Method and system for classifying electrocardiogram signal

Abstract

According to one aspect of the present invention, provided is a method for classifying an electrocardiogram signal, the method comprising the steps of: converting an electrocardiogram signal into first graph data including a P node corresponding to a P wave, a QRS node corresponding to a QRS complex, and a T node corresponding to a T wave; and outputting the classification result of the electrocardiogram signal by processing the first graph data by using a graph convolutional network (GCN)-based classification model.

Description

Method and system for classifying electrocardiogram signals

[0001] The present invention relates to a method and system for classifying an electrocardiogram signal.

[0002] The development of wearable patch-type electrocardiogram (ECG) devices has made it possible to measure and store ECGs over extended periods. This has led to a demand for fast and accurate analysis of large volumes of ECG data. AI-based ECG analysis technology, which enables this, is helping cardiologists diagnose patients more efficiently.

[0003] These AI-based ECG analysis technologies have developed primarily around convolutional neural networks (CNNs) that take short-term ECG signals as input. This is because most ECG signals are obtained by measuring for 10 seconds according to the standard 12-lead ECG, and this data is used to train CNN-based models. In other words, conventional CNN-based ECG analysis methods have the limitation of being difficult to apply to ECG signals measured continuously over long periods of time.

[0004] Meanwhile, graph representation methods that can describe relationships between specific objects and the Graph Convolutional Network (GCN) technology that trains these methods have recently been gaining attention. While there have been attempts to utilize these GCN-based models in the field of electrocardiogram analysis, the lack of an effective graph representation method for ECG signals has hindered their use at a level that can aid cardiologists in making diagnoses.

[0005] In addition, cardiologists analyze electrocardiogram signals by considering the relationship between beats and / or the relationship between P wave - QRS complex - T wave within a beat. However, conventional CNN-based electrocardiogram analysis methods or conventional GCN-based electrocardiogram analysis methods have limitations in that they cannot analyze electrocardiograms by considering these relationships because they use the entire electrocardiogram signal as input.

[0006] Accordingly, the inventor(s) of the present invention propose a technology for converting an electrocardiogram signal into first graph data including a P node corresponding to a P wave, a QRS node corresponding to a QRS complex, and a T node corresponding to a T wave, processing the first graph data using a GCN-based classification model to output a classification result of the electrocardiogram signal, thereby enabling analysis of the electrocardiogram signal by considering the relationship between beats and / or the relationship between P wave - QRS complex - T wave within a beat.

[0007] <Prior Art Literature>

[0008] <Non-patent literature>

[0009] (Non-patent Document 1) H. Ma and L. Xia, "Atrial Fibrillation Detection Algorithm Based on Graph Convolution Network," in IEEE Access, vol. 11, pp. 67191-67200, 2023

[0010] The purpose of the present invention is to solve all of the problems of the above-mentioned prior art.

[0011] In addition, another object of the present invention is to convert an electrocardiogram signal into first graph data including a P node corresponding to a P wave, a QRS node corresponding to a QRS complex, and a T node corresponding to a T wave, and to output a classification result of an electrocardiogram signal by processing the first graph data using a GCN-based classification model.

[0012] In addition, another object of the present invention is to enable analysis of an electrocardiogram signal by considering the relationship between beats and / or the relationship between P wave - QRS complex - T wave within a beat.

[0013] In addition, another object of the present invention is to enable analysis of electrocardiogram signals on a bit-by-bit basis.

[0014] In addition, another object of the present invention is to enable more efficient diagnosis by analyzing electrocardiogram signals (particularly, long-term electrocardiogram signals measured for a long period of time of 10 seconds or more) to classify a patient's arrhythmia on a beat-by-beat basis and providing the results to a cardiologist.

[0015] A representative configuration of the present invention to achieve the above purpose is as follows.

[0016] According to one aspect of the present invention, a method is provided, including a step of converting an electrocardiogram signal into first graph data including a P node corresponding to a P wave, a QRS node corresponding to a QRS complex, and a T node corresponding to a T wave, and a step of outputting a classification result of the electrocardiogram signal by processing the first graph data using a classification model based on a Graph Convolutional Network (GCN).

