KR20220098600A - System for generating multiple lead electrocardiogram based on deep learning through time-differential single lead electrocardiogram - Google Patents

Abstract

The present invention synthesizes a plurality of residual induced electrocardiogram data from two or more standard induced electrocardiogram data through an electrocardiogram measuring unit 110 that measures two or more standard induced electrocardiogram data with a time difference, and a pre-learned deep learning algorithm Deep learning-based through single-guided electrocardiogram with a lag, which includes the first deep learning model unit 120 that generates multiple-guided electrocardiogram data, and utilizes a variety of multiple-guided electrocardiogram information for diagnosing ECG-related diseases that require A multi-guided electrocardiogram generation system is disclosed.

Description

Deep learning-based multi-guided electrocardiogram generation system using single-guided electrocardiogram with time difference

The present invention is a deep learning-based multiple-guided ECG generation system through a single-guided ECG with a lag, capable of synthesizing multiple-guided ECGs through a deep learning model from two or more standard-guided ECGs measured by a single-guided ECG measuring device is about

As is well known, since the development of the electrocardiogram, electrocardiogram-related knowledge has expanded exponentially, and it is possible to obtain information on the electrical function of the heart from electrocardiography and to diagnose various heart diseases such as arrhythmias, coronary artery disease, and myocardial disease.

Recently, research on the AI algorithm of electrocardiogram is being actively conducted, and the AI algorithm detects heart failure, predicts atrial fibrillation among arrhythmic rhythms, or determines gender.

In this way, by overcoming the limitations of humans, it is possible to detect subtle changes in the ECG waveform by the AI algorithm, and further improve the ECG interpretation.

On the other hand, the electrocardiogram used in the medical field is a 12-lead electrocardiogram, which is measured by attaching 3 extremity electrodes, 6 chest electrodes, and 1 ground electrode, and 10 electrodes, and the measured electrocardiogram data can be remotely transmitted.

However, since it is inconvenient to expose the chest area and attach 10 electrodes in daily life, portable pad measuring devices that can measure single-guided electrocardiogram, Galaxy Watch, Apple Watch, etc. are sometimes used.

The single-guided electrocardiogram measured as described above can be used for diagnosing arrhythmias, but there is a limit to its use in diagnosing diseases that require ECG information of various inductions, such as myocardial infarction.

Therefore, there is a need for a technology capable of synthesizing multiple-guided ECGs including 12-guided ECGs by generating other guided ECGs based on single-guided ECG information.

Korean Patent Application Laid-Open No. 10-2022-0008448 (An electrocardiogram generating apparatus and method based on generative 　 hostile 　 neural network algorithm, January 21, 2022) Korean Patent Publication No. 10-2362679 (EKG signal-based disease prediction method, 2022.02.14)

The technical problem to be achieved by the spirit of the present invention is to synthesize multiple-guided electrocardiograms through a deep learning model from two or more standard-guided electrocardiograms measured by a single-guided electrocardiogram measuring device. An object of the present invention is to provide a system for generating a multi-guided electrocardiogram based on learning.

In order to achieve the above object, an embodiment of the present invention provides an electrocardiogram measuring unit for measuring two or more standard-guided electrocardiogram data with a time difference; and a first deep learning model unit for synthesizing a plurality of residual induced electrocardiogram data from the two or more standard induced electrocardiogram data through a pre-trained deep learning algorithm to generate plural induced electrocardiogram data; We provide a system for generating multiple guided ECGs based on deep learning through guided ECG.

Here, the electrocardiogram measuring unit is an electrocardiogram device of a medical institution having a remote communication module, a portable wearable single induction electrocardiogram measuring device, or a portable single induction electrocardiogram pad measuring device, and the remote communication module is one-lead electrocardiogram data with an effective time difference of 10 seconds or more and the 2-lead ECG data may be identified as valid data and transmitted remotely.

In addition, the first deep learning model unit may be composed of a generative opposing neural network consisting of one or more pairs of generators and discriminators.

