Method and system for predicting occurrence of abnormal waveform in electrocardiogram

Abstract

According to an aspect of the present invention, provided is a method for predicting the occurrence of an abnormal waveform in an electrocardiogram signal of a subject, the method comprising the steps of: acquiring an electrocardiogram signal at a first time point from the subject; inputting the acquired electrocardiogram signal at the first time point into a prediction model so as to analyze same; and determining, by using a result of the analysis, information related to an electrocardiogram signal at a second time point, wherein the second time point is a time point that is temporally subsequent to the first time point.

Description

Method and system for predicting the occurrence of abnormal waveforms in electrocardiogram signals

[0001] The present invention relates to a method and system for predicting the occurrence of an abnormal waveform in an electrocardiogram signal.

[0002] Intraventricular conduction abnormality (ICA) is a cardiovascular disease that can be caused by various reasons, including ischemic heart disease and surgery. It is a high-risk disease that carries a high risk of causing a serious decrease in heart rate and can cause sudden falls and cardiac death.

[0003] Meanwhile, intraventricular conduction disturbances can be identified or diagnosed through a wide QRS complex, an abnormal waveform observed in the electrocardiogram (ECG). However, because the severity of the disease is often already high when abnormalities are observed in the ECG signal, early diagnosis is crucial for effective treatment. However, despite the significant risk associated with intraventricular conduction disturbances, there are currently no methods for predicting their onset early.

[0004] The purpose of the present invention is to solve all of the problems of the above-mentioned prior art.

[0005] In addition, another purpose of the present invention is to predict future electrocardiogram signals based on currently measured normal electrocardiogram signals using an artificial intelligence model, thereby determining the possibility of abnormal waveforms appearing, thereby enabling early prediction or diagnosis of cardiovascular diseases.

[0006] A representative configuration of the present invention to achieve the above purpose is as follows.

[0007] According to one aspect of the present invention, a method for predicting the occurrence of an abnormal waveform in an electrocardiogram signal of a subject is provided, comprising the steps of: obtaining an electrocardiogram signal from the subject; analyzing an electrocardiogram signal at a first point in time among the obtained electrocardiogram signals by inputting it into a prediction model; and determining a possibility of an abnormal waveform appearing in an electrocardiogram signal at a second point in time using the analysis result, wherein the second point in time is a point in time that lags behind the first point in time.

[0008] According to another aspect of the present invention, a system for predicting the occurrence of an abnormal waveform in an electrocardiogram signal of a subject is provided, comprising: an acquisition unit for acquiring an electrocardiogram signal from the subject; an analysis unit for inputting an electrocardiogram signal at a first point in time among the acquired electrocardiogram signals into a prediction model and analyzing it; and a determination unit for determining the possibility of an abnormal waveform appearing in an electrocardiogram signal at a second point in time using the analysis result, wherein the second point in time is a point in time that lags behind the first point in time.

[0009] In addition, a non-transitory computer-readable recording medium recording another method for implementing the present invention, another system, and a computer program for executing the method are further provided.

[0010] According to the present invention, an artificial intelligence model is used to predict future electrocardiogram signals based on currently measured normal electrocardiogram signals, thereby determining the possibility of abnormal waveforms appearing, thereby enabling early prediction or diagnosis of cardiovascular disease.

[0011] FIG. 1 is a diagram schematically illustrating the configuration of an entire system for predicting the occurrence of an abnormal waveform in an electrocardiogram signal according to one embodiment of the present invention.

[0012] FIG. 2 is a drawing showing in detail the internal configuration of a prediction system according to one embodiment of the present invention.

[0013] FIG. 3 is a diagram illustrating one of the prediction models and its training process according to one embodiment of the present invention.

[0014] <Explanation of symbols>

[0015] 100: Communications network

[0016] 200: Prediction System

[0017] 210: Acquisition Department

[0018] 220: Analysis Department

[0019] 230: Decision

[0020] 240: Communications Department

[0021] 250: Control Unit

[0022] 300: Device

[0023] The following detailed description of the present invention refers to the accompanying drawings, which illustrate specific embodiments in which the present invention may be practiced. These embodiments are described in sufficient detail to enable those skilled in the art to practice the present invention. It should be understood that the various embodiments of the present invention, while different from each other, are not necessarily mutually exclusive. For example, specific shapes, structures, and characteristics described herein may be modified and implemented from one embodiment to another without departing from the spirit and scope of the present invention. Furthermore, it should be understood that the positions or arrangements of individual components within each embodiment may also be modified without departing from the spirit and scope of the present invention. Accordingly, the following detailed description is not to be taken in a limiting sense, and the scope of the present invention is to be construed to encompass the scope of the claims and all equivalents thereof. Like reference numerals in the drawings represent the same or similar elements throughout the several aspects.