[0017] According to another aspect of the present invention, a system is provided, including a graph data management unit that converts an electrocardiogram signal into first graph data including a P node corresponding to a P wave, a QRS node corresponding to a QRS complex, and a T node corresponding to a T wave, and a classification model management unit that processes the first graph data using a classification model based on a GCN (Graph Convolutional Network) to output a classification result of the electrocardiogram signal.

[0018] 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.

[0019] According to the present invention, an electrocardiogram signal is converted into first graph data including a P node corresponding to a P wave, a QRS node corresponding to a QRS complex, and a T node corresponding to a T wave, and the first graph data is processed using a GCN-based classification model, thereby outputting a classification result of the electrocardiogram signal.

[0020] In addition, according to the present invention, it is possible to analyze an electrocardiogram signal by considering the relationship between beats and / or the relationship between P wave - QRS complex - T wave within a beat.

[0021] Additionally, according to the present invention, it is possible to analyze an electrocardiogram signal on a bit-by-bit basis.

[0022] In addition, according to the present invention, by analyzing an electrocardiogram signal (particularly, a long-term electrocardiogram signal measured for a long period of time of 10 seconds or more), a patient's arrhythmia can be classified on a beat-by-beat basis, and the results can be provided to a cardiologist, thereby supporting more efficient diagnosis.

[0023] FIG. 1 is a diagram schematically illustrating the configuration of an entire system for classifying an electrocardiogram signal according to one embodiment of the present invention.

[0024] FIG. 2 is a drawing showing in detail the internal configuration of a signal processing system according to one embodiment of the present invention.

[0025] FIG. 3 and FIG. 4 are diagrams exemplarily showing a process of converting an electrocardiogram signal into first graph data according to one embodiment of the present invention.

[0026] FIG. 5 is a diagram exemplifying the structure of an artificial intelligence-based signal processing model for embedding an electrocardiogram signal according to one embodiment of the present invention.

[0027] FIG. 6 is a diagram visually illustrating a matrix representation of first graph data according to one embodiment of the present invention.

[0028] FIG. 7 is a diagram exemplarily showing the result of converting an electrocardiogram signal into second graph data according to one embodiment of the present invention.

[0029] FIG. 8 is a diagram visually illustrating a matrix representation of second graph data according to one embodiment of the present invention.

[0030] FIG. 9 is a diagram exemplarily showing a process in which pooling is performed based on a QRS node according to one embodiment of the present invention.

[0031] FIG. 10 is a diagram exemplarily showing the structure of a GCN-based classification model according to one embodiment of the present invention.

[0032] FIG. 11 is a diagram exemplarily showing a process for generating augmented graph data according to one embodiment of the present invention.

[0033] FIG. 12 is a diagram exemplarily showing the structure of a GCN-based classification model according to one embodiment of the present invention.

[0034] <Explanation of symbols>

[0035] 100: Communications network

[0036] 200: Signal processing system

[0037] 210: Graph Data Management Department

[0038] 220: Classification Model Management Department

[0039] 230: Communications Department

[0040] 240: Control Unit

[0041] 300: Device

[0042] 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.

[0043] 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.

[0044] In this specification, a beat refers to a unit that distinguishes an electrocardiogram signal, and generally, one beat is composed of a P wave, a QRS complex, and a T wave.

[0045] Composition of the entire system

[0046] FIG. 1 is a diagram schematically illustrating the configuration of an entire system for classifying an electrocardiogram signal according to one embodiment of the present invention.

[0047] As illustrated in FIG. 1, the entire system according to one embodiment of the present invention may include a communication network (100), a signal processing system (200), and a device (300).

[0048] 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.

[0049] 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. As another example, the communication network (100) may be an optical communication network that implements conventional communication methods such as LiFi (Light Fidelity), etc., at least in part.

[0050] Next, the signal processing 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 signal processing system (200) according to one embodiment of the present invention can perform a function of converting an electrocardiogram signal into first graph data including a P node corresponding to a P wave, a QRS node corresponding to a QRS complex, and a T node corresponding to a T wave, and outputting a classification result of the electrocardiogram signal by processing the first graph data using a GCN-based classification model. Meanwhile, the signal processing system (200) may be a digital device equipped with a memory means and a microprocessor to have a computing capability, and may be, for example, a server system operated on the communication network (100).

