KR20230099195A - An apparatus and method for predicting heart disease based on deep learning models using ECG data and body information - Google Patents

Abstract

An apparatus for predicting the onset of heart disease based on a deep learning model using electrocardiogram data and body information according to the present invention includes an acquisition unit acquiring first electrocardiogram data and first body information of a patient; and augmenting the first electrocardiogram data to generate second electrocardiogram data, pre-processing the first body information to generate second body information, and using a first deep learning model to generate second electrocardiogram data from the second electrocardiogram data to the patient. generating first incidence probability information of heart disease, and generating second incidence probability information of heart disease in the patient from the second electrocardiogram data and the second body information using a second deep learning model; , a processor generating integrated onset probability information by integrating the first onset probability information and the second onset probability information.

Description

An apparatus and method for predicting heart disease based on deep learning models using ECG data and body information}

The present invention relates to an apparatus and method for predicting the onset of heart disease based on a deep learning model using electrocardiogram data and body information, and more particularly, to a first onset generated from a first deep learning model composed of a convolutional neural network model to which a residual block is applied. Heart disease based on a deep learning model using electrocardiogram data and body information that calculates an integrated probability of heart disease by integrating the probability information and the second probability information generated from the second deep learning model constructed using the deep learning model configuration automation module. It relates to an onset prediction device and method.

Arrhythmia refers to a state in which the heartbeat becomes irregular because normal and regular contraction is not continued when there is an abnormality in electrical signal generation or transmission in the heart, or when an abnormal electrical signal occurs.

Among arrhythmias, atrial fibrillation is the most common arrhythmia, and its prevalence is increasing worldwide as the aging population accelerates, and it is also increasing in Korea.

Since such arrhythmia increases the occurrence of cardiovascular diseases such as stroke and heart failure, treatment and management through early diagnosis are very important.

To this end, it is necessary to diagnose cardiovascular disease at an early stage based on an electrocardiogram. Electrocardiogram is a signal that records the active current generated in the myocardium according to the heartbeat with an ammeter. It is measured by attaching a lead to the patient. is also widely used.

Specifically, the electrocardiogram used for diagnosis is a 12-lead electrocardiogram, which is obtained using 12 leads. Leads 1, 2, 3, aVR, aVL, and avF attached to the extremities of the limbs are called limb induction, and Leads 1, 2, and 3 are called standard induction. Looking at the waveform of an electrocardiogram, it consists of P wave, QRS complex wave, and T wave. P wave represents atrial depolarization process, QRS complex wave represents ventricular depolarization process, and T wave represents ventricular repolarization.

Recently, it has become possible to monitor such an electrocardiogram with a wearable device, and a technology for extracting features based on the electrocardiogram measured with the wearable device is being developed.

However, since the information of the electrocardiogram measured by the wearable device is poor and the accuracy is low, when the electrocardiogram measured by the wearable device is used as it is, there is a problem in that the accuracy of monitoring and feature extraction is reduced.

Korean Patent Publication No. 10-2020-0041697

The present invention provides first onset probability information generated from a first deep learning model composed of a convolutional neural network model to which a residual block is applied and second probability information generated from a second deep learning model constructed using a deep learning model configuration automation module. It is intended to provide an apparatus and method for predicting the onset of heart disease based on a deep learning model using electrocardiogram data and body information capable of integrating and calculating the combined onset probability of heart disease.

The objects of the present invention are not limited to the above-mentioned objects, and other objects and advantages of the present invention not mentioned above can be understood by the following description and will be more clearly understood by the embodiments of the present invention. It will also be readily apparent that the objects and advantages of the present invention may be realized by means of the instrumentalities and combinations indicated in the claims.

An apparatus for predicting the onset of heart disease based on a deep learning model using electrocardiogram data and body information according to the present invention includes an acquisition unit acquiring first electrocardiogram data and first body information of a patient; and augmenting the first electrocardiogram data to generate second electrocardiogram data, pre-processing the first body information to generate second body information, and using a first deep learning model to generate second electrocardiogram data from the second electrocardiogram data to the patient. generating first incidence probability information of heart disease, and generating second incidence probability information of heart disease in the patient from the second electrocardiogram data and the second body information using a second deep learning model; , a processor generating integrated onset probability information by integrating the first onset probability information and the second onset probability information.

Preferably, the processor may include a data enhancer configured to generate the second electrocardiogram data by augmenting the first electrocardiogram data so that the number of measurement leads of the electrocardiogram data is a preset number.