[0024] Hereinafter, various preferred embodiments of the present invention will be described in detail with reference to the attached drawings so that a person having ordinary skill in the art to which the present invention pertains can easily practice the present invention.

[0025] Composition of the entire system

[0026] FIG. 1 is a diagram schematically illustrating the configuration of an entire system for predicting the occurrence of an abnormal waveform in an electrocardiogram signal according to one embodiment of the present invention.

[0027] First, the communication network (100) according to one embodiment of the present invention can be configured regardless of the communication mode such as wired communication or wireless communication, and can be configured with various communication networks such as a local area network (LAN), a metropolitan area network (MAN), and a wide area network (WAN). Preferably, the communication network (100) referred to herein may be the well-known Internet or the World Wide Web (WWW). However, the communication network (100) is not necessarily limited thereto, and may include at least a portion of a well-known wired or wireless data communication network, a well-known telephone network, or a well-known wired or wireless television communication network.

[0028] For example, the communication network (100) may be a wireless data communication network that implements conventional communication methods such as WiFi communication, WiFi-Direct communication, Long Term Evolution (LTE) communication, 5G communication, Bluetooth communication (including Bluetooth Low Energy (BLE) communication), infrared communication, ultrasonic communication, etc., at least in part.

[0029] Next, the prediction system (200) according to one embodiment of the present invention can perform communication with the device (300) described later through the communication network (100). In addition, the prediction system (200) according to one embodiment of the present invention can perform a function of obtaining an electrocardiogram signal from a subject, inputting the electrocardiogram signal of the first time point among the obtained electrocardiogram signals into a prediction model and analyzing it, and determining the possibility of an abnormal waveform appearing in the electrocardiogram signal of the second time point using the analysis result, and wherein the second time point described above is a time point that lags behind the first time point in time. Meanwhile, such a prediction system (200) may be a digital device having a memory means and a microprocessor equipped with a computing capability, and may be, for example, a server system operated on the communication network (100).

[0030] The configuration and function of the prediction system (200) according to one embodiment of the present invention will be described in detail below.

[0031] Next, a device (300) according to one embodiment of the present invention is a digital device that includes a function for communicating after connecting to a prediction system (200), and any digital device that has a memory means, a microprocessor, and a computing capability, such as a smart phone, a tablet, a smart watch, a smart band, smart glasses, a desktop computer, a notebook computer, a workstation, a PDA, a web pad, a mobile phone, etc., can be adopted as the device (300) according to the present invention.

[0032] In addition, according to one embodiment of the present invention, the device (300) may further include an application program for performing a function according to the present invention. Such an application may exist in the form of a program module within the device (300). The nature of such a program module may be generally similar to the acquisition unit (210), analysis unit (220), decision unit (230), communication unit (240), and control unit (250) of the prediction system (200) described below. Here, at least a part of the application may be replaced with a hardware device or firmware device that can perform functions substantially identical to or equivalent thereto, as necessary.

[0033] Configuration of the prediction system

[0034] Below, the internal configuration and functions of each component of the prediction system (200) that performs important functions for implementing the present invention will be examined.

[0035] FIG. 2 is a drawing showing in detail the internal configuration of a prediction system (200) according to one embodiment of the present invention.

[0036] As illustrated in FIG. 2, a prediction system (200) according to one embodiment of the present invention may include an acquisition unit (210), an analysis unit (220), a decision unit (230), a communication unit (240), and a control unit (250). According to one embodiment of the present invention, at least some of the acquisition unit (210), the analysis unit (220), the decision unit (230), the communication unit (240), and the control unit (250) of the prediction system (200) may be program modules that communicate with an external system (not shown). These program modules may be included in the prediction system (200) in the form of an operating system, an application program module, or other program modules, and may be physically stored in various known memory devices. In addition, these program modules may also be stored in a remote memory device that can communicate with the prediction system (200). Meanwhile, these program modules include, but are not limited to, routines, subroutines, programs, objects, components, data structures, etc. that perform specific tasks or execute specific abstract data types, as described later in accordance with the present invention.