[0051] The configuration and function of the signal processing system (200) according to the present invention will be described in detail below.

[0052] Next, a device (300) according to one embodiment of the present invention is a digital device that includes a function for communicating after being connected to a signal processing system (200). Any digital device having a memory means, a microprocessor, and a computing capability, such as a smart patch, a smart watch, a smart band, or smart glasses, can be adopted as the device (300) according to the present invention.

[0053] According to one embodiment of the present invention, such a device (300) may be a wearable monitoring device including a sensing means (e.g., a contact electrode, etc.) for measuring a predetermined biosignal (e.g., an electrocardiogram signal) from a human body. Furthermore, such a device (300) may further include a display means for providing a user with various information regarding the measurement of the biosignal.

[0054] In particular, the device (300) may include an application (not shown) that supports a user to receive services from the signal processing system (200). Such an application may be downloaded from the signal processing system (200) or an external application distribution server (not shown). Meanwhile, the nature of such an application may be generally similar to the graph data management unit (210), classification model management unit (220), communication unit (230), and control unit (240) of the signal processing system (200), which will be described later. 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.

[0055] Configuration of signal processing system

[0056] Below, the internal configuration and functions of each component of the signal processing system (200) that performs important functions for implementing the present invention will be examined.

[0057] FIG. 2 is a drawing showing in detail the internal configuration of a signal processing system (200) according to one embodiment of the present invention.

[0058] As illustrated in FIG. 2, a signal processing system (200) according to one embodiment of the present invention may be configured to include a graph data management unit (210), a classification model management unit (220), a communication unit (230), and a control unit (240). According to one embodiment of the present invention, at least some of the graph data management unit (210), the classification model management unit (220), the communication unit (230), and the control unit (240) may be program modules that communicate with an external system (not shown). These program modules may be included in the signal processing 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 be stored in a remote memory device that can communicate with the signal processing 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, which will be described later, according to the present invention.

[0059] Meanwhile, although the signal processing 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 signal processing system (200) may be realized within a device (300) or a server (not shown) or included within an external system (not shown) as needed.

[0060] First, the graph data management unit (210) according to one embodiment of the present invention can perform a function of converting an electrocardiogram signal into first graph data including a P node corresponding to a P wave, a QRS node corresponding to a QRS complex, and a T node corresponding to a T wave.

[0061] Specifically, according to one embodiment of the present invention, the electrocardiogram signal may be measured for a predetermined period of time by the sensing means of the device (300). For example, according to one embodiment of the present invention, a long-term electrocardiogram signal (an electrocardiogram signal measured for a long period of time of 10 seconds or more) measured through a single-lead electrocardiogram test using a patch-type device (300) may correspond to such an electrocardiogram signal.

[0062] The graph data management unit (210) according to one embodiment of the present invention can segment an electrocardiogram signal into a P wave, a QRS complex, and a T wave according to its waveform in order to convert the P wave, the QRS complex, and the T wave into corresponding nodes, respectively. According to one embodiment of the present invention, such segmentation can be performed using rule-based segmentation software, or using an artificial intelligence, for example, a CNN-based model, or can also be performed by an expert (for example, a cardiologist). Of course, the segmentation according to one embodiment of the present invention is not limited to the method described above, and can be variously changed within the scope that can achieve the purpose of the present invention.

[0063] Continuing, the graph data management unit (210) according to one embodiment of the present invention can convert the P wave, QRS complex, and T wave included in the electrocardiogram signal into corresponding nodes after performing the above segmentation, which are referred to herein as P node, QRS node, and T node, respectively. Each of these nodes may include information about the corresponding signal (i.e., P wave, QRS complex, or T wave) or may be associated with such information, and may be included in the first graph (meaning a graph in data structure theory) data.

[0064] According to one embodiment of the present invention, such first graph data may also include edges defined based on the connection relationship between the P node, the QRS node, and the T node. In addition, according to one embodiment of the present invention, a lower weight may be assigned to an edge as the distance between the P node, the QRS node, and the T node becomes greater, and a higher weight may be assigned to an edge as the distance between the P node, the QRS node, and the T node becomes closer. According to one embodiment of the present invention, a higher weight assigned to an edge may mean a higher degree of correlation between nodes (or signals corresponding thereto) connected by the edge, and a lower weight may mean a lower degree of correlation between nodes (or signals corresponding thereto) connected by the edge.