Preferably, the processor calculates a body index based on the first body information, and the first body information and the body index are input to the second deep learning model. may include a body information processing unit that preprocesses and generates second body information.

Preferably, the processor may include a first predictor configured to generate the first onset probability information from the second electrocardiogram data by using the first deep learning model composed of a convolutional neural network model to which a residual block is applied. .

Preferably, the first predictor may generate influencing factor information indicating a factor influencing generation of the first onset probability information by using an attention module.

Preferably, the processor selects an arbitrary detailed deep learning model corresponding to the second electrocardiogram data and the second body information, and configures the second deep learning model with the selected detailed deep learning model. A second prediction unit configured to construct the second deep learning model using modules and to generate second onset probability information from the second electrocardiogram data and the second body information using the second deep learning model; can do.

Preferably, the processor checks heart disease type information that maximizes the combined incidence probability information, and if the combined incidence probability information corresponding to the identified heart disease type information exceeds a preset reference probability, the processor determines the identified heart disease type information. It may include; a prediction result integrator for generating heart disease type information into incidence prediction heart disease type information in which onset is predicted.

A heart disease prediction method based on a deep learning model using electrocardiogram data and body information according to the present invention includes the steps of acquiring, by an acquisition unit, first electrocardiogram data and first body information of a patient; generating, by a processor, second electrocardiogram data by augmenting the first electrocardiogram data; generating, by the processor, second body information by pre-processing the first body information; generating, by the processor, first onset probability information of a heart disease occurring in the patient from the second electrocardiogram data by using a first deep learning model; generating, by the processor, second onset probability information for the onset of a heart disease in the patient from the second electrocardiogram data and the second body information by using a second deep learning model; and generating, by the processor, final outbreak probability information by integrating the first outbreak probability information and the second outbreak probability information.

Preferably, the method according to the present invention further comprises generating the second electrocardiogram data by augmenting the first electrocardiogram data so that the number of measurement leads of the electrocardiogram data is a preset number by the data augmentation unit of the processor. can do.

Preferably, in the method according to the present invention, the body information processing unit of the processor calculates a body index based on the first body information, and the first body information and the body index are input to the second deep learning model. The method may further include generating second body information by pre-processing the first body information and the body index so as to be

Preferably, in the method according to the present invention, the first prediction unit of the processor generates the first onset probability information from the second electrocardiogram data by using the first deep learning model composed of a convolutional neural network model to which a residual block is applied. doing; and generating, by the first prediction unit, influencing factor information indicating a factor influencing generation of the first onset probability information by using an attention module.

Preferably, in the method according to the present invention, the second prediction unit of the processor selects an arbitrary detailed deep learning model corresponding to the second electrocardiogram data and the second body information, and uses the selected detailed deep learning model as the second prediction unit. configuring the second deep learning model using a deep learning model configuration automation module constituting the deep learning model; and generating second onset probability information from the second electrocardiogram data and the second body information by using the second deep learning model by a second prediction unit of the processor.

Preferably, in the method according to the present invention, the predictive result integration unit of the processor checks heart disease type information that maximizes the combined incidence probability information, and the combined incidence probability information corresponding to the identified heart disease type information. The method may further include generating the confirmed heart disease type information as incidence prediction heart disease type information in which onset is predicted when the probability exceeds a preset reference probability.

According to the present invention, first incidence probability information generated from a first deep learning model composed of a convolutional neural network model to which a residual block is applied and second probability generated from a second deep learning model constructed using a deep learning model configuration automation module By integrating the information to calculate the combined incidence probability of heart disease, it is possible to improve the accuracy of the predicted incidence probability.

1 is a diagram illustrating a connection configuration between a device for predicting the onset of heart disease based on a deep learning model using electrocardiogram data and body information according to an embodiment of the present invention and a measuring device.
FIG. 2 is a diagram showing the internal configuration of a measuring device included in an apparatus for predicting heart disease based on a deep learning model using electrocardiogram data and body information according to an embodiment of the present invention.
FIG. 3 is a diagram showing the internal configuration of a processor of an apparatus for predicting heart disease based on a deep learning model using electrocardiogram data and body information according to an embodiment of the present invention.
4 is a diagram for explaining a residual block used by an apparatus for predicting heart disease based on a deep learning model using electrocardiogram data and body information according to an embodiment of the present invention.
5 is a diagram for explaining an attention module used by an apparatus for predicting heart disease based on a deep learning model using electrocardiogram data and body information according to an embodiment of the present invention.
6 is a diagram for explaining a deep learning model configuration automation module used by an apparatus for predicting heart disease based on a deep learning model using electrocardiogram data and body information according to an embodiment of the present invention.
7 is a flow chart of an apparatus for predicting the onset of heart disease based on a deep learning model using electrocardiogram data and body information and a measurement method according to an embodiment of the present invention.

