Electrocardiogram generation system and method based on deep learning algorithms

Abstract

To address the difficulty of measuring a 12-lead electrocardiogram at home or in everyday life, since the chest must be exposed to attach chest electrodes and it is difficult for the general public to correctly attach nine electrodes to accurate positions.SOLUTION: The present invention relates to a system for generating electrocardiograms based on a deep learning algorithm and a method thereof. According to the invention, the system comprises: a data input unit that receives 12-lead electrocardiograms measured from a plurality of patients; a data extraction unit that extracts training data from the input 12-lead electrocardiograms; a training unit that inputs the extracted training data into a plurality of training models to train the models on characteristics of the electrocardiograms; an electrocardiogram generation unit that receives one or more reference electrocardiograms from a subject to be measured, and generates virtual electrocardiograms by inputting the reference electrocardiogram into the plurality of trained training models; and a control unit that synchronizes the reference electrocardiogram and the generated virtual electrocardiograms with each other, and outputs waveforms for the synchronized reference electrocardiogram and virtual electrocardiograms.SELECTED DRAWING: Figure 2

Description

Technical Field

[0001] The present invention relates to an electrocardiogram generation system and method based on a deep learning algorithm, and more particularly, to an electrocardiogram generation system and method that generate a plurality of electrocardiograms from one or more lead electrocardiograms using a deep learning algorithm.

Background Art

[0002] A standard 12-lead electrocardiogram used in hospitals collects all 12-lead information after attaching six electrodes to the entire anterior chest and three electrodes respectively to the four limbs (four electrodes including a ground electrode), and integrates the information to diagnose a disease.

[0003] A 12-lead electrocardiogram records the electrical potential of the heart in 12 electrical directions centering on the heart, and by determining the state of the heart in multiple directions through this, it is possible to accurately read heart diseases limited to one site.

[0004] Measuring the electrical potential of the heart in multiple directions is meaningful in that the characteristics of the heart can be grasped in each direction, and for this reason, medical practice recommends measuring a standard 12-lead electrocardiogram for the diagnosis of heart diseases such as myocardial infarction.

[0005] However, in order to record 12 leads, the chest must be exposed to attach chest electrodes, and it is difficult for ordinary people to attach nine electrodes (three on the limbs and six on the chest) at correct positions, making measurement difficult at home or in daily life. In addition, it is difficult to move after attaching ten electrodes, so it is difficult to use for real-time monitoring.

[0006] Accordingly, recently, devices capable of measuring 1-lead electrocardiograms or two or more lead electrocardiograms have been developed so that they can be used in daily life.

[0007] First, one-lead electrocardiogram devices that use two electrodes include watch-type electrocardiogram devices (Apple Watch or Galaxy Watch). In a watch-type electrocardiogram device, the back of the watch is in contact with the left wrist, and the right fingers are in contact with the watch's crown, creating contact between the left arm electrode and the right arm electrode. The potential difference between the two electrodes is used to measure the lead I electrocardiogram.

[0008] Furthermore, the watch-type electrocardiogram device is worn on the left arm. Lead I is measured by touching the crown with the right hand, Lead II is measured by touching the crown with the right hand while the watch is on the stomach, and Lead III is measured by touching the crown with the left hand while the watch is on the stomach. After that, the V1-6 lead electrocardiogram is measured by touching the back of the watch to the V1-6 electrode positions while the left hand is in contact with the watch crown.

[0009] The aforementioned method has drawbacks in that it requires the user to accurately contact the electrocardiogram at the V1-6 positions, and unlike standard chest lead electrocardiograms, which must illustrate the potential difference between a virtual center point and the V electrode even when the V1-6 lead is in contact with that area, the V1-6 lead electrocardiogram cannot emulate standard chest leads because it only illustrates the potential difference between the right arm electrode and the V electrode.