[0065] Figures 3 and 4 are diagrams exemplarily showing a process of converting an electrocardiogram signal into first graph data.

[0066] Referring to FIG. 3, as described above, the graph data management unit (210) according to one embodiment of the present invention can segment an electrocardiogram (ECG) signal into a P wave, a QRS complex, and a T wave according to its waveform.

[0067] Figure 4 shows an example of the result of converting the electrocardiogram signal shown in Figure 3 into first graph data.

[0068] Referring to FIG. 4, the first graph (PQRST Graph) data may include nodes (410p, 410qrs, and 410t) and edges (421, 422, and 423). According to one embodiment of the present invention, the types of nodes may be any one of a P node (410p), a QRS node (410qrs), and a T node (410t), and these nodes may be connected by edges. In addition, according to one embodiment of the present invention, the types of edges may be any one of edges connecting nodes (421, 422, and 423) and edges connecting each node to itself (i.e., self-loop edges; not shown).

[0069] As illustrated in FIG. 4, edges (421) may be connected between adjacent nodes. In addition, in the case of QRS nodes, edges (422) may be connected between adjacent QRS nodes, and edges (423) may also be connected between nearby QRS nodes located within a predetermined distance (or window size). In FIG. 4, edges (423) are connected between QRS nodes with a unit distance (or window size) of 2, but the unit distance is not limited thereto, and may be variously changed within a range that can achieve the purpose of the present invention.

[0070] Meanwhile, the graph data management unit (210) according to one embodiment of the present invention can associate values ​​obtained by embedding signals corresponding to each of the P node, QRS node, and T node with each of the P node, QRS node, and T node using an artificial intelligence-based signal processing model.

[0071] FIG. 5 is a diagram exemplarily showing the structure of an artificial intelligence-based signal processing model that embeds an electrocardiogram signal. As illustrated in FIG. 5, according to one embodiment of the present invention, the signal processing model may follow the structure of an auto-encoder model. According to one embodiment of the present invention, such a signal processing model may be learned by compressing (or embedding) an actual electrocardiogram signal and then restoring it, and the graph data management unit (210) according to one embodiment of the present invention may associate the embedded value (embedded signal) with each node using only the encoder after learning is completed.

[0072] However, the structure or hyperparameters (e.g., number of blocks, input / output dimensions, etc.) of the signal processing model according to one embodiment of the present invention are not limited to those illustrated in FIG. 5, and may be variously changed within a range that can achieve the purpose of the present invention.

[0073] Fig. 6 is a diagram visually illustrating a matrix representation of the first graph data. Specifically, Fig. 6 (a) represents an adjacency matrix of the first graph data, and Fig. 6 (b) represents a feature matrix of the first graph data.

[0074] Referring to (a) of Fig. 6, the adjacency matrix can express whether nodes are connected to each other and the degree of that connection. Here, the degree of connection may be a concept corresponding to the weight assigned to the edge (or the degree of association between connected nodes) described above. According to one embodiment of the present invention, this adjacency matrix may be a symmetric matrix.

[0075] As described above, since each node of the first graph data can be connected to itself by an edge (self-loop edge), the value of the diagonal element of the adjacency matrix (indicating the degree of connection; for example, 1) is the largest, and when two nodes are not connected in the adjacency matrix, the element indicating whether the two nodes are connected and the degree of the connection (for example, the adjacency matrix A[i][j] when node i and node j are not connected) may have the smallest value (for example, 0). In addition, according to one embodiment of the present invention, when two different nodes are connected, the value of the element indicating whether the two nodes are connected and the degree of the connection may be smaller as the distance between the two nodes is farther (for example, a value closer to 0 than 1), and larger as the two nodes are closer (for example, a value closer to 1 than 0). In Fig. 6 (a), the color is displayed darker as the value of the corresponding element is larger, and the color is displayed lighter as the value of the corresponding element is smaller.

