specifically, the heart failure severity determination unit 300 may determine the heart failure severity according to the interval of cardiac ejection fraction.

For example, the heart failure severity determination unit 300 may determine the severity of heart failure as normal when the cardiac ejection fraction is 30% to 35% (30 < EF ≤ 35).

For example, the heart failure severity determination unit 300 may determine the severity of heart failure as mild when the cardiac ejection fraction is 35% to 40% (35 < EF ≤ 40).

For example, the heart failure severity determination unit 300 may determine the severity of heart failure as severe when the cardiac ejection fraction is 40% to 50% (40 < EF ≤ 50).

For example, the heart failure severity determination unit 300 may determine the severity of heart failure as false if the cardiac ejection fraction is greater than 50% (50 < EF).

However, the section of cardiac ejection fraction that determines the severity of heart failure is only an example and does not limit the method of determining the severity of heart failure of the present invention.

Figure 8 is a diagram showing performance evaluation indicators and prediction results of the heart failure severity prediction system 10.

Referring to FIG. 8(a), the performance evaluation indicators of the heart failure severity prediction system 10 were measured based on precision, recall, and F1-score.

For example, precision, recall, and F1-score for each heart failure severity were evaluated by dividing them into a training data set (Train set) and a test data set (Test set).

The precision may be an indicator indicating the extent to which the actual value is True among the predicted values being True. The recall rate may be an indicator indicating the extent to which the predicted value is True among the actual values being True. In other words, precision and recall can have a trade-off.

The F1 score may represent the harmonic mean of precision and recall. In other words, the F1 score can be an indicator to evaluate whether precision and recall are appropriately balanced.

As shown in Figure 8(a), the F1 score for the training data set (Train set) is 0.84 for mild, 0.83 for severe, and 0.89 for perjury.

In addition, it can be seen that the F1 score for the test data set is 0.65 for mild cases, 0.76 for severe cases, and 0.78 for perjury.

That is, it can be confirmed that the heart failure severity prediction system 10 of the present invention outputs a heart failure severity prediction result with high accuracy from a single electrocardiogram signal.

Referring to Figure 8(b), the ROC curve and AUC obtained from a plurality of experimental data (A, B, C, D) can be confirmed.

In the experimental data (A, B, C, D), ROC curve of cloass 0 is for Mild, ROC curve of cloass 1 is for Moderate, and ROC curve of cloass 2 is for Severe. It can indicate a case.

Comparing each experimental data (A, B, C, D), it can be confirmed that the heart failure severity prediction system 10 stably outputs a constant ROC curve and AUC for a single ECG signal.

For example, as shown in D of FIG. 8(b), the heart failure severity prediction system 10 has an AUC value of 74% for Mild, an AUC value of 53% for Moderate, and Severe. ) It can be seen that the AUC value for ) is 72%.

As such, according to the heart failure severity prediction system and heart failure severity prediction method of the present invention, heart failure severity can be predicted with high accuracy based on only a single electrocardiogram signal through deep learning model learning.

In addition, according to the heart failure severity prediction system and heart failure severity prediction method of the present invention, heart failure severity can be predicted as one of Normal, Mild, Moderate, and Severe to provide quick diagnostic results. You can.

However, since this has been described above, redundant description thereof will be omitted.

The device described above may be implemented with hardware components, software components, and/or a combination of hardware components and software components. For example, devices and components described in embodiments may include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable array (FPA), It may be implemented using one or more general-purpose or special-purpose computers, such as a programmable logic unit (PLU), microprocessor, or any other device capable of executing and responding to instructions. A processing device may execute an operating system (OS) and one or more software applications that run on the operating system. Additionally, a processing device may access, store, manipulate, process, and generate data in response to the execution of software. For ease of understanding, a single processing device may be described as being used; however, those skilled in the art will understand that a processing device includes multiple processing elements and/or multiple types of processing elements. It can be seen that it may include. For example, a processing device may include a plurality of processors or one processor and one controller. Additionally, other processing configurations, such as parallel processors, are possible.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc., singly or in combination. Program instructions recorded on the medium may be specially designed and configured for the embodiment or may be known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -Includes optical media (magneto-optical media) and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, etc. Examples of program instructions include machine language code, such as that produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

Although the embodiments have been described with limited drawings as described above, various modifications and variations can be made by those skilled in the art from the above description. For example, the described techniques are performed in a different order than the described method, and/or components of the described system, structure, device, circuit, etc. are combined or combined in a different form than the described method, or other components are used. Alternatively, appropriate results may be achieved even if substituted or substituted by an equivalent. Therefore, other implementations, other embodiments, and equivalents of the claims also fall within the scope of the claims described below.

10: Heart failure severity prediction system
100: ECG signal input unit
200: Cardiac ejection fraction learning model
300: Heart failure severity judgment unit

Claims (8)

an electrocardiogram signal input unit that receives a single electrocardiogram signal and generates input data based on the single electrocardiogram signal;
A cardiac ejection fraction learning model that pre-learns the cardiac ejection fraction corresponding to the single electrocardiogram signal based on an electrocardiogram data set and calculates the cardiac ejection fraction based on spatial and temporal characteristics of the input data; and
Comprising a heart failure severity determination unit that determines the severity of heart failure as at least one of Normal, Mild, Moderate, and Severe based on the cardiac ejection fraction,
Heart failure severity prediction system. According to paragraph 1,
The cardiac ejection fraction learning model is,
Characterized by sequentially using a CNN model for extracting the spatial characteristics of the single ECG signal and a RNN model for extracting the temporal characteristics of the single ECG signal,
Heart failure severity prediction system. According to paragraph 2,
The CNN model included in the cardiac ejection fraction learning model is,
Characterized in that it is a 1-D CNN model that learns morphological characteristics of the QRS complex (QRS complex) from the single electrocardiogram signal.
Heart failure severity prediction system. According to paragraph 2,
The RNN model included in the cardiac ejection fraction learning model is,
Characterized in that it is a GRU (Gated Recurrent Unit) model that learns changes in the single electrocardiogram signal over time.
Heart failure severity prediction system. According to paragraph 1,
The single electrocardiogram signal is,
Characterized in that it is a primary parameter measured using an electrocardiogram measurement device and no separate data processing or data conversion is performed.
Heart failure severity prediction system. According to paragraph 1,
The heart failure severity determination unit,
The heart failure severity is normal when the ejection fraction is 30% to 35%, mild when the ejection fraction is 35% to 40%, severe when the ejection fraction is 40% to 50%, and heart failure. Characterized by determining perjury if the probability of occurrence is greater than 50%,
Heart failure severity prediction system. Receiving a single electrocardiogram signal;
Pre-learning the cardiac ejection fraction corresponding to the single electrocardiogram signal based on an electrocardiogram data set;
generating input data based on the single electrocardiogram signal;
calculating the cardiac ejection fraction based on spatial and temporal characteristics of the input data; and
Comprising the step of determining heart failure severity as at least one of Normal, Mild, Moderate, and Severe based on the cardiac ejection fraction,
How to predict heart failure severity. In clause 7,
The step of calculating the cardiac ejection fraction is,
Characterized in calculating the cardiac ejection fraction by sequentially using a CNN model for extracting the spatial characteristics of the single ECG signal and an RNN model for extracting the temporal characteristics of the single ECG signal.
How to predict heart failure severity.