[0010] Alternatively, it is possible to measure two or more lead electrocardiograms by holding electrodes in both hands, placing the electrodes on the back of the device against the leg or ankle, and placing electrodes on the left arm, right arm, and left leg, respectively. However, this method has the disadvantage of not being able to monitor for long periods (1 hour, 24 hours, 7 days, etc.) and requiring the device itself to have three electrodes.

[0011] The technology underlying this invention is disclosed in Korean Registered Patent No. 10-2180135 (published November 17, 2020). [Overview of the project] [Problems that the invention aims to solve]

[0012] Thus, the present invention provides an electrocardiogram generation system and method for generating multiple electrocardiograms from one or more lead electrocardiograms using a deep learning algorithm. [Means for solving the problem]

[0013] According to an embodiment of the present invention for achieving such technical challenges, the electrocardiogram generation system based on a deep learning algorithm includes: a data input unit that receives 12-lead electrocardiograms measured in multiple patients and patient information corresponding to the 12-lead electrocardiograms; a data extraction unit that classifies and stores the input 12-lead electrocardiograms according to the patient information and extracts learning data from the stored 12-lead electrocardiograms; a learning unit that inputs the extracted learning data into one or more learning models to learn the characteristics of the electrocardiograms; an electrocardiogram generation unit that receives one or more reference electrocardiograms from the subjects of measurement and inputs the input reference electrocardiograms into one or more learning models that have completed learning to generate a virtual electrocardiogram; and a control unit that synchronizes the reference electrocardiograms and the generated virtual electrocardiograms with each other and outputs waveforms for the synchronized reference electrocardiograms and virtual electrocardiograms.

[0014] The patient information may include at least one of the following: gender, age, presence or absence of cardiac disease, and the potential vector of the measured electrocardiogram.

[0015] The learning unit can be trained to generate 12-n electrocardiograms by inputting n leads of electrocardiograms, which are a portion of the 6-lead limb electrocardiograms and the 6-lead chest electrocardiograms, into the first learning model.

[0016] The learning unit constructs 12 second learning models for 6-lead electrocardiograms of the limbs and 6-lead electrocardiograms of the chest. When 12-n generated electrocardiograms from the constructed first learning model are input to the second learning model, it can be trained to convert the input lead electrocardiograms into a style having the corresponding potential vectors and output them as n virtual electrocardiograms.

[0017] The first and second learning models can be constructed by combining a generative adversarial network and an autoencoder method.

[0018] The electrocardiogram generation unit inputs the input reference electrocardiogram into a first learning model, extracts potential vectors for the reference electrocardiogram to generate the remaining virtual electrocardiograms, and inputs the generated virtual electrocardiograms into a second learning model that has been learned using the extracted potential vectors, thereby generating virtual electrocardiograms for leads that are identical to the reference electrocardiogram again.

[0019] The control unit matches and synchronizes the reference electrocardiogram with the plurality of virtual electrocardiograms, and outputs the synchronized electrocardiograms via a monitor.

[0020] Furthermore, according to an embodiment of the present invention, the electrocardiogram generation method using the electrocardiogram generation system includes the steps of: inputting 12-lead electrocardiograms measured in multiple patients and patient information corresponding to the 12-lead electrocardiograms; classifying and saving the input 12-lead electrocardiograms according to the patient information and extracting learning data from the saved 12-lead electrocardiograms; inputting the extracted learning data into one or more learning models to learn the characteristics of the electrocardiograms; inputting one or more reference electrocardiograms from the subjects of measurement and inputting the input reference electrocardiograms into one or more learning models that have completed learning to generate a virtual electrocardiogram; and synchronizing the reference electrocardiograms and the generated virtual electrocardiograms with each other and outputting waveforms for the synchronized reference electrocardiograms and virtual electrocardiograms. [Effects of the Invention]

[0021] Thus, according to the present invention, it is possible to generate multiple synchronized electrocardiograms from two electrocardiograms measured at different locations using a deep learning algorithm, allowing for accurate interpretation of cardiac diseases, similar to a 12-lead electrocardiogram.