[0076] Referring to (b) of Fig. 6, in the feature matrix of the first graph data, each row represents a node, and each column of each row represents a feature value of the node of the corresponding row. Specifically, the first three columns (621) of each row indicate whether the node of the corresponding row corresponds to a P node, a QRS node, or a T node (one-hot encoding), the fourth column (622) of each row indicates the duration of the signal corresponding to the node of the corresponding row (or the width between the start and end (onset - offset) of the signal), and the fifth to last columns (623) of each row may represent values ​​obtained by embedding the signal corresponding to the node of the corresponding row using the signal processing model described above.

[0077] Meanwhile, the graph data management unit (210) according to one embodiment of the present invention can convert first graph data, which is an electrocardiogram signal or a result of converting the electrocardiogram signal, into second graph data that includes a QRS node but does not include a P node or a T node.

[0078] That is, the graph data management unit (210) according to one embodiment of the present invention can generate second graph data including only the QRS node in addition to first graph data including all of the P node, QRS node, and T node.

[0079] FIG. 7 is a diagram exemplarily showing the result of converting the electrocardiogram signal illustrated in FIG. 3 into second graph data according to one embodiment of the present invention.

[0080] Referring to FIG. 7, the second graph (QRS Graph) data may include nodes (710qrs) and edges (721). According to one embodiment of the present invention, all types of nodes included in the second graph data are QRS nodes (410qrs), and each node may be connected by an edge. In addition, according to one embodiment of the present invention, the type of edge included in the second graph data may be any one of an edge (721) connecting adjacent nodes and an edge connecting each node itself (i.e., a self-loop edge; not shown). According to one embodiment of the present invention, unlike the first graph data in which an edge (423) may connect even between close QRS nodes located within a predetermined distance, as illustrated in FIGS. 3 and 7, the second graph data may only include edges connecting adjacent QRS nodes.

[0081] Fig. 8 is a diagram visually illustrating a matrix representation of the second graph data. Specifically, Fig. 8 (a) illustrates an adjacency matrix of the second graph data, and Fig. 8 (b) illustrates a feature matrix of the second graph data. The description of the adjacency matrix and feature matrix illustrated in Fig. 8 is identical to that described above regarding the adjacency matrix and feature matrix illustrated in Fig. 6, and therefore, any redundant description will be omitted.

[0082] Meanwhile, according to one embodiment of the present invention, the second graph data can be used to output the classification result of the electrocardiogram signal, and details thereof will be described later.

[0083] Next, the classification model management unit (220) according to one embodiment of the present invention can perform a function of outputting a classification result of an electrocardiogram signal by processing the first graph data using a GCN-based classification model.

[0084] Specifically, the classification model management unit (220) according to one embodiment of the present invention can classify an electrocardiogram signal in bit units by processing the first graph data using a GCN-based classification model and output the classification result. According to one embodiment of the present invention, one QRS node (QRS complex) or one bit can be classified into any one class of Normal Beat (hereinafter, N bit), Supraventricular Ectopic Beat (hereinafter, S bit), and Ventricular Ectopic Beat (hereinafter, V bit) (since one QRS complex is included in one bit, the classification for the QRS node and the classification for the bit may refer to the same thing). However, the class according to one embodiment of the present invention is not limited to those listed above, and may be variously changed within a scope that can achieve the purpose of the present invention.

[0085] More specifically, the classification model management unit (220) according to one embodiment of the present invention can output the classification result for each QRS node included in the first graph data as the classification result of the electrocardiogram signal.

[0086] Referring to FIG. 4, for example, the classification model management unit (220) according to one embodiment of the present invention can output the classification result for each of the five QRS nodes included in the first graph data as the classification result of the electrocardiogram signal.

[0087] While a conventional artificial neural network-based electrocardiogram analysis method treats the entire electrocardiogram signal, for example, the entire electrocardiogram signal measured for 10 seconds according to a standard 12-lead electrocardiogram, as one input and outputs one classification result, the classification model management unit (220) according to one embodiment of the present invention classifies the electrocardiogram signal by bit unit (or QRS node unit) and allows a cardiologist to refer to the result.

[0088] The classification model management unit (220) according to one embodiment of the present invention can perform pooling based on each QRS node included in the first graph data using a GCN-based classification model so that classification results for each QRS node are output as described above.

