Electrocardiogram generation device and method based on generative adversarial network algorithm

Abstract

The present invention relates to an electrocardiogram generating device and method based on a generative adversarial network algorithm. The electrocardiogram generating device based on a generative adversarial network algorithm according to the present invention includes an input unit that receives electrocardiogram data of a patient to be diagnosed, a control unit that inputs the input electrocardiogram data to a pre-trained generative adversarial network algorithm to generate a plurality of synthesized electrocardiogram data, and an output unit that outputs the input actual electrocardiogram data of the patient and the plurality of generated electrocardiogram data. As described above, according to the present invention, electrocardiogram data is trained by characteristics using a deep learning algorithm, and arrhythmia, one type of cardiac disease, is diagnosed using the trained model, thereby improving accuracy and providing the reason for arrhythmia diagnosis, thereby improving diagnostic reliability.

Description

[Technical Field]

[0001] The present invention relates to an electrocardiogram generation device and method based on a generative adversarial network algorithm, and more particularly to an electrocardiogram generation device and method based on a generative adversarial network algorithm that uses a deep learning algorithm to generate information on the remaining 12-n electrodes from information on n electrodes, and diagnoses the patient's condition using the generated electrode information. [Background technology]

[0002] An electrocardiogram is a graphic recording of the electrical potential associated with the heartbeat on the body's surface. In addition to the standard 12-lead electrocardiogram, there are exercise electrocardiograms and action electrocardiograms (Holter recording and event recording electrocardiograms). Many tests are used to diagnose cardiovascular disease, but the electrocardiogram has many advantages and is the most commonly used test in clinical practice. The electrocardiogram is accurate, simple, reproducible, can be easily recorded repeatedly, and is a non-invasive test with low testing costs. The electrocardiogram is most commonly used to diagnose arrhythmias and coronary artery disease.

[0003] A standard 12-lead electrocardiogram involves attaching six electrodes to the front of the chest and three to each of the limbs, collecting all 12 lead information, and combining this information to diagnose illness. However, since 12-lead electrodes require the chest to be exposed and it is difficult to attach all 12 electrodes, it is difficult to measure at home or in everyday life.

[0004] Recently, wearable electrocardiogram equipment has been developed that uses only six electrodes (three of the twelve electrodes on the hands and feet) or only one electrode, like a patch-type product, for measurement. FIG. 1 shows 12-lead electrocardiogram data.

[0005] For example, as shown in Figure 1, a 12-lead electrocardiogram used in hospitals simultaneously measures the electrocardiogram from a total of 12 leads on both the left and right sides. However, wearable electrocardiogram equipment only uses information from the leftmost limb electrode (three electrodes are attached to the limbs to measure information from six electrodes: I, II, III, aVL, aVF, and aVL), or collects electrocardiogram information from only one of these electrodes, such as I or II.

[0006] As mentioned above, if only information from six electrodes or only information from one electrode is used, only half or 1 / 12 of the information from the original 12-lead electrodes is used, which results in a problem of reduced accuracy. The technology behind the present invention is disclosed in Korean Patent Registration No. 10-1109738 (published on February 24, 2012). Summary of the Invention [Problem to be solved by the invention]

[0007] The technical problem to be solved by the present invention is to provide an electrocardiogram generating device and method based on a generative adversarial network algorithm that uses a deep learning algorithm to generate information on the remaining 12-n electrodes from information on n electrodes, and predicts the patient's condition using the generated electrode information. [Means for solving the problem]

[0008] To achieve this technical objective, an electrocardiogram generating device based on a deep learning algorithm (generative adversarial network algorithm) according to an embodiment of the present invention includes an input unit that receives input electrocardiogram data of a patient to be diagnosed with a disease, a data generating unit (control unit) that generates multiple synthesized electrocardiogram data by inputting the input electrocardiogram data into a pre-trained generative adversarial network algorithm, and an output unit that outputs the input actual electrocardiogram data of the patient and the multiple generated electrocardiogram data.

[0009] The system may further include a learning unit that extracts lead electrocardiogram data from all electrocardiogram data of a patient diagnosed with a heart disease, and inputs the extracted lead electrocardiogram data into a pre-constructed generative adversarial network algorithm to learn to generate a plurality of synthesized electrocardiogram data.

[0010] The learning unit may include a first generation model that generates n pieces of synthetic electrocardiogram data from lead electrocardiogram data extracted from all input electrocardiogram data, and a second generation model that generates m pieces of synthetic electrocardiogram data from the n pieces of synthetic electrocardiogram data generated by the first generation model.