[0022] Further, according to the present invention, since two leads of electrocardiogram are used, measurement can be performed at home and in daily life, measurement can be performed even in a moving state, and real-time monitoring is enabled; and since a synchronized electrocardiogram is generated by utilizing electrocardiogram information measured at different time points, medical personnel can interpret the synchronized lead electrocardiogram information corresponding to the relevant heartbeat, and more accurate diagnosis can be performed based on the information. [BRIEF DESCRIPTION OF THE DRAWINGS]

[0023] [Figure 1] FIG. 1 is a configuration diagram for explaining an electrocardiogram generation system according to an embodiment of the present invention. [Figure 2] FIG. 2 is a flowchart for explaining an electrocardiogram generation method using the electrocardiogram generation system according to an embodiment of the present invention. [Figure 3] FIG. 3 is an exemplary diagram for explaining a general method for measuring a 12-lead electrocardiogram. [Figure 4] FIG. 4 is an exemplary diagram for explaining step S240 shown in FIG. 2. [Figure 5] FIG. 5 is an exemplary diagram for explaining step S260 shown in FIG. 2. [MODE FOR CARRYING OUT THE INVENTION]

[0024] An electrocardiogram generation system 100 according to an embodiment of the present invention includes a data input unit 110, a data extraction unit 120, a learning unit 130, an electrocardiogram generation unit 140, and a control unit 150.

[0025] First, the data input unit 110 receives input of 12-lead electrocardiograms measured from a plurality of patients and patient information corresponding to the 12-lead electrocardiograms.

[0026] Here, the 12-lead electrocardiogram includes six limb leads and six precordial leads, and the patient information includes at least one of gender, age, presence or absence of heart disease, and a potential vector of the measured electrocardiogram.

[0027] The data extraction unit 120 classifies the 12-lead electrocardiograms using the input patient information and stores them in a database. The data extraction unit 120 then randomly selects multiple 12-lead electrocardiograms stored in the database to generate training data.

[0028] Subsequently, the learning unit 130 trains the learning model constructed using the training data. More specifically, the learning unit 130 constructs a first learning model that extracts potential vector information for the input lead electrocardiograms and generates a 12-n lead virtual electrocardiogram from an n lead reference electrocardiogram, and a second learning model that generates an n lead virtual electrocardiogram from the 12-n lead virtual electrocardiogram.

[0029] The learning unit 130 then inputs the generated training data into the constructed first or second learning model to train the first or second learning model.

[0030] The electrocardiogram generation unit 140 acquires one or more reference electrocardiograms from the subject being measured. The electrocardiogram generation unit 140 then inputs the acquired reference electrocardiograms into a first learning model that has completed learning, and generates a first virtual electrocardiogram based on the potential vector relative to the reference electrocardiogram. The electrocardiogram generation unit 140 then inputs the generated first virtual electrocardiogram into a second learning model that has also been learned, causing the second learning model to output a second virtual electrocardiogram.

[0031] Finally, the control unit 150 synchronizes the reference electrocardiogram and the virtual electrocardiogram, and outputs multiple synchronized electrocardiograms from the reference electrocardiogram and the virtual electrocardiogram via the monitoring device.

[0032] Modes for carrying out the invention

[0033] Hereinafter, preferred embodiments of the present invention will be described in detail based on the attached drawings. Note that the thickness of lines and the size of components shown in the drawings may be exaggerated for clarity and convenience of explanation.

[0034] Furthermore, the terms described later are defined in consideration of the functions of this invention, and these may change depending on the intent or convention of the user or operator. Therefore, the definitions of such terms should be based on the content throughout this specification.

[0035] Below, we will describe in more detail an electrocardiogram generation system based on a deep learning algorithm according to an embodiment of the present invention, based on Figure 1.

[0036] Figure 1 is a diagram illustrating the configuration of an electrocardiogram generation system according to an embodiment of the present invention.

[0037] As shown in Figure 1, the electrocardiogram generation system 100 according to an embodiment of the present invention includes a data input unit 110, a data extraction unit 120, a learning unit 130, an electrocardiogram generation unit 140, and a control unit 150.