[0089] FIG. 9 is a diagram exemplarily showing a process in which pooling is performed based on a QRS node according to one embodiment of the present invention.

[0090] Referring to FIG. 9, the classification model management unit (220) according to one embodiment of the present invention can perform pooling on information associated with surrounding nodes based on each QRS node (or centered on each QRS node) based on the index of the QRS node (910; QRS-Centered Pooling). According to one embodiment of the present invention, such pooling can be performed according to the Weighted Average Pooling technique, but is not limited thereto, and various pooling techniques can be used within the scope that can achieve the purpose of the present invention.

[0091] FIG. 10 is a diagram exemplarily showing the structure of a GCN-based classification model according to one embodiment of the present invention.

[0092] Referring to FIG. 10, the classification model management unit (220) according to one embodiment of the present invention can output a classification result of an electrocardiogram signal by processing the first graph (PQRST Graph) data in the above-described manner using the classification model. The output value (Output) can indicate which of the N bits, S bits, and V bits each QRS node included in the first graph data corresponds to. Meanwhile, since the graph convolution operation applied within the classification model can utilize various known operation methods, a detailed description thereof will be omitted.

[0093] However, the structure or hyperparameters (e.g., number of blocks, input / output dimensions, etc.) of the classification model according to one embodiment of the present invention are not limited to those illustrated in FIG. 10, and may be variously changed within a range that can achieve the purpose of the present invention.

[0094] Meanwhile, according to one embodiment of the present invention, the above classification model may be learned using augmented graph data generated by cutting graph data in which an electrocardiogram signal associated with arrhythmia is converted.

[0095] FIG. 11 is a diagram exemplarily showing a process for generating augmented graph data according to one embodiment of the present invention.

[0096] Specifically, referring to FIG. 11, the graph data (Original Graph) converted from an electrocardiogram signal associated with an arrhythmia may include a P node, a QRS node, and a T node, like the first graph data. Here, the electrocardiogram signal associated with an arrhythmia may mean that at least one of the QRS nodes included in the graph data converted from the corresponding signal is classified into a class associated with an arrhythmia (e.g., the S bit or V bit described above). According to one embodiment of the present invention, when the above graph data is cut, augmented graph data is generated, and since the graph data is associated with an arrhythmia, at least one QRS node included in this augmented graph data may also be classified into a class associated with an arrhythmia.

[0097] According to one embodiment of the present invention, the augmented graph data generated as described above can be used for training the above-described classification model, thereby improving the performance of the classification model. The reason for generating such augmented graph data is that, in general, in electrocardiogram signals (training data) collected for training a classification model, there are overwhelmingly more bits classified into classes not associated with arrhythmia (e.g., the above-described S bit or V bit) than bits classified into classes not associated with arrhythmia (e.g., the above-described N bit), so if a classification model is trained using only the collected electrocardiogram signals, the performance of the model may deteriorate due to class imbalance.

[0098] Here, the important point is that the augmented graph data can be generated in the above-described manner and used for training the classification model because the classification model has a structure capable of outputting the classification result for each QRS node included in the first graph data as the classification result of the electrocardiogram signal. That is, as described above, since the conventional artificial neural network-based electrocardiogram analysis method treats the entire electrocardiogram signal as a single input and outputs a single classification result, it is not possible to generate augmented graph data in the manner according to one embodiment of the present invention and use it for training the model.

[0099] Meanwhile, the classification model management unit (220) according to one embodiment of the present invention can output a classification result of an electrocardiogram signal by further processing the second graph data described above using a GCN-based classification model.

[0100] FIG. 12 is a diagram exemplarily showing the structure of a GCN-based classification model according to one embodiment of the present invention.

[0101] Referring to FIG. 12, the classification model management unit (220) according to one embodiment of the present invention can output a classification result of an electrocardiogram signal by processing the first graph (PQRST Graph) data and the second graph (QRS Graph) data in the above-described manner using the classification model.

[0102] Specifically, the classification model management unit (220) according to one embodiment of the present invention can output a classification result of an electrocardiogram signal by processing the second graph data through a QRS block, concatenating it with the pooling result of the first graph data, and then processing it. As illustrated in Fig. 12, by processing the first graph data and the second graph data together, there is an advantage in that a classification result that considers the importance of the QRS node higher than that of the P node or the T node can be obtained as an output value.

