Method, program and device for predicting health status using electrocardiogram

Abstract

According to an embodiment of the present disclosure, there is disclosed a method, a program, and an apparatus for predicting a health condition using an electrocardiogram, which are performed by a computing device. The method may include the steps of: acquiring electrocardiogram data and cardiac arrest data including whether or not a subject has experienced cardiac arrest and the time of cardiac arrest for which the electrocardiogram data was measured; analyzing missing values of the electrocardiogram data or the cardiac arrest data to extract valid data from the electrocardiogram data; and generating training data by labeling the extracted valid data based on whether or not a subject has experienced cardiac arrest and the time of cardiac arrest.

Description

[Technical Field]

[0001] The present disclosure relates to artificial intelligence technology in the medical field, and more particularly to a method for training an artificial intelligence model based on the results of filtering electrocardiogram data to reflect cardiac arrest conditions, and predicting health conditions using the trained model. [Background technology]

[0002] An electrocardiogram (ECG) is a test that records the electrical activity of the heart. An ECG is a relatively simple and cost-effective test that can confirm the health of the heart, which plays an important role in the early diagnosis and management of heart disease. For example, ECG signals measured through an ECG can be used to determine whether each part of the heart is functioning normally, whether the size and position of the heart are normal, and whether there is any damage to the myocardium. Based on these determinations, ECG signals can be used to diagnose various heart-related problems and predict a person's health status.

[0003] Meanwhile, as artificial intelligence technology develops, there are increasing attempts to use it to predict health conditions based on electrocardiograms. The most important aspect of introducing artificial intelligence technology is securing the appropriate data necessary for AI training. Since electrocardiograms record cardiac activity, cardiac arrest has a significant impact on ECG interpretation, but most conventional technologies often do not reflect the occurrence and current status of cardiac arrest in the data secured for AI training. Summary of the Invention [Problem to be solved by the invention]

[0004] The present disclosure aims to provide a method for training an artificial intelligence model based on the results of filtering electrocardiogram data to reflect cardiac arrest conditions, and predicting health conditions using the trained model.

[0005] However, the problems to be solved by the present disclosure are not limited to those mentioned above, and other problems not mentioned will be clearly understood based on the following description. [Means for solving the problem]

[0006] To achieve the above object, one embodiment of the present disclosure provides a method for predicting a health condition using an electrocardiogram, the method being performed by a computing device. The method includes the steps of: acquiring electrocardiogram data and cardiac arrest data including whether or not a subject has experienced cardiac arrest and the duration of cardiac arrest for which the electrocardiogram data was measured; analyzing missing values of the electrocardiogram data or the cardiac arrest data to extract valid data from the electrocardiogram data; and generating training data by labeling the extracted valid data based on whether or not a subject has experienced cardiac arrest and the duration of cardiac arrest.

[0007] Alternatively, the step of analyzing the missing values of the electrocardiogram data or the cardiac arrest data to extract valid data from the electrocardiogram data may include the steps of: analyzing the time of cardiac arrest and clinical reasons related to cardiac arrest recorded in the cardiac arrest data to determine a cutoff time for filtering the electrocardiogram data; and filtering the electrocardiogram data of subjects in which cardiac arrest occurred within the cutoff time from the acquired electrocardiogram data to extract valid data.

[0008] Alternatively, the step of determining a cutoff time for filtering the electrocardiogram data by analyzing the time of cardiac arrest and clinical evidence regarding cardiac arrest included in the cardiac arrest data may include determining, as the cutoff time, any one of a first candidate time calculated based on clinical judgment of electrocardiogram changes occurring before cardiac arrest occurs, a second candidate time clinically determined as a time when treatment to prevent cardiac arrest is possible when cardiac arrest is predicted, or a third candidate time corresponding to the difference between the time of cardiac arrest recorded in the cardiac arrest data and the actual time of cardiac arrest of the subject whose electrocardiogram data was measured.

[0009] Alternatively, the cutoff time may be determined to be the largest value among the first candidate time, the second candidate time, or the third candidate time.

[0010] Alternatively, the step of generating learning data by labeling the extracted valid data based on whether cardiac arrest occurred and the time of cardiac arrest may include labeling the valid data of subjects who have experienced cardiac arrest within a predetermined time after the cutoff time as a cardiac arrest group, and labeling the valid data of subjects who have not experienced cardiac arrest or who have not died within a predetermined time after the cutoff time as a normal group.

[0011] Alternatively, the step of analyzing the missing values of the electrocardiogram data or the cardiac arrest data to extract valid data from the electrocardiogram data may include, when one or more leads themselves are missing among the N leads of the electrocardiogram data, excluding the electrocardiogram data in which one or more leads themselves are missing among the N leads from the valid data.