[0037] Meanwhile, although the prediction system (200) has been described as above, this description is exemplary, and it is obvious to those skilled in the art that at least some of the components or functions of the prediction system (200) may be realized within a device (300) or a server (not shown) or included within an external system (not shown) as needed.

[0038] According to one embodiment of the present invention, the acquisition unit (210) can perform a function of acquiring an electrocardiogram signal from a subject.

[0039] An electrocardiogram (ECG) signal according to one embodiment of the present invention may refer to a signal in the form of a graph recording the electrical activity of the heart. Specifically, the ECG signal may comprise a plurality of continuous waves, each wave comprising a plurality of primary waveforms as constituent elements. For example, the plurality of primary waveforms described above may include P waves, QRS complexes, T waves, etc., and may further include U waves, which are rare.

[0040] According to one embodiment of the present invention, the subject of measurement may refer to a target entity for determining the possibility of occurrence of an abnormal waveform (or predicting the occurrence of an abnormal waveform) using the prediction system (200) according to one embodiment of the present invention. Meanwhile, the electrocardiogram signal measured from the subject of measurement described above is different from the electrocardiogram signal used as learning data (or input data) of the prediction model included in the prediction system (200) according to one embodiment of the present invention (i.e., the electrocardiogram signal measured from an entity other than the subject of measurement), and thus must be distinguished from each other.

[0041] Next, according to one embodiment of the present invention, the analysis unit (220) can perform a function of inputting an electrocardiogram signal at a first point in time among the electrocardiogram signals obtained from the subject into a prediction model and analyzing the same.

[0042] A predictive model according to one embodiment of the present invention is a type of artificial intelligence model, and can perform a function of determining the possibility of occurrence of an abnormal waveform in an electrocardiogram signal (or predicting the occurrence of an abnormal waveform). Specifically, the predictive model according to one embodiment of the present invention may refer to an artificial intelligence model that can be used to determine the possibility of occurrence of an abnormal waveform by taking an electrocardiogram signal at a first point in time among electrocardiogram signals obtained from a subject as input data, and predicting and outputting an electrocardiogram signal at a second point in time, which is a point in time later than the first point in time, as input data. The predictive model according to one embodiment of the present invention includes the first to fourth models described below, but is not limited thereto, and any type of artificial intelligence model can be used as long as it is consistent with the purpose of the present invention.

[0043] According to one embodiment of the present invention, the analysis unit (220) may analyze the electrocardiogram signal at the first time point by using the electrocardiogram signal at the first time point as input data among the electrocardiogram signals obtained from the subject, extracting various features therefrom, and using a prediction model that has completed learning about the correlation between the extracted features and information related to the electrocardiogram signal at the second time point, predicting (analyzing) the electrocardiogram signal at the second time point that has not yet arrived (or information related to the QRS complex at the second time point), and determining whether an abnormal waveform appears in the electrocardiogram signal at the second time point based on the prediction result (analysis result).

[0044] Specifically, according to one embodiment of the present invention, the analysis unit (220) may input an electrocardiogram signal at a first time point among electrocardiogram signals obtained from a subject into a prediction model and analyze it, and may cause the prediction model to generate information related to an electrocardiogram signal at a second time point or information related to a QRS complex at the second time point as output data.

[0045] According to one embodiment of the present invention, information related to a QRS complex may refer to all kinds of information (or characteristics) about a QRS complex included in an electrocardiogram signal (i.e., one of several waveforms appearing in the electrocardiogram signal). Specifically, information related to a QRS complex may include a QRS complex length (or QRS complex width, QRS complex width), a QRS voltage, a QRS axis, a Q wave width, a Q wave depth, information related to an R wave, information related to an S wave, and the like. Specifically, information related to a QRS complex at a first time point may refer to the above-described information at the first time point, and information related to a QRS complex at a second time point may refer to the above-described information at the second time point.

[0046] Continuing, information related to the QRS complex at the second time point may mean (a) information related to the QRS complex at the second time point (particularly, the QRS complex length) in the learning data, and (b) information related to the QRS complex included in the electrocardiogram signal predicted to occur at the second time point by the prediction model in the input data of the inference process (particularly, the QRS complex length).

[0047] Next, according to one embodiment of the present invention, the decision unit (230) can perform a function of inputting an electrocardiogram signal at a first time point into a prediction model and using the analyzed analysis result to determine the possibility of an abnormal waveform appearing in an electrocardiogram signal at a second time point.