In addition, the generative adversarial neural network includes a first generator that receives the first-guided electrocardiogram data and the second-guided electrocardiogram data and synthesizes the residual induced electrocardiogram data, and the degree of agreement between the residual induced electrocardiogram data and the measured induced electrocardiogram data a first neural network of a first delimiter that converts ? into a probability value; and a second generator generating the first-lead ECG data and the second-lead ECG data from the residual guided electrocardiogram data synthesized by the first generator; and a second neural network of a second delimiter that converts a degree of agreement with the one-lead electrocardiogram data and the two-lead electrocardiogram data regenerated by a second generator into a probability value.

In addition, using the electrocardiogram data for determining the electrocardiogram-based disease and disease as learning data, and having one or more corresponding analysis modules for interpreting one or more characteristics for determining the diagnosis name of the disease, A second deep learning model unit for deriving a diagnosis name and a judgment characteristic of a diagnosed disease may be further included.

In addition, by comparing the diagnostic accuracy between the disease derived from the generated multiple-induced electrocardiogram data and the disease derived from the multiple-induced electrocardiogram data measured by the electrocardiogram measuring unit through the second deep learning model unit, the second 1 It is possible to determine the validity of the plurality of induced electrocardiogram data generated by the deep learning model unit.

In addition, through the second deep learning model unit, the characteristic values of the height, width, and slope of the P wave, QRS wave, and T wave extracted from the generated multiple induced electrocardiogram data, and multiple induced values measured by the electrocardiogram measuring unit By comparing the feature values of the height, width, and slope of the P wave, QRS wave, and T wave extracted from the electrocardiogram data, it is possible to determine the validity of the plurality of induced electrocardiogram data generated by the first deep learning model unit.

According to the present invention, by synthesizing multiple-guided electrocardiograms through a deep learning model from two or more standard-guided electrocardiograms measured by a single-guided electrocardiogram measuring device that can be used in daily life, a variety of multiple-guided electrocardiogram information is required in addition to arrhythmia diagnosis. There is an effect that can be used to diagnose ECG-related diseases.

1 shows a schematic configuration diagram of a deep learning-based multi-guided electrocardiogram generating system through a single-guided electrocardiogram with a parallax according to an embodiment of the present invention.
FIG. 2 illustrates the configuration of the first deep learning model unit of the deep learning-based multi-guided electrocardiogram generation system through the single-guided electrocardiogram with a time difference of FIG. 1 .
3 and 4 exemplify the multi-guided electrocardiogram generation result by the first deep learning model unit of FIG. 2 .
FIG. 5 schematically illustrates a method for generating a multiple-guided electrocardiogram by the deep learning-based multiple-guided electrocardiogram generating system through the single-guided electrocardiogram with a time difference of FIG. 1 .

Hereinafter, embodiments of the present invention having the above-described characteristics with reference to the accompanying drawings will be described in more detail.

The deep learning-based multiple-guided electrocardiogram generation system through the single-guided electrocardiogram with a time difference according to an embodiment of the present invention is an electrocardiogram measuring unit 110 that measures two or more standard-guided electrocardiogram data with a time difference, and pre-learning A variety of induced electrocardiogram information, including the first deep learning model unit 120, which generates multiple induced electrocardiogram data by synthesizing a plurality of residual induced electrocardiogram data from two or more standard induced electrocardiogram data through a deep learning algorithm It is intended to be used for the diagnosis of electrocardiogram-related diseases that require

Hereinafter, with reference to FIGS. 1 to 5 , a deep learning-based multi-guided electrocardiogram generating system using the parallax single-guided electrocardiogram of the above-described configuration will be described in detail as follows.

First, the electrocardiogram measuring unit 110 measures two or more N standard-guided electrocardiogram data from the body of the examinee with a time difference and transmits the measured values to the first deep learning model unit 120 .

For reference, the 12-lead ECG includes standard-guided ECGs of 1-lead ECG (lead I), 2-lead ECG (lead II), and 3-lead ECG (lead III), extremity guidance of avR, avL, and avF, and V1, V2, and V3. and chest guidance of V4, V5, and V6. 1-lead ECG is recorded by the potential difference between the left arm and right arm, 2-lead ECG is recorded by the potential difference between the right arm and left foot, and 3-lead ECG is recorded by the potential difference between the left and left arm As an electrocardiogram recorded by , the (+) and (-) poles exist where the minute is visible, and the standard lead has a potential relationship according to Einthoven's equation of 'lead Ⅱ = lead Ⅰ + lead Ⅲ'.