[0012] Alternatively, the step of analyzing the missing values of the electrocardiogram data or the cardiac arrest data to extract valid data from the electrocardiogram data may include a step of excluding, from the valid data, electrocardiogram data in which there is a lead among the N leads of the electrocardiogram data that has a missing value for more than a predetermined time, if any of the N leads has a missing value for more than a predetermined time.

[0013] Alternatively, the method may further include inputting the training data into a deep learning model to train the deep learning model so that the deep learning model calculates the likelihood of cardiac arrest occurring.

[0014] To achieve the above-described object, one embodiment of the present disclosure discloses a method for predicting a health condition using an electrocardiogram, performed by a computing device. The method includes the steps of acquiring first electrocardiogram data and inputting the first electrocardiogram data into a pre-trained deep learning model to predict the health condition of a subject whose electrocardiogram data was measured. In this case, the deep learning model may be trained based on training data generated by labeling valid data extracted from second electrocardiogram data based on whether or not cardiac arrest occurred and the duration of cardiac arrest. The valid data may be extracted from the second electrocardiogram data based on missing values of the second electrocardiogram data or on the results of analyzing cardiac arrest data including whether or not cardiac arrest occurred and the duration of cardiac arrest in the subject whose second electrocardiogram data was measured.

[0015] Alternatively, the prediction result derived by inputting the first electrocardiogram into the pre-trained deep learning model may include a score value indicating a health state of a subject whose electrocardiogram data was measured, and the score value closer to a first critical value may indicate a healthier state, and the score value closer to a second critical value may indicate a less healthy state.

[0016] To achieve the above object, one embodiment of the present disclosure provides a computer program stored in a computer-readable storage medium. When executed by one or more processors, the computer program performs operations for predicting a health condition using an electrocardiogram. The operations may include: acquiring electrocardiogram data and cardiac arrest data including whether or not a subject has experienced cardiac arrest and the time of cardiac arrest; analyzing missing values of the electrocardiogram data or the cardiac arrest data to extract valid data from the electrocardiogram data; and labeling the extracted valid data based on whether or not a subject has experienced cardiac arrest and the time of cardiac arrest to generate training data.

[0017] To achieve the above-described object, one embodiment of the present disclosure discloses a computing device for predicting a health condition using an electrocardiogram. The device may include a processor including at least one core; a memory including program code executable by the processor; and a network unit for acquiring electrocardiogram data and cardiac arrest data including whether or not a subject has experienced cardiac arrest and the time of cardiac arrest. In this case, the processor may analyze missing values of the electrocardiogram data or the cardiac arrest data to extract valid data from the electrocardiogram data, and label the extracted valid data based on whether or not a subject has experienced cardiac arrest and the time of cardiac arrest to generate training data. [Effects of the Invention]

[0018] The present disclosure can provide a method for training an artificial intelligence model based on the results of filtering electrocardiogram data to reflect cardiac arrest conditions, and predicting health conditions using the trained model. [Brief explanation of the drawings]

[0019] [Figure 1] FIG. 1 is a block diagram of a computing device according to one embodiment of the present disclosure. [Figure 2] FIG. 10 is a block diagram illustrating a process of extracting valid data from electrocardiogram data according to an embodiment of the present disclosure. [Figure 3] FIG. 1 is a block diagram illustrating a process of training a deep learning model based on valid data according to an embodiment of the present disclosure. [Figure 4] 1 is a flowchart illustrating a method for predicting a health condition using an electrocardiogram according to an embodiment of the present disclosure. [Figure 5] 10 is a flowchart illustrating a method for predicting health status using an electrocardiogram according to an alternative embodiment of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0020] Hereinafter, with reference to the accompanying drawings, embodiments of the present disclosure will be described in detail so that those skilled in the art can easily implement the present disclosure. The embodiments presented in this disclosure are provided to enable those skilled in the art to use or practice the contents of the present disclosure. Therefore, various modifications to the embodiments of the present disclosure will be apparent to those skilled in the art. That is, the present disclosure may be embodied in various different forms and is not limited to the following embodiments.

[0021] Throughout the specification of the present disclosure, the same or similar reference numerals refer to the same or similar components. In addition, in order to clearly explain the present disclosure, reference numerals of parts in the drawings that are not related to the explanation of the present disclosure may be omitted.

[0022] The term "or" as used in this disclosure is intended to mean an inclusive "or" rather than an exclusive "or." That is, unless otherwise specified in this disclosure or the context makes the meaning clear, "X utilizes A or B" should be understood to mean one of the natural inclusive permutations. For example, unless otherwise specified in this disclosure or the context makes the meaning clear, "X utilizes A or B" can be interpreted as either X utilizes A, X utilizes B, or X utilizes both A and B.