[0103] However, the structure or hyperparameters (e.g., number of blocks, input / output dimensions, etc.) of the classification model according to one embodiment of the present invention are not limited to those illustrated in FIG. 12, and may be variously changed within a range that can achieve the purpose of the present invention.

[0104] Next, the communication unit (230) according to one embodiment of the present invention can perform a function that enables data transmission and reception from / to the graph data management unit (210) and the classification model management unit (220).

[0105] Finally, the control unit (240) according to one embodiment of the present invention can perform a function of controlling the flow of data between the graph data management unit (210), the classification model management unit (220), and the communication unit (230). That is, the control unit (240) according to one embodiment of the present invention can control the flow of data from / to the outside of the signal processing system (200) or the flow of data between each component of the signal processing system (200), thereby controlling the graph data management unit (210), the classification model management unit (220), and the communication unit (230) to perform their respective unique functions.

[0106] The embodiments of the present invention described above may be implemented in the form of program commands that can be executed by 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.

[0107] 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.

[0108] 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 classifying electrocardiogram signals, A step of converting an electrocardiogram signal into first graph data including a P node corresponding to a P wave, a QRS node corresponding to a QRS complex, and a T node corresponding to a T wave, and A step of outputting a classification result of the electrocardiogram signal by processing the first graph data using a classification model based on a GCN (Graph Convolutional Network) method.

2. In paragraph 1, The first graph data includes an edge defined based on the connection relationship between the P node, the QRS node, and the T node, and the edge is given a lower weight as the distance between the P node, the QRS node, and the T node is greater, and the edge is given a higher weight as the distance between the P node, the QRS node, and the T node is closer. method.

3. In paragraph 1, In the above conversion step, a value embedding a signal corresponding to each of the P node, the QRS node, and the T node is associated with each of the P node, the QRS node, and the T node using an artificial intelligence-based signal processing model. method.

4. In paragraph 1, In the above output step, the classification result for each of the QRS nodes included in the first graph data is output as the classification result of the electrocardiogram signal. method.

5. In paragraph 4, In the above output step, the classification result for each QRS node is output by performing pooling based on each QRS node using the classification model. method.

6. In paragraph 1, The above classification model is learned using augmented graph data generated by cutting graph data in which electrocardiogram signals associated with arrhythmia are transformed. method.

7. In paragraph 1, In the above conversion step, the electrocardiogram signal or the first graph data is converted into second graph data including the QRS node but not including the P node and the T node, In the above output step, the classification result of the electrocardiogram signal is output by further processing the second graph data using the classification model. 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 classifying electrocardiogram signals, A graph data management unit that converts an electrocardiogram signal into first graph data including a P node corresponding to a P wave, a QRS node corresponding to a QRS complex, and a T node corresponding to a T wave, and A classification model management unit that processes the first graph data using a classification model based on a GCN (Graph Convolutional Network) and outputs a classification result of the electrocardiogram signal. System.

10. In paragraph 9, The first graph data includes an edge defined based on the connection relationship between the P node, the QRS node, and the T node, and the edge is given a lower weight as the distance between the P node, the QRS node, and the T node is greater, and the edge is given a higher weight as the distance between the P node, the QRS node, and the T node is closer. System.

11. In paragraph 9, The above graph data management unit uses an artificial intelligence-based signal processing model to embed a signal corresponding to each of the P node, the QRS node, and the T node, and associates the value with each of the P node, the QRS node, and the T node. System.

12. In paragraph 9, The above classification model management unit outputs the classification result for each of the QRS nodes included in the first graph data as the classification result of the electrocardiogram signal. System.

13. In paragraph 12, The above classification model management unit performs pooling based on each QRS node using the classification model, thereby outputting classification results for each QRS node. System.

14. In paragraph 9, The above classification model is learned using augmented graph data generated by cutting graph data in which electrocardiogram signals associated with arrhythmia are transformed. System.

15. In paragraph 9, The graph data management unit converts the electrocardiogram signal or the first graph data into second graph data including the QRS node but not including the P node and the T node, The above classification model management unit further processes the second graph data using the classification model to output the classification result of the electrocardiogram signal. System.

Images