Accordingly, the electrocardiogram measuring unit 110 is an electrocardiogram device of a medical institution having a remote communication module, a portable wearable single induction electrocardiogram measuring device 111 or a portable single induction electrocardiogram pad measuring device 112, and single induction electrocardiogram data having a time difference and 2-lead ECG data may be measured, and 1-lead ECG data and 2-lead ECG data having random parallax may be applied as learning data of the first deep learning model unit 120 .

On the other hand, since one-lead ECG data and two-lead ECG data must be measured by changing the posture by single-guide measurement, 1-lead ECG data and 2-lead ECG data with a time difference of 10 seconds or more may be significant, preferably The remote communication module identifies the 1-lead ECG data and the 2-lead ECG data with an effective time difference within the range of 10 seconds to 3 minutes as valid data, along with an identifier indicating whether it is the 1-lead ECG data or the 2-lead ECG data, It may be remotely transmitted to the first deep learning model unit 120 .

Next, the first deep learning model unit 120, through a pre-trained deep learning algorithm, for example, a 12-guided electrocardiogram, a plurality of residual (12-N) induced electrocardiogram data from two or more N standard-guided electrocardiogram data synthesizing to generate multi-guided electrocardiogram data.

For example, the first deep learning model unit 120 is composed of a generative adversarial network (GAN) consisting of one or more pairs of generators and discriminators, so that one-lead ECG data, two-lead ECG data, and a plurality of inductions corresponding thereto Using the electrocardiogram data as learning data, it learns to generate multi-guided electrocardiogram data synthesized from the first-guided electrocardiogram data and the two-guided electrocardiogram data transmitted from the electrocardiogram measuring unit 110 .

As shown in FIG. 2 , the generative adversarial neural network includes a first generator 121 that receives 1-lead ECG data and 2-lead ECG data and synthesizes the residual induced ECG data, and the synthesized residual induced ECG data. The first neural network of the first delimiter 122 that converts the degree of agreement with the measured induced electrocardiogram data into a probability value, and the first induced electrocardiogram data and the second induced electrocardiogram data from the residual induced electrocardiogram data synthesized by the first generator 121 A second generator 123 that regenerates , converts the degree of agreement with the input 1-guide ECG data and 2-lead ECG data and the 1-lead ECG data and the 2-lead ECG data generated by the second generator 123 into a probability value. A second neural network of the second delimiter 124 may be included.

Accordingly, referring to FIGS. 3 and 4 , the first generator 121 of the first neural network receives the 1-lead ECG data and 2-lead ECG data (FIG. 3(a)) transmitted from the ECG measurement unit 110. , extremity induction of avR, avL, and avF (Fig. 3 (b)) and thoracic induction of V1, V2, V3, V4, V5, and V6 (Fig. 4) can be synthesized, and the first delimiter of the first neural network ( 122), it is possible to distinguish whether the synthesized residual induced ECG is valid (real) or invalid (fake) according to the degree of agreement between the synthesized residual induced ECG (indicated in orange) and the measured (actual) residual induced ECG (indicated in blue). have.

In addition, the second neural network determines the degree of agreement between the 1-lead ECG data and 2-lead ECG data transmitted from the electrocardiogram measuring unit 110 and the 1-lead ECG data and the 2-lead ECG data regenerated by the second generator 123 . Accordingly, the validity of the residual guided electrocardiogram synthesized by the first neural network may be determined.

On the other hand, it may further include a second deep learning model unit 130 capable of predicting an electrocardiogram-based disease from the plurality of induced electrocardiogram data synthesized by the first deep learning model unit 120 above, and a second deep learning model unit (130) uses the electrocardiogram data for determining the electrocardiogram-based disease and disease as learning data, and includes one or more corresponding analysis modules for interpreting the characteristics of one or more electrocardiogram data for determining the diagnosis name of the disease, the first deep learning model Diagnosis names of diseases such as brain disease, cardiovascular disease, and lung disease and judgment characteristics of the diagnosed disease may be derived from the multiple-induced electrocardiogram data generated by the unit 120 .