[0023] The term "and / or" as used in this disclosure should be understood to refer to and include all possible combinations of one or more of the associated listed concepts.

[0024] The terms "comprises" and / or "comprises" as used in this disclosure should be understood to mean that the specified feature and / or component is present. However, the terms "comprises" and / or "comprises" should be understood not to exclude the presence or addition of one or more other features, other components and / or combinations thereof.

[0025] Unless otherwise specified in this disclosure or clear from the context as indicating the singular form, the singular should generally be construed as including "one or more."

[0026] The term "Nth (N is a natural number)" used in the present disclosure may be understood as an expression used to distinguish components of the present disclosure from one another based on a predetermined criterion, such as functional, structural, or convenience of description. For example, components performing different functional roles in the present disclosure may be classified as a first component or a second component. However, components that are substantially the same within the technical concept of the present disclosure but must be distinguished for convenience of description may also be classified as a first component or a second component.

[0027] The term "acquire" as used in this disclosure may be understood to mean generating data in an on-device manner as well as receiving data over a wired or wireless communication network with an external device or system.

[0028] Meanwhile, the terms "module" or "unit" used in this disclosure may refer to an independent functional unit that processes computing resources, such as a computer-related entity, firmware, software or a portion thereof, hardware or a portion thereof, or a combination of software and hardware. In this regard, a "module" or "unit" may refer to a unit composed of a single element or a unit expressed as a combination or set of multiple elements. For example, in a narrow sense, a "module" or "unit" may refer to a hardware element or a set of elements of a computing device, an application program that performs a specific software function, a processing procedure implemented through the execution of software, or a set of instructions for program execution. In a broad sense, a "module" or "unit" may refer to a computing device itself that constitutes a system, or an application executed on a computing device. However, the above concepts are merely examples, and the concepts of "module" and "unit" may be defined in various ways within the scope of those skilled in the art based on the contents of this disclosure.

[0029] The term "model" as used in this disclosure may be understood as a system implemented using mathematical concepts and language to solve a specific problem, a set of software units for solving a specific problem, or an abstract model of a processing process for solving a specific problem. For example, a neural network "model" may refer to a system implemented as a neural network that has problem-solving capabilities through learning. In this case, a neural network can acquire problem-solving capabilities by optimizing parameters connecting nodes or neurons through learning. A neural network "model" may include a single neural network or a neural network ensemble in which multiple neural networks are combined.

[0030] The explanations of the above terms are intended to aid in understanding the present disclosure. Therefore, unless the above terms are explicitly stated as matters that limit the contents of the present disclosure, please note that they are not used to limit the technical ideas of the contents of the present disclosure.

[0031] FIG. 1 is a block diagram of a computing device according to an embodiment of the present disclosure.

[0032] The computing device 100 according to an embodiment of the present disclosure may be a hardware device or part of a hardware device that performs comprehensive data processing and calculations, or may be a software-based computing environment connected via a communication network. For example, the computing device 100 may be a server that performs intensive data processing functions and shares resources, or a client that shares resources through interaction with the server. The computing device 100 may also be a cloud system in which multiple servers and clients interact to process data comprehensively. The above description is merely an example of a type of computing device 100, and various types of computing devices 100 may be configured within a scope that would be understandable to one skilled in the art based on the contents of this disclosure.

[0033] 1, a computing device 100 according to an embodiment of the present disclosure may include a processor 110, a memory 120, and a network unit 130. However, since FIG. 1 is merely an example, the computing device 100 may include other components for implementing a computing environment. Also, only some of the disclosed components may be included in the computing device 100.

[0034] The processor 110 according to an embodiment of the present disclosure may be understood as a component including hardware and / or software for performing computing operations. For example, the processor 110 may read a computer program to perform data processing for machine learning. The processor 110 may process operations such as input data processing for machine learning, feature extraction for machine learning, and error calculation based on backpropagation. The processor 110 for performing such data processing may include a central processing unit (CPU), a general-purpose graphics processing unit (GPGPU), a tensor processing unit (TPU), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), etc. The types of processor 110 described above are merely examples, and various types of processor 110 may be configured within the scope of what would be understood by one skilled in the art based on the present disclosure.