Hereinafter, various embodiments of the present invention will be described with reference to the accompanying drawings. However, it should be understood that this is not intended to limit the present invention to the specific embodiments, and includes various modifications, equivalents, and/or alternatives of the embodiments of the present invention. In connection with the description of the drawings, like reference numerals may be used for like elements.

In this document, expressions such as "has", "may have", "includes", or "may include" refer to the presence of a corresponding feature (eg, numerical value, function, operation, or component such as a part). , which does not preclude the existence of additional features.

In this document, expressions such as "A or B", "at least one of A and/and B", or "one or more of A or/and B" may include all possible combinations of the items listed together. . For example, "A or B", "at least one of A and B", or "at least one of A or B" includes (1) at least one A, (2) at least one B, Or (3) may refer to all cases including at least one A and at least one B.

Expressions such as "first", "second", "first", or "second" as used herein may modify various elements, in any order and/or importance, and may refer to one element as another. It is used to distinguish from components, but does not limit the components. For example, a first user device and a second user device may represent different user devices regardless of order or importance. For example, without departing from the scope of rights described in this document, a first element may be called a second element, and similarly, the second element may also be renamed to the first element.

A component (e.g., a first component) is "(operatively or communicatively) coupled with/to" another component (e.g., a second component); When referred to as "connected to", it should be understood that the certain component may be directly connected to the other component or connected through another component (eg, a third component). On the other hand, when an element (eg, a first element) is referred to as being “directly connected” or “directly connected” to another element (eg, a second element), the element and the above It may be understood that other components (eg, third components) do not exist between the other components.

As used in this document, the expression "configured to" means "suitable for", "having the capacity to", depending on the situation. ", "designed to", "adapted to", "made to", or "capable of" can be used interchangeably. The term "configured (or set) to" may not necessarily mean only "specifically designed to" hardware. Instead, in some contexts, the phrase "a device configured to" may mean that the device is "capable of" in conjunction with other devices or components. For example, the phrases "~unit configured (or set) to perform A, B, and C" or "~module configured (or configured) to perform A, B, and C" can be used to perform the corresponding action. It may refer to a dedicated processor (eg, embedded processor, MCU) or a general-purpose processor (eg, CPU, AP) capable of performing corresponding operations by executing one or more software programs stored in memory. there is.

The term "unit" used in the specification means a hardware component such as software, FPGA or ASIC, and "unit" performs certain roles. However, "part" is not meant to be limited to software or hardware. A “unit” may be configured to reside in an addressable storage medium and may be configured to reproduce one or more processors. Thus, as an example, "unit" refers to components such as software components, object-oriented software components, class components and task components, processes, functions, properties, procedures, subroutines, segments of program code, drivers, firmware, microcode, circuitry, data, databases, data structures, tables, arrays and variables. Functionality provided within components and “parts” may be combined into fewer components and “parts” or further separated into additional components and “parts”.

Terms used in this document are only used to describe a specific embodiment, and may not be intended to limit the scope of other embodiments. Singular expressions may include plural expressions unless the context clearly dictates otherwise. Terms used herein, including technical or scientific terms, may have the same meaning as commonly understood by a person of ordinary skill in the technical field described in this document. Among the terms used in this document, terms defined in general dictionaries may be interpreted as having the same or similar meaning as the meaning in the context of the related art, and unless explicitly defined in this document, in an ideal or excessively formal meaning. not interpreted In some cases, even terms defined in this document cannot be interpreted to exclude the embodiments of this document.

1 is a diagram illustrating a connection configuration between a device for predicting the onset of heart disease based on a deep learning model using electrocardiogram data and body information according to an embodiment of the present invention and a measuring device.

Referring to FIG. 1 , an apparatus 100 for predicting the onset of heart disease based on a deep learning model using electrocardiogram data and body information according to an embodiment of the present invention obtains first electrocardiogram data and first body information, respectively, and conducts a second electrocardiogram. Data and second body information may be generated. Thereafter, the device 100 for predicting heart disease based on the deep learning model using the electrocardiogram data and body information generates first onset probability information from the second electrocardiogram data using the first deep learning model, and uses the second deep learning model. A second onset probability may be generated from the second electrocardiogram data and the second body information. Finally, the apparatus 100 for predicting the onset of heart disease based on the deep learning model using electrocardiogram data and body information may generate integrated onset probability information by integrating the first onset probability information and the second onset probability information.