[0048] The analysis result according to one embodiment of the present invention should be understood as a concept including all types of information related to an electrocardiogram signal or electrocardiogram waveform that is predicted to occur (or is likely to occur) at a second time point. Specifically, the information related to the above-described electrocardiogram signal may be an electrocardiogram signal (i.e., electrocardiogram waveform data) for a predetermined interval. Accordingly, if the information related to the electrocardiogram signal is electrocardiogram waveform data, all types of information that can be extracted from the electrocardiogram waveform data (i.e., waveform characteristics such as the length, height, and shape of the waveform, etc.) may also be included in the information related to the electrocardiogram signal.

[0049] For example, the analysis result described above may be an electrocardiogram signal and a waveform included therein (i.e., electrocardiogram waveform data at the second time point) for a predetermined interval predicted to occur at the second time point.

[0050] Meanwhile, the above-described predetermined interval may be defined in various ways as long as it does not conflict with the purpose of the present invention. For example, it may be defined as a time interval, such as 'a time of about 10 seconds', or it may be defined as the number of beats, such as '3 beats (i.e., when a beat is a cycle of a heartbeat including a P wave, a QRS complex, and a T wave, 3 beats are defined as a predetermined interval)'.

[0051] Specifically, the information related to the electrocardiogram signal may be electrocardiogram signal data measured by a 12-lead method, and the three bits included in the electrocardiogram signal during the above-described predetermined interval may mean waveforms measured in different leads or waveforms measured in the same lead.

[0052] Meanwhile, the first point in time according to one embodiment of the present invention may mean any point in time when the electrocardiogram signal is normal, and the second point in time may mean a point in time that is later in time than the first point in time described above. That is, the first point in time according to one embodiment of the present invention may be a time section included in an electrocardiogram signal measured from a subject, while the second point in time may be a time section that is later than the first point in time, and may mean a future point in time that has not yet arrived and is not included in the electrocardiogram signal measured from the subject. Specifically, the second point in time may mean a point in time that has passed by up to 31 days compared to the first point in time, and thus, the second point in time may be a section or time specified by the predetermined interval described above within 31 days from the first point in time.

[0053] Continuing, the analysis unit (220) according to one embodiment of the present invention can use the prediction model to take the electrocardiogram signal of the first time point as input data and produce information related to the electrocardiogram signal of the second time point (e.g., the electrocardiogram waveform predicted to occur at the second time point) as an analysis result, and the decision unit (230) according to one embodiment of the present invention can compare the electrocardiogram signal of the first time point described above (or, information related to the QRS complex of the first time point) with the electrocardiogram signal predicted to occur at the second time point (or, information related to the QRS complex of the second time point) to determine the possibility of an abnormal waveform appearing at the second time point (or, whether an abnormal waveform will appear at the second time point).

[0054] According to one embodiment of the present invention, comparing information related to the QRS complex of the second time point with information related to the QRS complex of the first time point may mean comparing information of the same type among information related to the QRS complexes of the first and second time points. Specifically, by comparing the length of the QRS complex of the second time point with the length of the QRS complex of the first time point, it is possible to determine whether the QRS complex of the second time point has a normal (or normal range) length or an abnormal (or abnormal range) length (i.e., whether it corresponds to an abnormal waveform).

[0055] An abnormal waveform according to one embodiment of the present invention may mean a waveform that exhibits at least one abnormality among abnormality of shape (i.e., when the QRS complex is wide or the R wave is abnormally large, etc.), abnormality of time interval (when the interval between electrocardiogram waveforms is abnormally long or short, etc.), abnormality of amplitude (when the amplitude of the waveform is abnormally high or low, etc.), and irregularity of waveform (atrial fibrillation, etc., in which the waveform is not regular or shows an irregular rhythm).

[0056] Specifically, the abnormal waveform according to one embodiment of the present invention may be a 'wide QRS complex', which is an abnormal waveform having an abnormal shape.

[0057] Continuing, a wide QRS complex according to one embodiment of the present invention may refer to a QRS complex that exhibits a wider duration compared to a normal QRS complex. More specifically, a normal QRS complex may have a duration of 0.12 seconds (120 milliseconds) or less, while a wide QRS complex, which is an abnormal waveform, may have a duration greater than 0.12 seconds (120 milliseconds).