[0035] The processor 110 can filter the electrocardiogram data based on the electrocardiogram data and cardiac arrest data of the subject for which the electrocardiogram data was measured. The processor 110 can analyze the electrocardiogram data and the corresponding cardiac arrest data to extract valid data that can be used for learning from the electrocardiogram data. For example, the processor 110 can analyze whether or not cardiac arrest occurred and the time of cardiac arrest included in the cardiac arrest data to filter out electrocardiogram data that has little learning utility and is difficult to address based on prediction results. The processor 110 can analyze the extent to which missing values exist in the electrocardiogram data to filter out electrocardiogram data that has no effect on learning or that may even have a negative impact on learning. In this way, the processor 110 can filter electrocardiogram data based on information about cardiac arrest and information about missing electrocardiogram data itself to select valid data that can be used for learning.

[0036] The processor 110 may generate training data for a deep learning model that predicts a health state based on the filtered electrocardiogram data. The processor 110 may label valid data extracted from the electrocardiogram data and generate training data using the labeled valid data. For example, the processor 110 may label valid data based on information included in cardiac arrest data corresponding to the valid data. The processor 110 may label information included in the cardiac arrest data, such as whether or not cardiac arrest has occurred and the time of cardiac arrest, as valid data. The processor 110 may determine the labeled valid data as training data for a model that predicts the occurrence of cardiac arrest.

[0037] The processor 110 can train a deep learning model that predicts a health condition using an electrocardiogram based on the generated training data. Since the training data includes input electrocardiogram data and labels, when the deep learning model outputs a prediction result for a health condition based on the input included in the training data, the processor 110 can compare the output with the label and adjust the neural network parameters included in the deep learning model. The processor 110 can train the deep learning model by repeatedly adjusting the neural network parameters to minimize the error between the output and the label. In this case, the deep learning model can include a neural network that extracts features from an electrocardiogram and a neural network that calculates a prediction value for a health condition based on the extracted features. For example, the processor 110 can input valid data labeled with information about cardiac arrest to the deep learning model and output a probability value for the possibility of cardiac arrest (or the possibility of death) in the form of a score between 0 and 100. The processor 110 can calculate an error using a loss function with the score and label output by the deep learning model as input variables and adjust the neural network parameters included in the deep learning model in a direction that reduces the error. The processor 110 can repeatedly perform this process to train the deep learning model to predict the occurrence of cardiac arrest.

[0038] The processor 110 can use the trained deep learning model to predict the health state of a subject whose electrocardiogram data is measured based on newly acquired electrocardiogram data. The processor 110 can input the electrocardiogram data to the trained deep learning model and calculate a predicted health state of the subject whose electrocardiogram data is measured. At this time, the predicted health state result may include a score representing the degree of deterioration of the overall health state, such as the probability of death of the subject whose electrocardiogram data is measured. For example, the processor 110 can input the newly acquired electrocardiogram data to the trained deep learning model and output a probability value for the possibility of cardiac arrest (or the possibility of death) in the form of a score between 0 and 100. Here, 0 can be understood to represent a healthy state with an extremely low probability of death, and 100 can be understood to represent a dangerous state with the highest probability of death. In other words, the processor 110 can use the trained deep learning model to calculate a score that allows an intuitive understanding of a person's overall health state from electrocardiogram data.

[0039] The memory 120 according to an embodiment of the present disclosure may be understood as a component including hardware and / or software for storing and managing data processed by the computing device 100. That is, the memory 120 may store any type of data generated or determined by the processor 110 and any type of data received by the network unit 130. For example, the memory 120 may include at least one type of storage medium selected from the group consisting of flash memory, hard disk, multimedia card micro, card-type memory, random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, and optical disk. The memory 120 may also include a database system for managing data in a predetermined manner. The types of memory 120 described above are merely examples, and the types of memory 120 may be configured in various ways within the scope of what would be understood by a person skilled in the art based on the contents of this disclosure.

[0040] The memory 120 may structure, organize, and manage data, combinations of data, and program code executable by the processor 110 required for the processor 110 to perform calculations. For example, the memory 120 may store medical data received through the network unit 130 (described below). The memory 120 may store program code for operating a neural network model to receive input medical data and perform learning, program code for operating a neural network model to receive input medical data and perform inference according to the intended use of the computing device 100, and processed data generated by executing the program code.

[0041] The network unit 130 according to an embodiment of the present disclosure may be understood as a component that transmits and receives data through any type of publicly-known wired / wireless communication system. For example, the network unit 130 may transmit and receive data using a wired / wireless communication system such as a local area network (LAN), wideband code division multiple access (WCDMA), long term evolution (LTE), wireless broadband internet (WiBro), 5th generation mobile communication (5G), ultra wideband, ZigBee, radio frequency (RF) communication, wireless LAN, wireless fidelity (WiFi), near field communication (NFC), or Bluetooth. The above-mentioned communication systems are merely examples, and various wired / wireless communication systems for transmitting and receiving data by the network unit 130 may be applied other than the above-mentioned examples.