That is, the apparatus 100 for predicting the onset of heart disease based on a deep learning model using electrocardiogram data and body information uses first and second onset probability information by using different first deep learning models and second deep learning models, respectively. By generating each, finally, it is possible to generate integrated onset probability information.

Hereinafter, the device 100 for predicting heart disease based on a deep learning model using electrocardiogram data and body information performs 1) a process of acquiring first electrocardiogram data and first body information, and 2) obtaining second electrocardiogram data and second body information. A process of generating, 3) a process of generating first outbreak probability information, 4) a process of generating second outbreak probability information, and 4) a process of generating integrated outbreak probability information will be described.

FIG. 2 is a diagram showing the internal configuration of a measuring device included in an apparatus for predicting the onset of heart disease based on a deep learning model using electrocardiogram data and body information according to an embodiment of the present invention, and FIG. It is a diagram showing the internal configuration of the processor of the device for predicting heart disease based on a deep learning model using electrocardiogram data and body information according to.

1 to 3, the device 100 for predicting heart disease based on a deep learning model using electrocardiogram data and body information includes an acquisition unit 110, a processor 120, a communication unit 130, a display unit 140, and A storage unit 150 may be included.

The acquisition unit 110 may obtain first electrocardiogram data and first body information of the patient.

In detail, the acquisition unit 110 may obtain first electrocardiogram data and first body information of the patient through wired/wireless communication with the measuring device 10 .

Here, the measuring device 10 may measure first electrocardiogram data from a patient through one or more leads or may be worn on a patient's body to measure first electrocardiogram data.

In one embodiment, the measurement device 10 may be one or more of an electrocardiogram measurement device and a wearable device.

Meanwhile, the acquisition unit 110 may receive first body information from the patient, the medical staff in charge of the patient, and the patient's guardian.

To this end, the acquisition unit 110 may include an input interface such as a touch panel, keyboard, and mouse.

Here, the first body information may be information about the patient's body in addition to the first electrocardiogram data. For example, the first body information may include one or more of the patient's weight information, age information, and blood pressure information.

The processor 120 may generate second electrocardiogram data by augmenting the first electrocardiogram data, and generate second body information by pre-processing the first body information.

In addition, the processor 120 generates first onset probability information for the patient to develop a heart disease from the second electrocardiogram data using the first deep learning model, and uses the second deep learning model to generate the second electrocardiogram data and the second electrocardiogram data. 2 Second onset probability information for the patient to develop a heart disease may be generated from the body information.

Finally, the processor 120 may generate integrated onset probability information by integrating the first onset probability information and the second onset probability information.

To this end, the processor 120 may include a plurality of components as shown in FIG. 3 . Specifically, the processor 120 may include a data augmentation unit 121, a body information processing unit 122, a first prediction unit 123, a second prediction unit 124, and a prediction result integration unit 125. .

The data enhancer 121 may generate second electrocardiogram data by augmenting the first electrocardiogram data so that the number of measurement leads of the electrocardiogram data is a preset number.

To this end, the data augmentation unit 121 may pre-process the first ECG data using a Stacked Denoising AutoEncoder (SDAE) structure in which a hidden layer and noise are added to an autoencoder.

Here, an autoencoder is a neural network capable of learning dense representations of data, called latent representations, in an unsupervised manner, configured to copy inputs to outputs, encoding the data into lower-dimensional latent representations, and then converting the latent representations back into data. By decoding with , data can be compressed while minimizing reconstruction errors.

SDAE can be used to remove noise from data simply and efficiently by learning useful characteristics of data and simultaneously learning to remove noise with a structure in which a hidden layer and noise are added to the above-described autoencoder.

The data enhancer 121 may generate second electrocardiogram data by augmenting the first electrocardiogram data so that the number of measurement leads of the first electrocardiogram data from which noise is removed is a preset number. Here, the preset number may be 12. That is, the data enhancer 121 may generate second electrocardiogram data by augmenting the first electrocardiogram data so that the number of measurement leads is 12.

To this end, the data enhancer 121 may generate second ECG data by learning a Gaussian mixture model and augmenting the noise-removed first ECG data using the Gaussian mixture model.

Here, the Gaussian mixture model may be approximated as one distribution by weighting a plurality of Gaussian probability density functions, and may be implemented through Equation 1 below.