[0011] The learning unit may include a first division model that receives input of lead electrocardiogram data or m pieces of synthetic electrocardiogram data and classifies the data as real or synthetic, and a second division model that receives input of all (remaining) electrocardiogram data excluding the lead electrocardiogram data or n pieces of synthetic electrocardiogram data and classifies the data as real or synthetic.

[0012] In addition, an electrocardiogram generating method based on a deep learning algorithm (generative adversarial network algorithm) using an electrocardiogram generating device according to another embodiment of the present invention includes the steps of receiving input of electrocardiogram data of a patient to be diagnosed with a disease, inputting the input electrocardiogram data into a previously trained generative adversarial network algorithm to generate a plurality of synthesized electrocardiogram data, and outputting the input actual electrocardiogram data of the patient and the generated plurality of electrocardiogram data. [Effects of the Invention]

[0013] As described above, according to the present invention, the accuracy of diagnosing heart-related diseases can be improved by generating n additional electrocardiograms using electrocardiograms measured with one electrode or three electrodes using a deep learning algorithm.Furthermore, since the present invention is applicable to portable wearable electrocardiogram equipment, it can be used at home or in daily life. [Brief explanation of the drawings]

[0014] [Figure 1] FIG. 1 is a diagram for explaining a general standard 12-lead electrocardiogram method. [Figure 2] 1 is a block diagram illustrating an electrocardiogram generation device based on a generative adversarial network algorithm according to an embodiment of the present invention; [Figure 3] 1 is a flowchart illustrating the operation flow of an electrocardiogram generation method based on a generative adversarial network algorithm according to an embodiment of the present invention. [Figure 4] FIG. 10 is a diagram showing types of electrocardiogram data. [Figure 5] 1 is a diagram illustrating a generative model and a segmented model according to an embodiment of the present invention; DETAILED DESCRIPTION OF THE INVENTION

[0015] An electrocardiogram generating device based on a deep learning algorithm, the electrocardiogram generating device including: an input unit that receives electrocardiogram data of a patient to be diagnosed; a data generating unit that generates a plurality of synthesized electrocardiogram data by inputting the input electrocardiogram data to a trained generative adversarial network algorithm; and an output unit that outputs the input actual electrocardiogram data of the patient, the generated plurality of electrocardiogram data, and a diagnosis result. MODE FOR CARRYING OUT THE INVENTION

[0016] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the following description, the thickness of lines and the size of components shown in the drawings may be exaggerated for clarity and convenience.

[0017] Furthermore, the terms used below are defined in consideration of the functions of the present invention, and may vary depending on the intentions or practices of users or operators. Therefore, the definitions of such terms should be determined based on the overall content of this specification. First, an electrocardiogram generating device based on a deep learning algorithm according to an embodiment of the present invention will be described with reference to FIG. FIG. 2 is a block diagram illustrating an electrocardiogram generating device based on a generative adversarial network algorithm according to an embodiment of the present invention.

[0018] As shown in FIG. 2, an electrocardiogram generating device 100 based on a generative adversarial network algorithm according to an embodiment of the present invention includes an input unit 110, a data generating unit (control unit) 120, a learning unit 130, and an output unit 140.

[0019] First, the input unit 110 receives the electrocardiogram data of a patient, and further receives information (indicator) indicating which signal the input electrocardiogram data is.

[0020] The data generator 120 inputs the input electrocardiogram data to a trained generative adversarial network algorithm, thereby generating a plurality of synthesized electrocardiogram data.

[0021] The learning unit 130 receives electrocardiogram data measured from a patient with a heart disease as input, extracts lead data from the input electrocardiogram data, and inputs the extracted lead data to a generative adversarial network algorithm, thereby learning to output synthesized electrocardiogram data.

[0022] Here, the learning unit 130 constructs a plurality of generative models and segmentation models based on a generative adversarial network algorithm. Thus, the learning unit 130 trains the generative models to generate synthesized electrocardiogram data, and trains the segmentation models to analyze the generated synthetic electrocardiogram data and real electrocardiogram data and distinguish whether the data is synthetic or not. Finally, the output unit outputs a diagnosis result using the input electrocardiogram data and the plurality of electrocardiogram data generated by synthesis. The method for generating electrocardiogram data will now be described in more detail with reference to FIGS.

[0023] Figure 3 is a flowchart showing the operational flow of an electrocardiogram generation method based on a generative adversarial network algorithm according to an embodiment of the present invention, Figure 4 is a diagram showing types of electrocardiogram data, and Figure 5 is a diagram for explaining a generative model and a segmentation model according to an embodiment of the present invention. As shown in FIG. 3, the input unit 110 according to the embodiment of the present invention receives electrocardiogram data measured using 12 lead electrodes (S310).