[0042] The network unit 130 may receive data necessary for the processor 110 to perform calculations via wired / wireless communication with any system or any client. The network unit 130 may also transmit data generated through calculations by the processor 110 via wired / wireless communication with any system or any client. For example, the network unit 130 may receive medical data through communication with a database in a hospital environment, a cloud server that performs tasks such as medical data standardization, a client such as a smart watch, or a medical computing device. The network unit 130 may transmit output data of the neural network model, and intermediate data and processed data derived during the calculation process of the processor 110 via communication with the aforementioned database, server, client, or computing device.

[0043] FIG. 2 is a block diagram illustrating a process of extracting valid data from electrocardiogram data according to one embodiment of the present disclosure.

[0044] 2, a computing device 100 according to an embodiment of the present disclosure can analyze electrocardiogram data 10 and cardiac arrest data 20 to extract valid data 40 from the electrocardiogram data 10. Specifically, the computing device 100 can analyze the time of cardiac arrest and clinical evidence related to cardiac arrest recorded in the cardiac arrest data 20 to determine a cutoff time 39 for filtering the electrocardiogram data. The computing device 100 can then filter the electrocardiogram data of subjects who have experienced cardiac arrest within the cutoff time 39 from the electrocardiogram data 10 to extract valid data 40 from the electrocardiogram data 10. That is, the computing device 100 can calculate the cutoff time 39 based on information related to cardiac arrest and extract data of subjects who have experienced cardiac arrest within a predetermined time after the cutoff time 39 as valid data 40. Because electrocardiogram signals vary greatly before the time of death, the usefulness of analysis is limited for predictions close to the time of death. Thus, the computing device 100 can filter valid data 40 from the electrocardiogram data 10 based on a cutoff time 39 related to the time of cardiac arrest to provide data worthy of analysis for predicting health status.

[0045] For example, the computing device 100 can analyze the time of cardiac arrest recorded in the cardiac arrest data 20 and clinical evidence related to cardiac arrest to calculate candidate times for deriving a cutoff time 39. The candidate times can include a first candidate time 31 calculated based on clinical judgment of electrocardiogram changes occurring before cardiac arrest occurs, a second candidate time 33 clinically determined as a time when treatment to prevent cardiac arrest can be performed when cardiac arrest is predicted to occur, and a third candidate time 35 corresponding to the error between the time of cardiac arrest recorded in the cardiac arrest data and the actual time of cardiac arrest in the subject whose electrocardiogram data was measured.

[0046] The computing device 100 may calculate, as the first candidate time 31, the N minutes at which the probability of cardiac arrest is high, the changes are large, and the accuracy of the analysis through the deep learning model is low. The electrocardiogram measured for N minutes from N minutes before cardiac arrest occurs to the occurrence of cardiac arrest may generally have a larger range of changes than a healthy state. Furthermore, even if the analysis is performed through a deep learning model for this portion, the accuracy of the analysis results regarding the possibility of cardiac arrest may tend to be low. Therefore, the computing device 100 may calculate, based on clinical judgment regarding cardiac arrest, the time at which the electrocardiogram changes are estimated to begin to occur most significantly as the first candidate time 31.

[0047] The computing device 100 may calculate, as the second candidate time 33, M minutes, which is a time period determined to be possible for treatment to prevent cardiac arrest before the time when cardiac arrest is predicted to occur. Even if the deep learning model predicts a time shorter than M minutes as the expected time of cardiac arrest, if the time period is so short that treatment to prevent cardiac arrest is impossible, the prediction itself, which is performed for the purpose of preventing cardiac arrest in advance, becomes meaningless. Therefore, the computing device 100 may calculate, as the second candidate time 33, a clinical time period determined to be possible for treatment to prevent cardiac arrest. In this case, the time period determined to be possible for treatment to prevent cardiac arrest may be determined based on personal information included in the cardiac arrest data and clinical evidence related to cardiac arrest. The computing device 100 may analyze how many minutes from the time when cardiac arrest is clinically predicted to occur to prevent cardiac arrest, taking into account the age, gender, etc. of the subject included in the personal information. The computing device 100 may then determine the time derived as a result of the analysis as the second candidate time 33.

[0048] The computing device 100 may calculate, as the second candidate time 35, the difference L minutes between the actual time of cardiac arrest and the time recorded in the cardiac arrest data by medical staff. The actual time of cardiac arrest may be determined based on electrocardiogram data. Because an electrocardiogram signal indicates cardiac activity, the computing device 100 may determine the time when the electrocardiogram signal included in the electrocardiogram data remains unchanged as the actual time of cardiac arrest. Even if a patient's heart actually stopped at 12:24, medical staff may subjectively record the cardiac arrest as occurring at 12:30. In this case, an electrocardiogram measured after 12:26 may be misinterpreted as an electrocardiogram measured four minutes before the onset of cardiac arrest when viewed in conjunction with the cardiac arrest data, even though it was taken after the onset of cardiac arrest. Therefore, to prevent misinterpretation due to such an error, the computing device 100 may determine, as the third candidate time 35, the difference between the actual time of cardiac arrest and the time of cardiac arrest included in the cardiac arrest data.