[0058] Accordingly, the decision unit (230) according to one embodiment of the present invention can extract information (i.e., length) related to a normal QRS complex from an electrocardiogram signal at a first time point, and compare it with information (i.e., length) related to a (predicted) QRS complex at a second time point generated through a prediction model, thereby determining the possibility of an abnormal waveform appearing at the second time point (i.e., determining whether it is a wide QRS complex). For example, if the length of the QRS complex at the first time point is 0.12 seconds, and the length of the predicted QRS complex at the second time point is greater than 0.12 seconds (e.g., 0.13 seconds or longer), it can be determined that a wide QRS complex, which is an abnormal waveform, will occur at the second time point. In addition, the decision unit (230) according to one embodiment of the present invention may calculate the probability (e.g., percentage probability, etc.) of the appearance of an abnormal waveform at a second time point based on information including the degree to which the length of the predicted QRS complex is large (e.g., the probability that an abnormal waveform will appear is high in proportion to the degree to which the length is large or the frequency with which a waveform with a large length appears).

[0059] Meanwhile, in response to the decision unit (230) according to one embodiment of the present invention determining (or predicting) whether a wide QRS complex will occur at the second time point, the decision unit (230) or an administrator (e.g., medical staff, etc.) can predict (or diagnose) that an intraventricular conduction abnormality (ICA) will occur in the subject, or calculate the possibility (or risk) of an intraventricular conduction abnormality occurring in the subject in the future based on calculating the possibility of a wide QRS complex occurring at the second time point.

[0060] Intraventricular conduction abnormality (ICA), according to one embodiment of the present invention, may refer to a condition or disease in which the electrical signals of the heart are not conducted normally within the ventricles, resulting in delayed or abnormal ventricular contraction. Such intraventricular conduction abnormality can be diagnosed by the appearance of an abnormal QRS complex (e.g., a widened or abnormally shaped QRS complex) in an electrocardiogram signal.

[0061] A prediction model according to one embodiment of the present invention may be any one of a first model based on an autoencoder, a second model based on a U-Net, a third model including a pre-trained encoder and a feedforward neural network (FNN), and a fourth model based on a convolutional neural network (CNN).

[0062] The first model according to one embodiment of the present invention may refer to one of the model structures that can be used as a predictive model according to one embodiment of the present invention. Specifically, the first model may refer to a model based on an autoencoder, and may refer to a model that uses an electrocardiogram signal at a first time point as input data and generates an electrocardiogram signal expected to occur at a second time point as output data based on the encoder and decoder of the autoencoder.

[0063] Continuing, an autoencoder according to an embodiment of the present invention is a type of artificial neural network that operates in an unsupervised learning manner, and may refer to an artificial neural network that detects important features of data (i.e., latent vectors that express high-dimensional data well) by learning a method to compress input data to express it in a low-dimensional space and then restore it back to the original data (i.e., generate output data similar to the input data). The encoder included in the autoencoder may perform the function of compressing input data into a low-dimensional latent space, and the decoder may perform the function of restoring the low-dimensional representation obtained from the encoder back to the original high-dimensional data. Meanwhile, the encoding and decoding parts of the autoencoder may have a convolutional neural network (CNN) structure, and such an autoencoder may be referred to as a convolutional autoencoder (or convolutional autoencoder). The above-described convolutional autoencoder may also be used as an autoencoder according to an embodiment of the present invention.

[0064] The second model according to one embodiment of the present invention may refer to one of the model structures that can be used as a predictive model according to one embodiment of the present invention. Specifically, the second model may refer to a model based on U-Net, and based on the encoder and decoder of U-Net, a model that takes an electrocardiogram signal at a first time point as input data and generates an electrocardiogram signal expected to occur at a second time point as output data.

[0065] Continuing, the U-Net according to one embodiment of the present invention may mean an artificial neural network structure based on a convolutional neural network (CNN) designed for image segmentation, and in particular, may mean a model optimized for predicting the class (background or object) to which each pixel of an input image belongs. The U-Net is a model having a symmetrical "U" shape, and may include an encoding path that receives an input image, gradually reduces spatial information (i.e., reduces the size of a feature map), and extracts abstract features (or, features), and a decoding path that gradually restores the abstract features obtained in the encoding path to the original resolution. In particular, the U-Net has the effect of combining low-resolution features and high-resolution features to restore the original more precisely by concatenating the output (i.e., feature map) of each stage obtained in the encoding path with the corresponding stage in the decoding path (such connection may be referred to as a skip connection).

[0066] A third model according to one embodiment of the present invention may be a model including an autoencoder or U-Net encoder and a feed-forward neural network for which pre-training has been completed, which may mean a model that takes an electrocardiogram signal at a first time point as input data and generates an electrocardiogram signal expected to occur at a second time point as output data.