For example, the second deep learning model unit 130 is a convolutional neural network (CNN) model, which extracts characteristic values of diseases from an ECG image corresponding to ECG data, and interprets the characteristic values by an analysis module to characterize the characteristics. The diagnosis name of the electrocardiogram-based disease corresponding to the value and the judgment characteristic (reason for judgment) of the diagnosed disease may be derived and provided.

On the other hand, through the second deep learning model unit 130, the disease derived from the multiple induction electrocardiogram data generated by the first deep learning model unit 120, and the multiple induction actually measured by the electrocardiogram measuring unit 110 By comparing the diagnostic accuracy between diseases derived from the electrocardiogram data, the validity of the plurality of induced electrocardiogram data generated by the first deep learning model unit 120 may be additionally determined.

Or, through the second deep learning model unit 130, the characteristics of the height, width and slope of the P wave, QRS wave and T wave extracted from the plurality of induced electrocardiogram data generated by the first deep learning model unit 120 . By comparing the value and the characteristic values of the height, width, and slope of the P wave, QRS wave and T wave extracted from the multiple induced electrocardiogram data measured by the electrocardiogram measuring unit 110, the first deep learning model unit 120 It is also possible to determine the validity of the multi-guided electrocardiogram data generated by .

5 schematically illustrates a method for generating a multiple-guided electrocardiogram by a deep learning-based multiple-guided electrocardiogram generating system through a single-guided electrocardiogram with a time difference of FIG. 1, which will be briefly described with reference to this.

First, by the electrocardiogram measuring unit 110, two or more N standard-guided electrocardiogram data from the body of the examinee are respectively measured with a time difference and transmitted to the first deep learning model unit 120 (S110).

For example, the electrocardiogram measuring unit 110 may measure the 1-lead electrocardiogram data and the 2-lead electrocardiogram data having a disparity, and the 1-lead electrocardiogram data and the 2-lead electrocardiogram data having a random disparity are the first deep learning model unit 120 . It can be applied as the learning data of

Then, through the first deep learning model unit 120 having a pre-trained deep learning algorithm, a plurality of residual (12-N) (12-induction as an example) from two or more N standard-guided electrocardiogram data A plurality of guided electrocardiogram data is generated by synthesizing the guided electrocardiogram data (S120).

Here, the first deep learning model unit 120 is composed of a generative opposing neural network consisting of one or more pairs of generators and discriminators, so that one-lead ECG data, two-lead ECG data, and multiple-guided ECG data corresponding thereto are used as learning data. , learn to generate multi-guided electrocardiogram data synthesized from the 1-guided electrocardiogram data and the 2-guided electrocardiogram data transmitted from the electrocardiogram measuring unit 110 .

That is, the first deep learning model unit 120, from the one-lead electrocardiogram data and two-lead electrocardiogram data (FIG. 3 (a)) transmitted from the electrocardiogram measuring unit 110, avR, avL, and avF extremity induction ( 3(b)) and chest guidance of V1, V2, V3, V4, V5, and V6 (FIG. 4) can be synthesized, and the synthesized residual guided electrocardiogram (indicated in orange) and measured (actual) residual electrocardiogram (Indicated in blue), it is possible to distinguish whether the synthesized residual induced electrocardiogram is valid (real) or invalid (fake) according to the degree of correspondence with the first lead electrocardiogram data and the second lead electrocardiogram data transmitted from the electrocardiogram measuring unit 110 and The validity of the residual induced electrocardiogram synthesized by the first neural network may be determined according to the degree of agreement between the 1-lead ECG data and the 2-lead ECG data regenerated by the second generator 123 .

Then, through the second deep learning model unit 130, the ECG-based disease is predicted from the plurality of induced electrocardiogram data synthesized by the first deep learning model unit 120 (S130).

Here, the second deep learning model unit 130 uses the electrocardiogram data for determining the electrocardiogram-based disease and disease as learning data, and at least one corresponding analysis module for interpreting the characteristics of one or more electrocardiogram data for determining the diagnosis name of the disease. provided, the diagnosis name of diseases such as brain disease, cardiovascular disease, lung disease, and the like, and judgment characteristics (reason for judgment) of the diagnosed disease can be derived from the plurality of induced electrocardiogram data generated by the first deep learning model unit 120 . .