[0049] The computing device 100 can compare the magnitudes of the first candidate time 31, the second candidate time 33, and the third candidate time 35 to determine the cutoff time 39, which is the time before cardiac arrest occurs when analysis is least useful. Then, the computing device 100 can determine the largest value among the first candidate time 31, the second candidate time 33, and the third candidate time 35 as the cutoff time 39 based on the comparison result. For example, the largest value among the first candidate time 31 (N minutes), the second candidate time 33 (M minutes), and the third candidate time 35 (L minutes) can include the remaining time. Therefore, the computing device 100 can determine the largest value among the candidate times 31, 33, and 35 that includes the remaining time as the cutoff time 39 (T minutes).

[0050] The computing device 100 can exclude data within a cutoff time 39 from the electrocardiogram data 10 and extract the remaining data as valid data 40. For example, if the cutoff time 39 is determined to be T minutes, the computing device 100 can determine that the data within T minutes from the electrocardiogram data 10 is data with low learning utility. The computing device 100 can determine that the data after T minutes from the electrocardiogram data 10 is data with high learning utility. Furthermore, the computing device 100 can determine that the data from the electrocardiogram data 10 for which cardiac arrest has not occurred is data with high learning utility. The computing device 100 can then determine that the data from the electrocardiogram data 10 with high learning utility is valid data 40.

[0051] Meanwhile, although not shown in FIG. 2 , the computing device 100 may analyze missing values in the electrocardiogram data 10 to extract valid data 40 from the electrocardiogram data 10. The computing device 100 may analyze the number of missing values based on the leads themselves or the measurement time of the electrocardiogram signal included in the electrocardiogram data 10. The computing device 100 may then filter the electrocardiogram data 10 based on the analysis results to generate valid data 40. For example, if one or more of the N leads of the electrocardiogram data 10 are missing, the computing device 100 may exclude from the valid data 40 the electrocardiogram data in which one or more of the N leads are missing. If any of the N leads of the electrocardiogram data 10 has missing values for a predetermined period of time or more, the computing device 100 may exclude from the valid data 40 the electrocardiogram data in which one or more of the N leads have missing values for a predetermined period of time or more. That is, the computing device 100 may exclude from the valid data 40 data in which one or more of the N leads are missing.

[0052] The computing device 100 can use at least one of a method of generating valid data 40 based on a cutoff time 39 and a method of generating valid data 40 based on missing values of the electrocardiogram data 10 according to user input. For example, if a user intends to generate training data with high purity, the computing device 100 can generate valid data 40 using both the method of generating valid data 40 based on a cutoff time 39 and the method of generating valid data 40 based on missing values of the electrocardiogram data 10 according to the user's intended input.

[0053] FIG. 3 is a block diagram illustrating a process of training a deep learning model based on valid data according to an embodiment of the present disclosure.

[0054] 2 and 3, a computing device 100 according to an embodiment of the present disclosure may generate training data 50 by labeling the valid data 40 calculated through FIG. 2. The computing device 100 may generate training data 50 by labeling the valid data 40 based on whether cardiac arrest occurred and the time of cardiac arrest. For example, the computing device 100 may label the valid data 40 for which cardiac arrest occurred within 24 hours after the cutoff time 39 as a cardiac arrest group. The computing device 100 may label the valid data 40 for which cardiac arrest did not occur or the subject did not die within 24 hours after the cutoff time 39 as a normal group. That is, the computing device 100 may not classify data for which the time of cardiac arrest occurred more than 24 hours after the cutoff time 39 into any group to reduce prediction errors of the model during the training process. Meanwhile, the aforementioned value of 24 hours is merely an example, and the computing device 100 may set various corresponding times based on user input.

[0055] The computing device 100 may perform training of the deep learning model 200 based on the training data 50 generated after labeling. In this case, the deep learning model 200 may be trained to calculate the likelihood of cardiac arrest occurring within 24 hours from the time of receiving the electrocardiogram data. For example, the computing device 100 may input the electrocardiogram data included in the training data 50 into the deep learning model 200 and calculate a score indicating the likelihood of cardiac arrest occurring within 24 hours. In this case, the score may represent the likelihood of cardiac arrest occurring within 24 hours as a value between 0 and 100. The computing device 100 may calculate an error by comparing the score calculated by the deep learning model 200 with the score corresponding to the label through a loss function. The computing device 100 may adjust the neural network parameters of the deep learning model 200 in a direction that reduces the error. The computing device 100 may repeat this process and terminate training of the deep learning model 200 if the error satisfies a minimum criterion. Meanwhile, although the learning of the deep learning model 200 in FIG. 3 is described based on supervised learning, due to the neural network structure of the deep learning model 200, the learning of the deep learning model 200 may be performed based on self-supervised learning.