[0067] Specifically, the third model according to one embodiment of the present invention may refer to one of the model structures that can be used as a predictive model according to one embodiment of the present invention, and as described below, may be generated by pre-training an autoencoder or U-Net, then freezing the weights and adding a feedforward neural network. The specific structure of the third model and the method for obtaining it will be described in more detail in the relevant section below.

[0068] The fourth model according to one embodiment of the present invention may refer to one of the model structures that can be used as a predictive model according to one embodiment of the present invention. Specifically, the fourth model may refer to a model based on a convolutional neural network (CNN), which takes an electrocardiogram signal from a first time point as input data and generates an electrocardiogram signal expected to occur at a second time point as output data.

[0069] Continuing, a convolutional neural network (CNN) according to one embodiment of the present invention is a model mainly designed to process visual data such as images, and is characterized by extracting local features (e.g., edges or patterns) from input data using convolutional layers. The convolutional neural network may include, as components, a convolutional layer that detects a specific pattern (e.g., edges or textures) of an image by applying a filter (or kernel) to the input image, an activation function that nonlinearly transforms the output data of each neuron to improve the expressiveness of the model, a pooling layer that extracts key features while reducing the size of the image, and a fully connected layer that performs the final classification or prediction based on the image features. In addition, the output of each layer may be adjusted to a certain range to stabilize learning and increase the training speed, and the dropout method that randomly inactivates neurons during learning to prevent overfitting may be applied to improve generalization performance.

[0070] The first to fourth models described above can be obtained by a method including a step of training a model to be able to identify a correlation between the electrocardiogram signal measured at the first time point and the electrocardiogram signal measured at the second time point, using electrocardiogram signals measured from an individual other than the subject as learning data, and including the electrocardiogram signal measured at the first time point and the electrocardiogram signal measured at the second time point.

[0071] Specifically, all of the above-described first to fourth models can be obtained by completing learning about the correlation between the electrocardiogram signal at a first time point and the electrocardiogram signal at a second time point that is temporally later than the first time point, using electrocardiogram signals measured from an individual other than the subject of the measurement.

[0072] Meanwhile, the subject of measurement may refer to an entity that only has an electrocardiogram signal measured at the first point in time, but has not yet reached the second point in time and thus has no measured electrocardiogram signal, and is the target entity for determining whether an abnormal waveform will appear at the second point in time.

[0073] Continuing, the first to fourth models, for which learning has been completed, can use the measured electrocardiogram signal of the subject at the first point in time as input data to predict the electrocardiogram signal of the second point in time that has not yet arrived. In other words, by utilizing the prediction system (200) according to one embodiment of the present invention, it is possible to predict electrocardiogram signals that will occur in the future before they occur, and through this, it is also possible to predict whether abnormal waveforms will appear in the future.

[0074] Meanwhile, a third model according to an embodiment of the present invention can be obtained by a method including a step of pre-training an encoder of an autoencoder or a U-Net using an electrocardiogram signal including an electrocardiogram signal measured at a first time point and an electrocardiogram signal measured at a second time point as learning data, which is an electrocardiogram signal measured from an object other than the subject of the measurement, a step of freezing the weights of an encoder for which pre-training has been completed, and a step of adding a feed-forward neural network to the fixed encoder.

[0075] Specifically, a third model according to an embodiment of the present invention can be obtained through a method including (a) a pre-learning and weight fixing step for reconstructing an electrocardiogram signal at a second time point from an electrocardiogram signal at a first time point using an autoencoder or a U-Net, and training the model to reduce loss between the reconstructed electrocardiogram signal at the second time point (i.e., a predicted value) and the actual electrocardiogram signal at the second time point (i.e., an actual value), thereby enabling the model to effectively extract inherent features included in the electrocardiogram signal at the first time point, and (b) a step for training a feedforward neural network model to predict an electrocardiogram signal at the second time point from the electrocardiogram signal at the first time point using a feature extraction part (i.e., an encoder) of the pre-learned autoencoder or U-Net obtained through the above-described pre-learning and weight fixing step.

[0076] According to one embodiment of the present invention, "freezing" the encoder weights may mean fixing the weights (parameters) of the encoder portion of the neural network so that they are not updated during the training process. This allows training of only other portions (e.g., the feedforward neural network portion of the third model described above) while maintaining the encoder weights unchanged. By fixing the encoder weights, the number of parameters to be trained is reduced, making training faster and more efficient, and also reducing the risk of overfitting.