는 가중치 파라미터로, 각 분포의 가중치를 의미하며 가중치 값들은 기댓값 최대화 알고리즘을 통해 최적화될 수 있다.here,

is a weight parameter, meaning the weight of each distribution, and weight values can be optimized through an expected value maximization algorithm.

Meanwhile, the expectation maximization algorithm may be divided into a prediction step and a maximization step.

에 대해 파라미터

를 정의한 상태에서 각 레이블에 대한 확률 분포를 가정했을 경우, 해당 분포의 높이에 따라 i번째 데이터가 j번째 그룹에 속할 확률을 계산할 수 있다.In the prediction phase, the data

parameter for

Assuming a probability distribution for each label in the state where is defined, the probability that the i-th data belongs to the j-th group can be calculated according to the height of the distribution.

Specifically, the prediction step may be implemented through Equation 2 below.

값들을 이용하여 모수를 추정할 수 있다.In the maximization step, the calculated in the prediction step

Parameters can be estimated using the values.

Specifically, the maximization step may be implemented through Equation 3 below.

Meanwhile, the body information processor 122 may calculate a body index based on the first body information. For example, the body information processing unit 122 may calculate a BMI index as a body index based on weight information and height information included in the first body information.

Thereafter, the body information processing unit 122 may pre-process the first body information and the body index so that the first body information and the calculated body index are input to a second deep learning model to be described later.

At this time, the body information processing unit 122 may generate second body information by pre-processing the first body information and the body index corresponding to the input form of the second deep learning model.

That is, the body information processing unit 122 may generate both the preprocessed first body information and the body index as second body information.

Meanwhile, the first prediction unit 123 may generate first onset probability information from the second electrocardiogram data by using a first deep learning model composed of a convolutional neural network model to which a residual block is applied.

Here, the convolutional neural network (CNN) model consists of an input layer, a hidden layer, and an output layer, and a plurality of convolutional layers, a pooling layer, an activation function, and a fully connected layer exist in the hidden layer, and a batch normalization layer or drop An out layer may be included.

The convolution layer performs convolution on the input data, and can extract features through weights and calculations.

The pooling layer may be an operation for reducing the dimension of data, and the data whose dimension is reduced through the pooling operation may pass through an activation function and be transferred to the next layer.

The convolutional neural network according to the present invention may apply a ReLU activation function implemented through Equation 4 below as an activation function.

The fully connected layer may serve to connect one layer to another layer.

The batch normalization layer and the dropout layer are intended to prevent overfitting. The batch normalization of the batch normalization layer is a technique for performing normalization by subtracting the mean and dividing by the standard deviation for the batch input data, and the dropout of the dropout layer is one layer. It may be a technique of selecting only a certain percentage of nodes connected to the next layer in .

The output layer performs softmax regression, and softmax regression is a generalization of logistic regression used in binary classification for multinomial classification and can be used for multinomial classification supervised learning.

The convolutional neural network model according to the present invention may return a probability belonging to a class (when each of a plurality of heart diseases is affected or not) by performing softmax regression.

The convolutional neural network model according to the present invention may perform softmax regression using the softmax function implemented through Equation 5 below.

는 샘플 X에 대한 각 클래스의 점수를 담은 벡터이고,

는 샘플 x에 대한 각 클래스의 점수가 주어졌을 때, 해당 샘플이 i번째 클래스에 속할 추정 확률이다.where K is the number of classes,

is a vector containing the scores of each class for sample X,

is an estimated probability that the sample belongs to the i-th class given the score of each class for sample x.

가 가장 높은 클래스를 반환할 수 있다.At this time, the softmax regression classifier according to the present invention is calculated by the softmax function

can return the highest class.

In the convolutional neural network model according to the present invention, learning proceeds in the direction of minimizing the cost function implemented through Equation 6 below, overfitting is prevented by applying regulation to the cost function, and L2 regulation is used as a regulation method. can

Meanwhile, the first predictor 123 may construct a first deep learning model by applying a residual block to the convolutional neural network model described above.

The convolutional neural network model to which the residual block is applied includes a residual block with a shortcut connection, and has the advantage of solving the vanishing gradient using this. Here, the shortened connection is a connection that transcends one layer, and a total of two paths may exist: a path that moves according to the actually designed connection and a path that jumps over the path and is directly connected to the next step.

Meanwhile, the first predictor 123 according to another embodiment may configure the first deep learning model by further applying an attention module to the convolutional neural network model to which the residual block is applied.

As shown in FIG. 5, the attention module can predict the output by referring again to all input information of the encoder at each point in time when the output is predicted by the decoder, but only by referring to the input information associated with the part to be predicted at that point.