[0056] FIG. 4 is a flowchart illustrating a method for predicting a health condition using an electrocardiogram according to one embodiment of the present disclosure.

[0057] 4, a computing device 100 according to an embodiment of the present disclosure may acquire electrocardiogram data and cardiac arrest data including whether or not cardiac arrest occurred in a subject whose electrocardiogram data was measured and the time of cardiac arrest (S110). If the computing device 100 is a server that performs electrocardiogram interpretation, the computing device 100 may acquire the electrocardiogram data and the corresponding cardiac arrest data through communication with equipment that generates electrocardiogram data and cardiac arrest data. If the computing device 100 is a device that can perform electrocardiogram measurement and recording of cardiac arrest-related information, the computing device 100 may generate electrocardiogram data and cardiac arrest data based on signals and information input through a measurement unit and an input / output unit.

[0058] The computing device 100 can analyze missing values of the electrocardiogram data or cardiac arrest data to extract valid data from the electrocardiogram data (S120). The computing device 100 can analyze the time of cardiac arrest recorded in the cardiac arrest data and clinical reasons for cardiac arrest to determine a cutoff time for filtering the electrocardiogram data. The computing device 100 can then filter the electrocardiogram data of subjects in which cardiac arrest occurred within the cutoff time to extract valid data.

[0059] The computing device 100 can generate learning data by labeling the valid data extracted in step S120 based on whether cardiac arrest occurred and the time of cardiac arrest (S130). The computing device 100 can classify the valid data into data of subjects who have experienced cardiac arrest and data of subjects who have not experienced cardiac arrest or who have not died. The computing device 100 can then label each of the data classified into the two groups.

[0060] 4, the computing device 100 can input the training data into the deep learning model to train the deep learning model so that the deep learning model calculates the likelihood of cardiac arrest. The explanation regarding the training of the deep learning model is substituted for the explanation of FIG. 3 above.

[0061] FIG. 5 is a flowchart illustrating a method for predicting health status using an electrocardiogram according to an alternative embodiment of the present disclosure.

[0062] 5, a computing device 100 according to an embodiment of the present disclosure may acquire electrocardiogram data. If the computing device 100 is a server that performs electrocardiogram interpretation, the computing device 100 may acquire the electrocardiogram data through communication with equipment, such as a wearable, that measures electrocardiogram data. If the computing device 100 is a device capable of measuring electrocardiograms, the computing device 100 may generate electrocardiogram data based on signals and information input through a measurement unit, such as an electrode.

[0063] The computing device 100 can input electrocardiogram data into the pre-trained deep learning model and predict the health condition of the subject whose electrocardiogram data was measured (S220). The computing device 100 can calculate a score representing the likelihood that the subject whose electrocardiogram data was measured will experience cardiac arrest within 24 hours through the deep learning model to which the electrocardiogram data was input. The computing device 100 can then analyze how much the health condition has deteriorated based on the score and generate a report. If the computing device 100 is a server that performs electrocardiogram interpretation, the computing device 100 can share the report through communication with equipment such as a wearable that measures electrocardiogram data. If the computing device 100 has an input / output unit, the computing device 100 can directly output the report through the input / output unit.

[0064] The various embodiments of the present disclosure described above may be combined with additional embodiments and may be modified within the scope that would be understandable to one skilled in the art in light of the above detailed description. The embodiments of the present disclosure are illustrative in all respects and should not be considered limiting. For example, components described as a single type may be implemented in a distributed form, and similarly, components described as distributed may be implemented in a combined form. Therefore, all modifications and variations derived from the meaning, scope, and equivalent concepts of the claims of the present disclosure should be construed as being within the scope of the present disclosure.

Claims

1. 1. A method for predicting a health condition using an electrocardiogram, performed by a computing device including at least one processor, comprising: acquiring electrocardiogram data and cardiac arrest data including whether or not cardiac arrest occurred in the subject for whom the electrocardiogram data was measured and the time of cardiac arrest; analyzing the missing values of the electrocardiogram data or the cardiac arrest data to extract valid data from the electrocardiogram data; and generating training data by labeling the extracted valid data based on whether cardiac arrest occurred and the time of cardiac arrest.