[0077] A feedforward neural network according to one embodiment of the present invention is an artificial neural network that processes input data by transmitting it in one direction, and may refer to a neural network having a structure in which data starts from an input layer and is sequentially transmitted to an output layer through a plurality of hidden layers. Specifically, the input layer may perform the function of receiving input data as the first layer of the neural network, the hidden layer may perform the function of improving the learning ability and expressiveness of the neural network by nonlinearly transforming the input data through an activation function, and the output layer may perform the function of receiving data processed in the hidden layer and finally outputting a result.

[0078] More specifically, the prediction model of the present invention can be obtained through a structure and training process as illustrated in FIG. 3. However, the structure of the prediction model illustrated in FIG. 3 is merely an example, and it should be understood that the first to fourth models described above or the prediction model according to one embodiment of the present invention are not limited by the contents of FIG. 3.

[0079] Referring to FIG. 3, a prediction model according to one embodiment of the present invention can proceed with learning by using an electrocardiogram signal (410) at a first time point as input data. The electrocardiogram signal (410) at a first time point is data composed of a waveform representing a normal electrocardiogram signal, and may have a predetermined time or a predetermined number of bits. For example, as illustrated in FIG. 3, it may have three bits.

[0080] Continuing, the prediction model according to one embodiment of the present invention can generate an electrocardiogram signal (420) predicted to occur at a second point in time, compare it with an electrocardiogram signal actually occurring at the second point in time, calculate a difference (loss), and proceed with learning in the direction of reducing the calculated difference.

[0081] Training a model in the direction of reducing the above-described difference may mean training a model through a method including, as illustrated in FIG. 3, (a) a step of inputting an ECG signal at a first time point into a U-Net model (430) including an encoder and a decoder, generating (reconstructing) an ECG signal at a second time point, and comparing it with an actual ECG signal to minimize loss (unsupervised learning), (b) a step of fixing the weights of the encoder after the pre-training is completed, (c) a step of extracting features from the ECG signal at the first time point through an encoder with fixed weights, and (d) a step of inputting the extracted features into a new feedforward neural network (440) to perform supervised learning to produce final output data (i.e., predicting the ECG signal at the second time point, or predicting whether a wide QRS complex occurs at the second time point).

[0082] In conclusion, a prediction model according to one embodiment of the present invention can be trained using the learning method illustrated in FIG. 3, and the trained prediction model can use the electrocardiogram signal at the first time point as input data and predict the electrocardiogram signal to be generated at the second time point with high accuracy before the second time point is reached.

[0083] The prediction system (200) according to one embodiment of the present invention may further include a warning unit (not shown) in addition to the acquisition unit (210), analysis unit (220), decision unit (230), communication unit (240), and control unit (250).

[0084] A warning unit (not shown) according to one embodiment of the present invention may perform a function of notifying an administrator of the determined information in response to the determination of the possibility of an abnormal waveform appearing at a second time point (or whether an abnormal waveform will appear at all). For example, a warning unit (not shown) according to one embodiment of the present invention may notify an administrator (e.g., medical staff, etc.) of the occurrence (or probability of occurrence) of a wide QRS complex in response to the prediction that a wide QRS complex will occur at a second time point (or the probability of occurrence is determined to exceed a predetermined standard).

[0085] Next, the communication unit (240) according to one embodiment of the present invention can perform a function that enables data transmission and reception from / to the acquisition unit (210), the analysis unit (220), and the decision unit (230).

[0086] Finally, the control unit (250) according to one embodiment of the present invention can perform a function of controlling the flow of data between the acquisition unit (210), the analysis unit (220), the decision unit (230), and the communication unit (240). That is, the control unit (250) according to one embodiment of the present invention can control the flow of data from / to the outside of the prediction system (200) or the flow of data between each component of the prediction system (200), thereby controlling the acquisition unit (210), the analysis unit (220), the decision unit (230), and the communication unit (240) to perform their own functions.

[0087] The embodiments of the present invention described above may be implemented in the form of program commands that can be executed through various computer components and recorded on a computer-readable recording medium. The computer-readable recording medium may include program commands, data files, data structures, etc., either singly or in combination. The program commands recorded on the computer-readable recording medium may be specially designed and configured for the present invention or may be known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical recording media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, and hardware devices specifically configured to store and execute program commands, such as ROMs, RAMs, and flash memories. Examples of program commands include not only machine language codes generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter, etc. Hardware devices may be changed into one or more software modules to perform processing according to the present invention, and vice versa.