The attention module according to the present invention inputs the second electrocardiogram data to a convolutional neural network model to which a residual block is applied, receives a feature map as an output, divides the feature map into a plurality of patches, and inputs them to an encoder.

Through this, the first prediction unit 123 may generate first onset probability information indicating a probability that a corresponding heart disease will occur or not to occur in a patient for each of a plurality of types of heart disease information (class).

In addition, the first prediction unit 123 may generate influence factor information of the first outbreak probability information by using the attention module. Here, the influencing factor information may be information indicating a factor that has an influence on the generation of the first onset probability information. For example, the influencing factor information may be information indicating a factor influencing the generation of the first onset probability information among the second electrocardiogram data. 2 It may be measurement lead identification information of electrocardiogram data.

In this case, the first deep learning model may be composed of a plurality of models instead of one model.

That is, the first prediction unit 123 may generate first onset probability information by using each of a plurality of models constituting the first deep learning model for each of a plurality of heart disease type information (classes).

For example, when there are two types of heart disease information and the first deep learning model is composed of two models, the first prediction unit 123 may generate a total of four pieces of first onset probability information.

The second prediction unit 124 configures a second deep learning model using the deep learning model configuration automation module, and uses the second deep learning model to obtain second onset probability information from the second electrocardiogram data and the second body information. can create

Here, the deep learning model configuration automation module may be a module that selects an arbitrary detailed deep learning model in response to the second electrocardiogram data and the second body information and configures the second deep learning model with the selected detailed deep learning model.

Specifically, as shown in FIG. 6, the deep learning model construction automation module performs tasks such as deep learning model selection, feature design, and data segmentation, which were previously performed by workers to construct a deep learning model, on behalf of workers. to construct a deep learning model.

Accordingly, the second prediction unit 124 may transfer the second electrocardiogram data and the second body information to the deep learning model configuration automation module, and configure the second deep learning model using the deep learning model configuration automation module. .

Then, the second prediction unit 124 may generate second onset probability information from the second electrocardiogram data and the second body information by using the second deep learning model.

At this time, the second deep learning model may be composed of a plurality of models instead of one model.

That is, the second prediction unit 124 may generate second onset probability information by using each of a plurality of models constituting the second deep learning model for each of a plurality of heart disease type information (class).

For example, when there are two pieces of heart disease type information and the second deep learning model is composed of two models, the second prediction unit 124 may generate a total of four pieces of second onset probability information.

The prediction result integrator 125 may generate integrated onset probability information by integrating the first onset probability information and the second onset probability information.

Specifically, the prediction result integration unit 125 may generate integrated outbreak probability information by adding the first outbreak probability information and the second outbreak probability information. In this case, the prediction result integrator 125 may generate integrated onset probability information by adding the first onset probability information and the second onset probability information having the same corresponding heart disease type information.

For example, the first deep learning model is composed of the A1 model and the A2 model, the second deep learning model is composed of the B1 model and the B2 model, and the heart predicted from the A1 model, the A2 model, the B1 model, and the B2 model, respectively. When the disease type information is C1 heart disease and C2 heart disease, the prediction result integration unit 125 predicts the C1 heart disease, and the first incidence probability information generated from the A1 model and the A2 model, respectively, and the B1 model and the B2 model By summing up the second incidence probability information generated from each, a single integrated incidence probability information may be generated. Subsequently, the prediction result integration unit 125 predicts the C2 heart disease, sums the first incidence probability information generated from each of the A1 model and the A2 model, and the second incidence probability information generated from each of the B1 model and the B2 model, Another integrated onset probability information may be generated.

Thereafter, the prediction result integrator 125 checks the heart disease type information of the maximum combined incidence probability information, and if the integrated incidence probability information corresponding to the identified heart disease type information exceeds a preset reference probability, the confirmed heart disease The type information may be generated as onset prediction heart disease type information in which onset is predicted.

Meanwhile, the prediction result integration unit 125 according to another embodiment applies a first weight corresponding to the first deep learning model to the first onset probability information, and applies a second weight corresponding to the second deep learning model to the second By applying to the incidence probability information, integrated incidence probability information may be generated.

At this time, the prediction result integrator 125 according to another embodiment sets a first weight corresponding to the first deep learning model for each heart disease type information, and sets a second weight corresponding to the second deep learning model for each heart disease type information. Weight can be set.

On the other hand, the processor 120 may perform the operation of each of the above-described components, one or more cores (core, not shown) and a graphic processing unit (not shown) and / or a connection path for transmitting and receiving signals to and from other components ( For example, a bus, etc.) may be included.