2. The step of analyzing the missing values of the electrocardiogram data or the cardiac arrest data to extract valid data from the electrocardiogram data includes: determining a cutoff time for filtering the electrocardiogram data by analyzing the cardiac arrest occurrence time recorded in the cardiac arrest data and clinical evidence for cardiac arrest; and 2. The method of claim 1, further comprising: filtering the electrocardiogram data of a subject who has experienced cardiac arrest within the cutoff time from the acquired electrocardiogram data to extract valid data.

3. determining a cutoff time for filtering the electrocardiogram data by analyzing a time of cardiac arrest and clinical evidence related to cardiac arrest included in the cardiac arrest data, 3. The method of claim 2, comprising: determining as the cutoff time one of a first candidate time calculated based on clinical judgment of electrocardiogram changes occurring before cardiac arrest occurs, a second candidate time clinically judged as the time when treatment to prevent cardiac arrest is possible when cardiac arrest is predicted, or a third candidate time corresponding to the error between the time of cardiac arrest recorded in the cardiac arrest data and the actual time of cardiac arrest of the subject whose electrocardiogram data was measured.

4. The cutoff time is The method of claim 3 , wherein the first candidate time, the second candidate time, or the third candidate time is determined to be the largest value.

5. generating learning data by labeling the extracted valid data based on whether cardiac arrest occurred and the time of cardiac arrest, 5. The method of claim 4, further comprising the step of labeling data of subjects who have experienced cardiac arrest within a predetermined time period after the cutoff time in the valid data as a cardiac arrest group, and labeling data of subjects who have not experienced cardiac arrest or died within a predetermined time period after the cutoff time in the valid data as a normal group.

6. The step of analyzing the missing values of the electrocardiogram data or the cardiac arrest data to extract valid data from the electrocardiogram data includes:

2. The method according to claim 1, further comprising: excluding, when one or more of the N leads of the electrocardiogram data are missing, the electrocardiogram data in which one or more of the N leads are missing from the valid data.

7. The step of analyzing the missing values of the electrocardiogram data or the cardiac arrest data to extract valid data from the electrocardiogram data includes:

2. The method of claim 1, further comprising: excluding, from the valid data, electrocardiogram data in which a lead in the N leads has missing values for a predetermined period of time or more, if any of the N leads in the electrocardiogram data has missing values for a predetermined period of time or more.

8. 2. The method of claim 1, further comprising: inputting the training data into a deep learning model to train the deep learning model so that the deep learning model calculates the likelihood of cardiac arrest occurring.

9. 1. A method for predicting health status using an electrocardiogram, performed by a computing device including at least one processor, comprising: acquiring first electrocardiogram data; inputting the first electrocardiogram data into a pre-trained deep learning model to predict a health state of a subject whose first electrocardiogram data was measured; Although including The deep learning model is Learning is performed based on learning data generated by performing labeling on valid data extracted from the second electrocardiogram data based on whether cardiac arrest occurred and the time of cardiac arrest, The valid data is A method in which missing values of the second electrocardiogram data or cardiac arrest data including the presence or absence of cardiac arrest and the time of cardiac arrest in the subject for whom the second electrocardiogram data was measured are extracted from the second electrocardiogram data based on the results of analyzing the cardiac arrest data.

10. A prediction result derived by inputting the first electrocardiogram data into the pre-trained deep learning model is a score value indicating a health state of the subject for whom the first electrocardiogram data was measured; The method of claim 9 , wherein the score value closer to a first critical value indicates a healthier state, and the score value closer to a second critical value indicates a less healthy state.

11. A computer program stored in a computer-readable storage medium, the computer program performing an operation for predicting a health condition using an electrocardiogram when executed by one or more processors, The operation is acquiring electrocardiogram data and cardiac arrest data including whether or not cardiac arrest occurred in the subject for whom the electrocardiogram data was measured and the time of cardiac arrest; analyzing the missing values of the electrocardiogram data or the cardiac arrest data to extract valid data from the electrocardiogram data; and generating learning data by labeling the extracted valid data based on whether cardiac arrest occurred and the time of cardiac arrest.

12. 1. A computing device for predicting a health condition using an electrocardiogram, comprising: a processor including at least one core; a memory containing program code executable by the processor; and a network unit for acquiring electrocardiogram data and cardiac arrest data including whether or not cardiac arrest occurred in the subject whose electrocardiogram data was measured and the time of cardiac arrest; Including, The processor: analyzing the missing values of the electrocardiogram data or the cardiac arrest data to extract valid data from the electrocardiogram data; The apparatus generates learning data by labeling the extracted valid data based on whether cardiac arrest occurred and the time of cardiac arrest.

Images