[0088] Although the present invention has been described above with specific details such as specific components and limited examples and drawings, these are provided only to help a more general understanding of the present invention, and the present invention is not limited to the above examples, and those with ordinary knowledge in the technical field to which the present invention pertains can make various modifications and changes based on this description.

[0089] Therefore, the idea of ​​the present invention should not be limited to the embodiments described above, and not only the scope of the patent claims described below but also all scopes equivalent to or equivalently modified from the scope of the patent claims are considered to fall within the scope of the idea of ​​the present invention.

Claims

1. A method for predicting the occurrence of abnormal waveforms in the electrocardiogram signal of a subject, A step of obtaining an electrocardiogram signal from the subject; A step of inputting the first point-in-time ECG signal among the above-mentioned acquired ECG signals into a prediction model and analyzing it, and A step of determining the possibility of an abnormal waveform appearing in an electrocardiogram signal at a second time point using the above analysis results is included. The above second point in time is a point in time later than the first point in time. method.

2. In paragraph 1, In the above analysis step, the prediction model generates information related to the QRS complex at the second time point as output data. method.

3. In paragraph 2, In the above decision step, the information related to the generated QRS complex of the second time point is compared with the information related to the QRS complex of the first time point to determine whether an abnormal waveform appears at the second time point. method.

4. In paragraph 1, The above prediction model is one of the first model based on an autoencoder, the second model based on a U-Net, the third model including a pre-trained encoder and a feedforward neural network (FNN), and the fourth model based on a convolutional neural network (CNN). method.

5. In paragraph 4, The above first to fourth models are: An electrocardiogram signal measured from an object other than the subject of measurement, the electrocardiogram signal including the electrocardiogram signal measured at a first time point and the electrocardiogram signal measured at a second time point, is acquired by a method including a step of learning using the electrocardiogram signal as learning data. method.

6. In paragraph 5, The third model above is, A step of pre-training an encoder of an autoencoder or U-Net using an electrocardiogram signal measured from an object other than the subject, including an electrocardiogram signal measured at a first time point and an electrocardiogram signal measured at a second time point, as learning data; A step of fixing the weights of the encoder for which the above pre-learning has been completed, and Obtained by a method comprising the step of adding a feed forward neural network to the above fixed encoder. method.

7. In paragraph 1, Further comprising a step of notifying the administrator of the determined information in response to the information related to the electrocardiogram signal at the second point in time being determined. method.

8. A non-transitory computer-readable recording medium recording a computer program for executing the method according to paragraph 1.

9. A system for predicting the occurrence of abnormal waveforms in the electrocardiogram signal of a subject, An acquisition unit that acquires an electrocardiogram signal from the subject; An analysis unit that inputs the electrocardiogram signal at the first point in time among the above-mentioned acquired electrocardiogram signals into a prediction model and analyzes it, and A decision unit is included that determines the possibility of an abnormal waveform appearing in an electrocardiogram signal at a second time point using the above analysis results, The above second point in time is a point in time later than the first point in time. System.

10. In paragraph 9, The above analysis unit generates information related to the QRS complex of the second time point as output data from the prediction model. System.

11. In paragraph 10, The above decision unit compares the information related to the generated QRS complex of the second time point with the information related to the QRS complex of the first time point to determine whether an abnormal waveform appears at the second time point. System.

12. In paragraph 9, The above prediction model is one of the first model based on an autoencoder, the second model based on a U-Net, the third model including a pre-trained encoder and a feedforward neural network (FNN), and the fourth model based on a convolutional neural network (CNN). System.

13. In paragraph 12, The above first to fourth models are: An electrocardiogram signal measured from an object other than the subject of measurement, the electrocardiogram signal including the electrocardiogram signal measured at a first time point and the electrocardiogram signal measured at a second time point, is acquired by a method including a step of learning using the electrocardiogram signal as learning data. System.

14. In paragraph 13, The third model above is, A step of pre-training an encoder of an autoencoder or U-Net using an electrocardiogram signal measured from an object other than the subject, including an electrocardiogram signal measured at a first time point and an electrocardiogram signal measured at a second time point, as learning data; A step of fixing the weights of the encoder for which the above pre-learning has been completed, and Obtained by a method comprising the step of adding a feed forward neural network to the above fixed encoder. System.

15. In paragraph 9, In response to the information related to the electrocardiogram signal at the second point in time being determined, a warning unit is further included to notify the administrator of the determined information. System.

Images