[0038] First, the data input unit 110 receives 12-lead electrocardiograms measured in multiple patients and patient information corresponding to the 12-lead electrocardiograms.

[0039] Here, the 12-lead electrocardiogram includes a 6-lead extremity electrocardiogram and a 6-lead chest electrocardiogram, and the patient information includes at least one of the following: sex, age, presence or absence of cardiac disease, and the potential vector of the measured electrocardiogram.

[0040] The data extraction unit 120 classifies the 12-lead electrocardiograms using the input patient information and stores them in a database. The data extraction unit 120 then randomly selects multiple 12-lead electrocardiograms stored in the database to generate training data.

[0041] Subsequently, the learning unit 130 trains the learning model constructed using the training data. More specifically, the learning unit 130 extracts potential vector information for the input lead electrocardiograms and constructs a first learning model that generates a 12-n lead virtual electrocardiogram from an n lead reference electrocardiogram, and a second learning model that generates an n lead virtual electrocardiogram from the 12-n lead virtual electrocardiogram.

[0042] The learning unit 130 then inputs the generated training data into the constructed first or second learning model to train the first or second learning model.

[0043] The electrocardiogram generation unit 140 acquires one or more reference electrocardiograms from the subject being measured. The electrocardiogram generation unit 140 then inputs the acquired reference electrocardiograms into a first learning model that has completed learning, and generates a first virtual electrocardiogram based on the potential vector relative to the reference electrocardiogram. The electrocardiogram generation unit 140 then inputs the generated first virtual electrocardiogram into a second learning model that has also been learned, causing the second learning model to output a second virtual electrocardiogram.

[0044] Finally, the control unit 150 synchronizes the reference electrocardiogram and the virtual electrocardiogram, and outputs multiple synchronized electrocardiograms from the reference electrocardiogram and the virtual electrocardiogram via the monitoring device.

[0045] The following section will describe in more detail the electrocardiogram generation method using the electrocardiogram generation system 100 according to an embodiment of the present invention, based on Figures 2 to 5.

[0046] Figure 2 is a flowchart illustrating an electrocardiogram generation method using an electrocardiogram generation system according to an embodiment of the present invention.

[0047] As shown in Figure 2, the electrocardiogram generation method using the electrocardiogram generation system according to an embodiment of the present invention is divided into two stages: a stage in which a learning model is trained, and a stage in which an electrocardiogram is generated using the trained model after training is complete.

[0048] In the training phase of the learning model, the electrocardiogram generation system 100 first receives lead electrocardiograms measured from multiple patients and patient information (S210).

[0049] First, a 12-lead electrocardiogram records the electrical potential of the heart in 12 different directions, centered around the heart.

[0050] A common method for taking an electrocardiogram involves attaching four electrodes (limb leads) to both of the patient's arms and legs. The electrocardiogram device measures the potential using the arm electrodes and the left leg electrode, while the right leg electrode acts as the ground electrode.

[0051] Figure 3 is an illustrative diagram illustrating a typical method for measuring a 12-lead electrocardiogram.

[0052] As shown in Figure 3, the electrocardiogram measuring device generates a lead I electrocardiogram by subtracting the potential of the right arm from the potential of the left arm, a lead II electrocardiogram by subtracting the potential of the right arm from the potential of the left bundle branch, and a lead III electrocardiogram by subtracting the potential of the left arm from the potential of the left bundle branch.

[0053] The electrocardiogram (ECG) device calculates the average potential of the electrodes on both arms and the left leg to determine the potential of a virtual center point. Then, the ECG device generates the aVL lead ECG by subtracting the virtual center point potential from the left arm potential, and the aVR lead ECG by subtracting the virtual center point potential from the right arm potential. Furthermore, the ECG device generates the aVF lead ECG by subtracting the virtual center point potential from the left leg electrode potential. As described above, the ECG device generates a total of six-lead (six-row) electrocardiograms using three limb electrodes (four including the ground electrode).