[0024] More specifically, the input unit 110 receives electrocardiogram data measured by attaching 12-lead electrodes to a body part of a patient diagnosed with arrhythmia or heart disease, and further receives information (indicator) indicating which signal the input electrocardiogram data is.

[0025] In the embodiment of the present invention, the input electrocardiogram data is described as electrocardiogram data acquired using 12-lead electrodes, but this is not limited to this, and electrocardiogram data acquired using 6-lead electrodes, 18-lead electrodes, or 24-lead electrodes may also be used as needed. Thereafter, the input unit 110 transmits the input 12 pieces of electrocardiogram data and the corresponding diagnosis results to the learning unit 130.

[0026] Then, the learning unit 130 extracts n lead electrocardiogram data from the 12 electrocardiogram data, inputs the extracted n lead electrocardiogram data into the first generation model, and trains it to extract 12-n synthetic electrocardiogram data (S320).

[0027] To explain this again, the learning unit 130 collects electrocardiogram data measured from multiple patients. Then, the learning unit 130 extracts n lead electrocardiogram data from each group. The extracted lead electrocardiogram data is input to a first generative model based on a generative adversarial network algorithm.

[0028] To explain this again, as shown in Figure 4, an electrocardiogram is recorded as a graph of three standard leads (I / II / III), three limb leads (aVR / aVR / aVF), and six chest leads (V1-V6).

[0029] Therefore, the first generative model receives input of n pieces of input electrocardiogram data and information (indicator) about the signal corresponding to the lead electrocardiogram data. Then, the first generative model is generated by synthesizing the remaining electrocardiogram data excluding the input lead electrocardiogram data.

[0030] For example, if it is assumed that the input lead electrocardiogram data is three limb lead data, the first generative model generates nine pieces of synthetic electrocardiogram data excluding the three limb lead data.

[0031] Thereafter, the learning unit 130 inputs the synthesized 12-n pieces of electrocardiogram data and the actual 12-n pieces of electrocardiogram data into the first section model, and causes the model to learn to determine whether the data is synthesized (S330).

[0032] Specifically, the first division model cross-analyzes nine pieces of synthetic electrocardiogram data and nine pieces of actual electrocardiogram data excluding three limb lead data, and determines whether the nine pieces of synthetic electrocardiogram data are synthetic electrocardiogram data or actual electrocardiogram data.

[0033] Thereafter, the learning unit 130 inputs the 12-n pieces of composite electrocardiogram data generated in step S320 into the second generation model, and trains the second generation model to generate n pieces of composite electrocardiogram data again (S340).

[0034] In step S32O, the first generative model generates nine pieces of synthetic electrocardiogram data from three limb lead data, and the second generative model synthesizes and generates three limb lead data using the nine pieces of synthetic electrocardiogram data.

[0035] After completing step S340, the learning unit 130 inputs the n pieces of synthetic electrocardiogram data and the n pieces of lead electrocardiogram data into the second section model, and trains the second section model to determine whether the data are synthetic (S350).

[0036] The learning unit 130 repeatedly learns a generative model that generates electrocardiogram data and a segmented model that determines whether the data is synthetic, thereby learning to generate an electrocardiogram model that is close to the actual one.

[0037] Meanwhile, in an embodiment of the present invention, as shown in FIG. 5, two generative models and two divisional models are constructed so that they can compete with each other, but the number of generative models and divisional models can also be increased or decreased. When step S350 is completed, the input unit 110 receives the electrocardiogram data of the subject (S360).

[0038] Here, the input electrocardiogram data is not all of the 12 electrocardiogram data but some of the electrocardiogram data. The input unit 110 transmits the electrocardiogram data to the control unit 120.

[0039] Then, the control unit 120 inputs the transferred n electrocardiogram data as lead electrocardiogram data into the trained generative adversarial network algorithm to generate 12-n synthetic electrocardiogram data (S370).

[0040] In the case of a wearable device, an electrocardiogram is measured using one to three electrodes, and therefore, the control unit 120 inputs one or three pieces of electrocardiogram data into a generative adversarial network algorithm.

[0041] The Generative Adversarial Network algorithm then generates 11 synthetic ECG data from one lead ECG data, and when three lead ECG data are input, the Generative Adversarial Network algorithm generates nine synthetic ECG data. Thereafter, the output unit 140 outputs the generated electrocardiogram data (S380).