Therefore, by the configuration of the deep learning-based multiple-guided electrocardiogram generation system through the single-guided electrocardiogram with a time difference as described above, it is possible to deep By synthesizing multiple-guided electrocardiograms through the learning model, it is possible to utilize various kinds of multiple-guided electrocardiogram information in addition to diagnosing arrhythmias to diagnose ECG-related diseases.

The embodiments described in this specification and the configurations shown in the drawings are only the most preferred embodiment of the present invention, and do not represent all of the technical spirit of the present invention, so various equivalents that can replace them at the time of the present application It should be understood that there may be water and variations.

110: electrocardiogram measuring unit 111: wearable single induction electrocardiogram measuring device
112: single induction electrocardiogram pad measuring device
120: first deep learning model unit 121: first generator
122: first delimiter 123: second constructor
124: second delimiter 130: second deep learning model unit

Claims (7)

an electrocardiogram measuring unit that measures two or more standard-guided electrocardiogram data with a time difference; and
A first deep learning model unit for synthesizing a plurality of residual induced electrocardiogram data from the two or more standard induced electrocardiogram data to generate plural induced electrocardiogram data through a pre-trained deep learning algorithm; Containing,
Deep learning-based multi-guided electrocardiogram generation system using staggered single-guided electrocardiogram.
According to claim 1,
The electrocardiogram measuring unit is an electrocardiogram device of a medical institution having a remote communication module, a portable wearable single induction electrocardiogram measuring device, or a portable single induction electrocardiogram pad measuring device, and the remote communication module includes 1-lead electrocardiogram data and 2 with an effective time difference of 10 seconds or more. Characterized in that the induced electrocardiogram data is identified as valid data and transmitted remotely,
Deep learning-based multi-guided electrocardiogram generation system using staggered single-guided electrocardiogram.
3. The method of claim 2,
The first deep learning model unit is characterized in that it is composed of a generative antagonistic neural network consisting of one or more pairs of generators and discriminators,
Deep learning-based multi-guided electrocardiogram generation system using staggered single-guided electrocardiogram.
4. The method of claim 3,
The generative adversarial neural network is
a first generator for synthesizing the residual induced electrocardiogram data by receiving the 1-lead electrocardiogram data and the two-lead electrocardiogram data as input; a first neural network; and
a second generator for generating the first-lead ECG data and the second-lead ECG data from the residual guided electrocardiogram data synthesized by the first generator; It characterized in that it comprises a second neural network of a second delimiter that converts the degree of agreement with the one-lead electrocardiogram data and the two-lead electrocardiogram data regenerated by two generators into a probability value,
Deep learning-based multi-guided electrocardiogram generation system using staggered single-guided electrocardiogram.
According to claim 1,
Using the electrocardiogram data to determine the electrocardiogram-based disease and disease as learning data, and having one or more corresponding analysis modules for interpreting one or more characteristics for determining the diagnosis name of the disease, the diagnosis name of the disease and the Deriving the judgment characteristics of the diagnosed disease, characterized in that it further comprises a second deep learning model unit,
Deep learning-based multi-guided electrocardiogram generation system using staggered single-guided electrocardiogram.
6. The method of claim 5,
By comparing the diagnostic accuracy between the disease derived from the generated multiple induced electrocardiogram data and the disease derived from the multiple induced electrocardiogram data measured by the electrocardiogram measuring unit through the second deep learning model unit, the first deep Characterized in determining the validity of the plurality of induced electrocardiogram data generated by the learning model unit,
Deep learning-based multi-guided electrocardiogram generation system using staggered single-guided electrocardiogram.
6. The method of claim 5,
Through the second deep learning model unit, the characteristic values of the height, width, and slope of the P wave, QRS wave, and T wave extracted from the generated multiple induced electrocardiogram data, and multiple induced electrocardiogram data measured by the electrocardiogram measuring unit Comparing the feature values of the height, width and slope of the P wave, QRS wave and T wave extracted from
Deep learning-based multi-guided electrocardiogram generation system using staggered single-guided electrocardiogram.

Images