[0054] Subsequently, the electrocardiogram measuring device generates a six-lead chest electrocardiogram using the difference between the potential of a virtual center point determined by the limb leads and the potentials of the six electrodes attached to the chest. Specifically, six electrodes, V1, V2, V3, V4, V5, and V6, are attached to the chest from the front to the left side of the chest at predetermined positions.

[0055] The electrocardiogram measuring device then generates a V1 lead electrocardiogram by subtracting the potential measured by the V1 electrode from the potential of a virtual center point obtained by averaging the limb electrodes.

[0056] As described above, the electrocardiogram measuring device generates a 12-lead electrocardiogram using electrodes attached to multiple patients. The generated 12-lead electrocardiogram is then transmitted to the electrocardiogram generation system 100.

[0057] At this point, the electrocardiogram generation system 100 receives the generated 12-lead electrocardiogram and its corresponding patient information as input.

[0058] Here, patient information includes at least one of the following: gender, age, presence or absence of cardiac disease, and the potential vector of the measured electrocardiogram.

[0059] Once step S210 is completed, the electrocardiogram generation system 100 extracts training data using the collected 12-lead electrocardiogram and patient information (S220).

[0060] To explain this further, the data extraction unit 120 classifies the collected 12-lead electrocardiograms according to patient information and stores them in a database. Then, the data extraction unit 120 randomly selects from the multiple stored 12-lead electrocardiograms to generate training data.

[0061] Subsequently, the learning unit 130 uses the generated training data to train the first learning model and the second learning model, respectively (S230).

[0062] First, the learning unit 130 inputs learning data composed of 12-lead electrocardiograms into the first and second learning models to learn the characteristics of the electrocardiograms. To explain this further, the direction of electrical flow differs depending on the potential vector in the lead electrocardiogram, and the patient's age and gender affect the electrocardiogram style. In other words, as people get older, the cardiac muscle decreases, so the amplitude of the electrocardiogram tends to decrease, and in the case of women, the position of the electrocardiogram electrodes is lower due to the breasts or the distance between the heart and the electrodes is larger, causing deformation of the electrocardiogram shape.

[0063] Furthermore, in cases of chronic lung disease (late-onset obstructive pulmonary disease), lung volume increases, and the heart, located between the lungs, is positioned vertically, causing the electrical flow in the heart to change vertically in three-dimensional space.

[0064] Therefore, the learning unit 130 inputs the lead electrocardiograms, which have been separated and stored according to patient information, namely the patient's age, sex, health status, and potential vector, into the first and second learning models. The first and second learning models then learn the characteristics of the input lead electrocardiograms and extract the potential vector information of the input lead electrocardiograms. However, it is also possible to receive and learn electrocardiograms without separating them by patient information, and this method allows for the learning of learning models that are not limited to a single characteristic.

[0065] Furthermore, the learning unit 130 constructs 12 second learning models corresponding to 12-lead electrocardiograms. The learning unit 130 then trains each of these second learning models based on the characteristics of the electrocardiogram.

[0066] To explain this further, the first learning model, based on a deep learning algorithm, learns the relationship between the input electrocardiogram and the 12-lead electrocardiogram, extracts at least one of the features of the input electrocardiogram, namely age, sex, and potential vectors, and thereby generates a 12-n lead virtual electrocardiogram from an n-lead reference electrocardiogram. On the other hand, the second learning model generates an n-lead virtual electrocardiogram from the 12-n lead virtual electrocardiogram. For example, assuming that a V1 lead electrocardiogram is to be generated, the learning unit 130 learns a style in the first learning model to generate virtual I, II, III, aVL, aVR, aVF, V2, V3, V4, V5, and V6 lead electrocardiograms from the V1 lead electrocardiogram, and the second learning model learns a style to generate a virtual V1 lead electrocardiogram from the virtual I, II, III, aVL, aVR, aVF, V2, V3, V4, V5, and V6 lead electrocardiograms. Then, the second learning model converts the input electrocardiogram into a V1 lead style to generate a virtual electrocardiogram.