The processor 120 may be configured to perform the operation of each component described above by executing one or more instructions stored in the storage unit 150 .

Meanwhile, the communication unit 130 may provide a communication interface necessary to provide a transmission/reception signal between an external device and each device in the form of packet data in conjunction with a communication network. Furthermore, the communication unit 130 may serve to transmit data in response to data requests from respective devices.

Here, the communication unit 130 may be a device including hardware and software necessary for transmitting and receiving signals such as control signals or data signals with other network devices through wired or wireless connections.

The display unit 140 may graphically display the above-described onset probability information and influencing factor information.

The storage unit 150 may store programs (one or more instructions) for processing and control of the processor 120 . Programs stored in the storage unit 150 may be divided into a plurality of modules according to functions.

7 is a flow chart of an apparatus for predicting the onset of heart disease based on a deep learning model using electrocardiogram data and body information and a measurement method according to an embodiment of the present invention.

Referring to FIG. 7 , in step S1, the acquisition unit may acquire first electrocardiogram data and first body information of the patient.

Then, in step S2, the processor may augment the first electrocardiogram data to generate second electrocardiogram data.

Step S2 may include generating second electrocardiogram data by augmenting the first electrocardiogram data so that the number of measurement leads of the electrocardiogram data is a preset number, by a data enhancer of the processor.

Subsequently, in step S3, the processor may pre-process the first body information to generate second body information.

In step S3, in the method according to the present invention, the body information processing unit of the processor calculates the body index based on the first body information, and the first body information and the body index are input to the second deep learning model. A step of generating second body information by pre-processing the information and the body index may be included.

Subsequently, in step S4 , the processor may generate first onset probability information for the patient to develop a heart disease from the second electrocardiogram data using the first deep learning model.

Step S4 may include generating first onset probability information from second electrocardiogram data by using a first deep learning model composed of a convolutional neural network model to which a residual block is applied by a first prediction unit of the processor.

In step S5 , the processor may generate second onset probability information for the patient to develop a heart disease from the second electrocardiogram data and the second body information by using the second deep learning model.

In step S5, the second prediction unit of the processor selects an arbitrary detailed deep learning model corresponding to the second electrocardiogram data and the second body information and configures the second deep learning model with the selected detailed deep learning model. constructing a second deep learning model using modules; and generating, by a second prediction unit of the processor, second onset probability information from second electrocardiogram data and second body information by using a second deep learning model.

In step S6, the processor may generate final outbreak probability information by integrating the first outbreak probability information and the second outbreak probability information.

In step S6, the prediction result integration unit of the processor checks heart disease type information that maximizes the integrated incidence probability information, and if the integrated incidence probability information corresponding to the identified heart disease type information exceeds a preset reference probability, The method may include generating heart disease type information as incidence prediction heart disease type information in which onset is predicted.

Steps of a method or algorithm described in relation to an embodiment of the present invention may be directly implemented as hardware, implemented as a software module executed by hardware, or implemented by a combination thereof. A software module may include random access memory (RAM), read only memory (ROM), erasable programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory, hard disk, removable disk, CD-ROM, or It may reside in any form of computer readable recording medium well known in the art to which the present invention pertains.

So far, the present invention has been mainly looked at with respect to preferred embodiments. Those skilled in the art to which the present invention belongs will understand that the present invention can be implemented in a modified form without departing from the essential characteristics of the present invention. Therefore, the disclosed embodiments should be considered from a descriptive point of view rather than a limiting point of view. The scope of the present invention is shown in the claims rather than the foregoing description, and all differences within the equivalent scope should be construed as being included in the present invention.

As described above, although the present invention has been described by the limited embodiments and drawings, the present invention is not limited thereto, and the technical spirit of the present invention and the following by those skilled in the art to which the present invention belongs Of course, various modifications and variations are possible within the scope of equivalents of the claims to be described.