[0042] That is, the output unit 140 outputs the input electrocardiogram data and the generated electrocardiogram data, and the output 12 pieces of electrocardiogram data can be output through the terminal of the subject or to a terminal used by a medical professional, who then uses the output electrocardiogram data to diagnose heart disease.

[0043] In this way, according to the present invention, the accuracy of diagnosing heart-related diseases can be improved by generating n additional electrocardiograms using electrocardiograms measured with one electrode or three electrodes using a deep learning algorithm.Furthermore, according to the present invention, since it can be applied to portable wearable electrocardiogram equipment, it can be used at home or in daily life.

[0044] Although the present invention has been described with reference to the embodiments shown in the drawings, these are merely illustrative, and those skilled in the art will recognize that various modifications and equivalent embodiments are possible. Therefore, the true technical scope of the present invention should be determined by the technical spirit of the following claims. [Explanation of symbols]

[0045] 100 Electrocardiogram generating device 110 Input section 120 control section 130 Learning Department 140 Output section

Claims

1. An electrocardiogram generating device that generates information on the remaining 12-n electrodes from information on n electrodes based on a generative adversarial network algorithm, an input unit for receiving n pieces of electrocardiogram data of a patient who is to be diagnosed with a disease; a control unit that generates a plurality of synthetic electrocardiogram data by inputting the n pieces of input electrocardiogram data to a previously constructed generative adversarial network algorithm; an output unit that outputs the input electrocardiogram data and the generated plurality of composite electrocardiogram data; The generative adversarial network algorithm includes a first generative model that generates 12-n pieces of synthetic electrocardiogram data from the n pieces of input electrocardiogram data, and a second generative model that generates n pieces of synthetic electrocardiogram data from the 12-n pieces of synthetic electrocardiogram data. Electrocardiogram generating device.

2. The system further includes a learning unit that extracts lead electrocardiogram data from all electrocardiogram data of a patient diagnosed with a cardiac disease, and inputs the extracted lead electrocardiogram data into the previously constructed generative adversarial network algorithm to learn to generate a plurality of the synthetic electrocardiogram data, the first and second generative models generate a plurality of pieces of synthetic electrocardiogram data using the extracted lead electrocardiogram data; The electrocardiogram generating device according to claim 1 .

3. The learning unit constructing the first and second generative models and a segmental model based on the generative adversarial network algorithm; a first division model included in the division model, which receives input of the n lead electrocardiogram data extracted from all input electrocardiogram data and the n synthetic electrocardiogram data and classifies the data as real or synthetic; a second division model included in the division model, which receives input of 12-n pieces of lead electrocardiogram data excluding the lead electrocardiogram data input to the first division model and 12-n pieces of synthetic electrocardiogram data, and classifies the data as real or synthetic; The electrocardiogram generating device according to claim 2 .

4. An electrocardiogram generation method using an electrocardiogram generation device that generates information on the remaining 12-n electrodes from information on n electrodes based on a generative adversarial network algorithm, receiving n pieces of electrocardiogram data of a patient to be diagnosed; generating a plurality of synthetic electrocardiogram data by inputting the n pieces of electrocardiogram data to a previously constructed generative adversarial network algorithm; and outputting the input electrocardiogram data and the generated plurality of composite electrocardiogram data, The generative adversarial network algorithm includes a first generative model that generates 12-n pieces of synthetic electrocardiogram data from the n pieces of input electrocardiogram data, and a second generative model that generates n pieces of synthetic electrocardiogram data from the 12-n pieces of synthetic electrocardiogram data. Electrocardiogram generation method.

5. The method further includes a step of extracting lead electrocardiogram data from all electrocardiogram data of a patient diagnosed with a cardiac disease, and inputting the extracted lead electrocardiogram data into the previously constructed generative adversarial network algorithm to learn to generate a plurality of the synthetic electrocardiogram data; the first and second generative models generate a plurality of pieces of synthetic electrocardiogram data using the extracted lead electrocardiogram data; The method for generating an electrocardiogram according to claim 4.

6. The step of training to generate electrocardiogram data includes: constructing the first and second generative models and a segmental model based on the generative adversarial network algorithm; receiving the n lead electrocardiogram data extracted from all input electrocardiogram data and the n synthetic electrocardiogram data using a first division model included in the division model, and classifying the n lead electrocardiogram data as real or synthetic; and using a second division model included in the division model, receiving input of 12-n pieces of lead electrocardiogram data excluding the lead electrocardiogram data input to the first division model and 12-n pieces of synthetic electrocardiogram data, and classifying the data as real or synthetic. The method for generating an electrocardiogram according to claim 5.

Images