[0067] The first learning model learns to identify which lead an input electrocardiogram (ECG) belongs to by inputting the lead type along with the ECG. This allows the model to generate a 12-lead ECG even when an arbitrary ECG with unknown leads is input.

[0068] Here, the first and second learning models are based on deep learning algorithms composed of autoencoders or generative adversarial networks, and the deep learning algorithm can be implemented using one of the autoencoders or generative adversarial networks selected from among them, or it can be implemented by mixing autoencoders and generative adversarial networks.

[0069] Once learning of the learning model is complete in stages S210 and S230, the electrocardiogram generation system 100 according to the embodiment of the present invention generates an electrocardiogram using the learned learning model.

[0070] First, the electrocardiogram generation system 100 receives a reference electrocardiogram measured via electrodes attached to the body of the person being measured (S240).

[0071] In this case, the reference electrocardiogram does not contain information about the potential vector.

[0072] Subsequently, the electrocardiogram generation unit 140 inputs the input reference electrocardiogram into the first learning model and the second learning model to generate multiple virtual electrocardiograms (S250).

[0073] First, the electrocardiogram generation unit 140 inputs a reference electrocardiogram into the first learning model and extracts the characteristics of the reference electrocardiogram.

[0074] Here, the characteristics include at least one of the following: the age of the person being measured, their sex, and their potential vector.

[0075] This generates a first virtual electrocardiogram from the reference electrocardiogram.

[0076] Subsequently, the electrocardiogram generation unit 140 inputs the generated first virtual electrocardiogram into the second learning model. The second learning model then generates a second virtual electrocardiogram based on the input first virtual electrocardiogram.

[0077] Figure 4 is an illustrative diagram illustrating the S240 stage shown in Figure 2.

[0078] As shown in Figure 4, the first electrocardiogram shows a reference electrocardiogram. Here, the first learning model assumes that the characteristic of the reference electrocardiogram is L1. Then, the electrocardiogram generation unit 140 generates virtual electrocardiograms for the remaining 11 leads based on the L1 electrocardiogram generated by the first learning model. After inputting the generated 11-lead virtual electrocardiograms into the second learning model, a virtual L1 lead electrocardiogram is generated. Each learning model is configured with a generator that generates electrocardiograms, along with a discriminator that determines whether the generated electrocardiogram was generated accurately and improves the accuracy of the generated electrocardiogram through feedback.

[0079] When step S250 is completed, the control unit 150 outputs the reference electrocardiogram and 11 virtual electrocardiograms via the monitoring device (S260).

[0080] Figure 5 is an illustrative diagram illustrating the S260 step shown in Figure 2.

[0081] The control unit 150 outputs a reference electrocardiogram and 11 virtual electrocardiograms. Then, as shown in Figure 5, the control unit 150 synchronizes the outputted reference electrocardiogram and the 11 virtual electrocardiograms to determine whether or not there is a health abnormality in the subject being measured. Here, since virtual electrocardiograms for leads identical to the reference electrocardiogram are also generated by the second learning model, it is also possible to make a determination using 12 virtual electrocardiograms.

[0082] As mentioned above, the electrocardiogram (ECG) output will differ in terms of slope, amplitude, etc., depending on the age, gender, and presence or absence of health abnormalities of the person being measured. Therefore, the control unit 150 outputs 12 ECGs synchronized with the reference ECG and 11 virtual ECGs via the monitoring device.

[0083] Thus, according to the present invention, it is possible to generate multiple synchronized electrocardiograms from two electrocardiograms measured at different locations using a deep learning algorithm, allowing for accurate interpretation of cardiac diseases, similar to a 12-lead electrocardiogram.

[0084] Furthermore, since the electrocardiogram generation system according to the present invention uses two lead electrocardiograms, it can be measured at home or in daily life, and can be measured even while moving, enabling real-time monitoring. It generates a synchronized electrocardiogram by utilizing electrocardiogram information measured at different points in time, so medical personnel can interpret the synchronized lead electrocardiogram information in accordance with the corresponding heartbeat, and based on this, a more accurate diagnosis can be made.