10: measuring device
100: Heart disease prediction device based on deep learning model using electrocardiogram data and body information
110: acquisition unit
120: processor
130: communication department
140: display unit
150: storage unit

Claims (12)

an acquiring unit acquiring first electrocardiogram data and first body information of the patient; and
The first electrocardiogram data is augmented to generate second electrocardiogram data, the first body information is preprocessed to generate second body information, and a heart is provided to the patient from the second electrocardiogram data by using a first deep learning model. generating first incidence probability information of the onset of a disease, and generating second onset probability information of the onset of a heart disease in the patient from the second electrocardiogram data and the second body information using a second deep learning model; And a processor for integrating the first incidence probability information and the second incidence probability information to generate integrated incidence probability information.
A heart disease prediction device based on deep learning model using electrocardiogram data and body information.
According to claim 1,
The processor
and a data enhancer configured to generate the second electrocardiogram data by augmenting the first electrocardiogram data so that the number of measurement leads of the electrocardiogram data is a preset number.
A heart disease prediction device based on deep learning model using electrocardiogram data and body information.
According to claim 1,
The processor
A body index is calculated based on the first body information, and the first body information and the body index are pre-processed so that the first body information and the body index can be input to the second deep learning model to obtain a second body index. Characterized in that it comprises a; body information processing unit for generating information
A heart disease prediction device based on deep learning model using electrocardiogram data and body information.
According to claim 1,
The processor
A first predictor configured to generate the first onset probability information from the second electrocardiogram data using the first deep learning model composed of a convolutional neural network model to which a residual block is applied;
The first prediction unit
Characterized in that generating influencing factor information indicating a factor influencing the generation of the first onset probability information using an attention module
A heart disease prediction device based on deep learning model using electrocardiogram data and body information.
According to claim 1,
The processor
Using a deep learning model configuration automation module that selects an arbitrary detailed deep learning model corresponding to the second electrocardiogram data and the second body information and configures the second deep learning model with the selected detailed deep learning model, and a second predictor configured to construct 2 deep learning models and generate second onset probability information from the second electrocardiogram data and the second body information using the second deep learning model.
A heart disease prediction device based on deep learning model using electrocardiogram data and body information.
According to claim 1,
The processor
Heart disease type information that maximizes the combined incidence probability information is checked, and if the combined incidence probability information corresponding to the identified heart disease type information exceeds a preset standard probability, the identified heart disease type information is changed to onset probability. Characterized in that it comprises; a prediction result integration unit that generates the predicted onset prediction heart disease type information
A heart disease prediction device based on deep learning model using electrocardiogram data and body information.
obtaining, by an acquisition unit, first electrocardiogram data and first body information of the patient; and
generating, by a processor, second electrocardiogram data by augmenting the first electrocardiogram data;
generating, by the processor, second body information by pre-processing the first body information;
generating, by the processor, first onset probability information of a heart disease occurring in the patient from the second electrocardiogram data by using a first deep learning model;
generating, by the processor, second onset probability information for the onset of a heart disease in the patient from the second electrocardiogram data and the second body information by using a second deep learning model; and
characterized in that it comprises; generating integrated outbreak probability information by integrating, by the processor, the first outbreak probability information and the second outbreak probability information.
A heart disease prediction method based on deep learning model using electrocardiogram data and body information.
According to claim 7,
and generating the second electrocardiogram data by augmenting the first electrocardiogram data so that the number of measurement leads of the electrocardiogram data is a preset number by the data enhancer of the processor.
A heart disease prediction method based on deep learning model using electrocardiogram data and body information.
According to claim 7,
The body information processing unit of the processor calculates a body index based on the first body information, and the first body information and the body index are input to the second deep learning model. Characterized in that it further comprises; generating second body information by pre-processing the index
A heart disease prediction method based on deep learning model using electrocardiogram data and body information.
According to claim 7,
generating the first onset probability information from the second electrocardiogram data by using the first deep learning model composed of a convolutional neural network model to which a residual block is applied by a first prediction unit of the processor; and
Generating, by the first prediction unit, influencing factor information indicating a factor influencing generation of the first onset probability information by using an attention module; characterized in that it further comprises
A heart disease prediction method based on deep learning model using electrocardiogram data and body information.
According to claim 7,
The second prediction unit of the processor selects an arbitrary detailed deep learning model corresponding to the second electrocardiogram data and the second body information, and constructs a deep learning model for constructing the second deep learning model with the selected detailed deep learning model. configuring the second deep learning model using an automation module; and
Generating, by a second prediction unit of the processor, second onset probability information from the second electrocardiogram data and the second body information using the second deep learning model; characterized in that it further comprises
A heart disease prediction method based on deep learning model using electrocardiogram data and body information.
According to claim 7,
The prediction result integration unit of the processor checks heart disease type information that maximizes the combined incidence probability information, and if the combined incidence probability information corresponding to the checked heart disease type information exceeds a preset reference probability, the confirmation is made. Generating the heart disease type information as the onset prediction heart disease type information in which the onset is predicted; characterized in that it further comprises
A heart disease prediction method based on deep learning model using electrocardiogram data and body information.

Images