[0085] Although the present invention has been described based on the embodiments shown in the drawings, these are merely illustrative, and it will be understood by those with ordinary skill in the art to which the present invention pertains that a variety of modifications and equivalent other embodiments are possible. Therefore, the true scope of technical protection of the present invention must be determined by the technical idea of ​​the following claims. [Industrial applicability]

[0086] This invention uses a deep learning algorithm to generate multiple synchronized electrocardiograms from two electrocardiograms measured at different locations, allowing for accurate interpretation of cardiac diseases, similar to a 12-lead electrocardiogram, and making it industrially applicable to electrocardiogram generation systems based on various deep learning algorithms. [Explanation of Symbols]

[0087] 100 ECG generation systems 110 Data Input Section 120 Data Extraction Unit 130 Learning Department 140 ECG generation unit 150 Control Unit

Claims

1. A method for generating an electrocardiogram, performed by an electrocardiogram generation system, The data input unit receives the n-lead reference electrocardiogram measured via electrodes attached to the subject's body. The electrocardiogram generation unit inputs the n-lead reference electrocardiogram into a first learning model among multiple learning models trained based on a deep learning algorithm, thereby generating a 12-n lead virtual electrocardiogram based on the characteristics including the potential vectors of the n-lead reference electrocardiogram. The aforementioned n-lead reference electrocardiogram is a portion of the 6-lead limb electrocardiogram and the 6-lead chest electrocardiogram. The plurality of learning models include a first learning model trained to generate a 12-n lead electrocardiogram from an n lead electrocardiogram, and a second learning model trained to generate an n lead electrocardiogram from the 12-n lead electrocardiogram. method.

2. In the method according to claim 1, The first learning model extracts the characteristics from the n-lead reference electrocardiogram and generates a 12-n lead first virtual electrocardiogram as the 12-n lead virtual electrocardiogram based on these characteristics. method.

3. The method according to claim 1, The aforementioned characteristics further include at least one of gender and age. method.

4. The method according to claim 2, The second learning model generates a second virtual electrocardiogram with n leads by transforming the waveforms corresponding to the characteristics of the first virtual electrocardiogram with 12-n leads. method.

5. The method according to claim 2 or 4, The first and second learning models are constructed by mixing or using a generative adversarial network and an autoencoder, method.

6. The method according to claim 2, The first learning model receives a reference electrocardiogram of n leads, which is a portion of the six-lead electrocardiograms of the limbs and the six-lead electrocardiograms of the chest, as input, and learns to generate a 12-n lead electrocardiogram by determining the potential vectors of the n-lead reference electrocardiograms input to the first learning model. method.

7. The method according to claim 4, The second learning model receives a 12-n lead electrocardiogram generated by the first learning model as input, and is learned to generate an n-lead electrocardiogram by converting the 12-n lead electrocardiogram into waveforms corresponding to the potential vectors corresponding to the n-lead electrocardiograms. method.

8. The method according to claim 1, The reference electrocardiogram of n leads and the virtual electrocardiogram of 12-n leads are synchronized. method.

9. An electrocardiogram generation system, It comprises a data input unit and an electrocardiogram generation unit. The data input unit is configured to receive a reference electrocardiogram of n leads measured via electrodes attached to the body of the person being measured. The electrocardiogram generation unit is configured to generate a 12-n lead virtual electrocardiogram based on the characteristics including the potential vectors of the n lead reference electrocardiogram by inputting the n lead reference electrocardiogram into a first learning model among a plurality of learning models trained based on a deep learning algorithm. The aforementioned n-lead reference electrocardiogram is a portion of the 6-lead limb electrocardiogram and the 6-lead chest electrocardiogram. The plurality of learning models include a first learning model trained to generate a 12-n lead electrocardiogram from an n lead electrocardiogram, and a second learning model trained to generate an n lead electrocardiogram from the 12-n lead electrocardiogram. ECG generation system.

Images