Method and system for supporting electrocardiogram signal reading

Abstract

According to one aspect of the present invention, provided is a method for supporting the reading of an electrocardiogram signal of a measurement subject, the method comprising the steps of: extracting first feature information and second feature information from a measured electrocardiogram signal of the measurement subject and a guide electrocardiogram signal, respectively, by using a personalized reading model; and generating a reading result for the measured electrocardiogram signal by referring to information about the correlation between the first feature information and the second feature information, wherein the personalized reading model is a model trained by using the electrocardiogram signal of the measurement subject as training data, and the guide electrocardiogram signal is an already read electrocardiogram signal.

Description

Method and system for supporting electrocardiogram signal interpretation

[0001] The present invention relates to a method and system for assisting in reading an electrocardiogram signal of a subject.

[0002] Recently, there has been an increasing number of attempts to introduce artificial intelligence models to replace or support experienced clinical experts in diagnosing abnormal symptoms by interpreting electrocardiogram signals.

[0003] Accurately interpreting an ECG requires comprehensive consideration of various characteristics, such as the width, height, and shape of each ECG waveform (P wave, QRS complex, T wave, etc.). Utilizing an AI model for ECG interpretation can improve the accuracy and consistency of analysis by training the AI ​​model on the correlation between various waveform characteristics and their underlying pathological implications before interpreting the ECG.

[0004] However, even within the same waveform, subtle differences exist across subjects, and these differences can significantly impact interpretation results. However, existing ECG interpretation models fail to account for these differences as important factors, significantly reducing the accuracy of analysis results.

[0005] The purpose of the present invention is to solve all of the problems of the above-mentioned prior art.

[0006] In addition, another purpose of the present invention is to provide a personalized electrocardiogram reading model that can read electrocardiogram signals by taking into account individual characteristics of each subject by using a unique guide electrocardiogram signal for each subject, and further, can update the guide electrocardiogram signal as needed.

[0007] A representative configuration of the present invention to achieve the above purpose is as follows.

[0008] According to one aspect of the present invention, a method for supporting reading of an electrocardiogram signal of a subject is provided, comprising the steps of extracting first feature information and second feature information from a measured electrocardiogram signal of the subject and a guide electrocardiogram signal of the subject, respectively, using a personalized reading model, and generating a reading result for the measured electrocardiogram signal by referring to information regarding a correlation between the first feature information and the second feature information, wherein the personalized reading model is a model learned by using a pair of a measured electrocardiogram signal and a guide electrocardiogram signal for at least one other subject as learning data, and the guide electrocardiogram signal is an electrocardiogram signal that has already been read.

[0009] According to another aspect of the present invention, a system for supporting reading of an electrocardiogram signal of a subject is provided, comprising: a feature information extraction unit for extracting first feature information and second feature information from a measured electrocardiogram signal of the subject and a guide electrocardiogram signal of the subject, respectively, using a personalized reading model; and a reading result generation unit for generating a reading result for the measured electrocardiogram signal by referring to information regarding a correlation between the first feature information and the second feature information, wherein the personalized reading model is a model learned by using a pair of a measured electrocardiogram signal and a guide electrocardiogram for at least one other subject as learning data, and the guide electrocardiogram signal is an electrocardiogram signal that has already been read.

[0010] 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.

[0011] According to the present invention, it is possible to read an electrocardiogram signal by considering the individual characteristics of each subject using a unique guide electrocardiogram signal for each subject, and further, it is possible to provide a personalized electrocardiogram reading model that can update the guide electrocardiogram signal as needed.

[0012] FIG. 1 is a drawing schematically showing the configuration of an entire system for reading an electrocardiogram waveform of a subject according to one embodiment of the present invention.

[0013] FIG. 2 is a drawing detailing the internal configuration of a support system according to one embodiment of the present invention.

[0014] FIG. 3 is a diagram illustrating a detailed structure of a personalized reading model according to one embodiment of the present invention.

[0015] <Explanation of symbols>

[0016] 100: Communications network

[0017] 200: Support System

[0018] 210: Feature information extraction unit

[0019] 220: Reading result generation unit

[0020] 230: Update Department

[0021] 240: Communications Department

[0022] 250: Control Unit

[0023] 300: Device

[0024] 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 understood 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.

[0025] 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.

[0026] Composition of the entire system

[0027] FIG. 1 is a diagram schematically illustrating the configuration of an entire system for supporting reading of an electrocardiogram signal of a subject according to one embodiment of the present invention.

[0028] 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.

[0029] 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.

[0030] Next, the support system (200) according to one embodiment of the present invention can communicate with the device (300) described later through the communication network (100). In addition, the support system (200) according to one embodiment of the present invention can perform a function of extracting first characteristic information and second characteristic information from the measured electrocardiogram signal of the subject and the guide electrocardiogram signal of the subject, respectively, using a personalized reading model, and generating a reading result for the measured electrocardiogram signal by referring to information regarding the correlation between the first characteristic information and the second characteristic information. Meanwhile, the support 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).

[0031] The configuration and function of the support system (200) according to one embodiment of the present invention will be described in detail below.

[0032] 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 support 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.

[0033] 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 feature information extraction unit (210), reading result generation unit (220), update unit (230), communication unit (240), and control unit (250) of the support 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.

[0034] Composition of the support system

[0035] Below, the internal configuration of the support system (200) that performs important functions for implementing the present invention and the functions of each component will be examined.

[0036] FIG. 2 is a drawing showing in detail the internal configuration of a support system (200) according to one embodiment of the present invention.

[0037] As illustrated in FIG. 2, a support system (200) according to one embodiment of the present invention may include a feature information extraction unit (210), a reading result generation unit (220), an update unit (230), a communication unit (240), and a control unit (250). According to one embodiment of the present invention, at least some of the feature information extraction unit (210), the reading result generation unit (220), the update unit (230), the communication unit (240), and the control unit (250) of the support system (200) may be program modules that communicate with an external system (not shown). These program modules may be included in the support 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 support 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.

[0038] Meanwhile, although the support 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 support system (200) may be realized within a device (300) or a server (not shown) or included within an external system (not shown) as needed.

[0039] First, according to one embodiment of the present invention, the feature information extraction unit (210) can perform a function of extracting first feature information and second feature information from the measured electrocardiogram signal of the subject and the guide electrocardiogram signal of the subject, respectively, using a personalized reading model.

[0040] An electrocardiogram signal according to one embodiment of the present invention refers to data that records the activity current generated in the heart in the form of waves. The training data for a personalized reading model according to one embodiment of the present invention may be data including the aforementioned electrocardiogram signal. Specifically, the data including the electrocardiogram signal used as training data for a personalized reading model according to one embodiment of the present invention may include one or more bit sequences.

[0041] A bit sequence according to one embodiment of the present invention is an electrocardiogram signal representing one heartbeat cycle, and is an electrocardiogram signal including a P wave, a QRS wave (or QRS complex), and a T wave. However, if the subject has an abnormal symptom, a bit sequence that does not include at least one of the P wave, QRS wave, and T wave may also exist.

[0042] The training data (particularly, the measured electrocardiogram signal) of the personalized reading model according to one embodiment of the present invention may be composed of five consecutive bit sequences. Specifically, the personalized reading model according to one embodiment of the present invention may learn features included in five consecutive bit sequences in order to read the bit sequence located in the middle (i.e., the third bit sequence among the bit sequences arranged in time order) among the five consecutive bit sequences.

[0043] According to one embodiment of the present invention, reading an electrocardiogram signal or supporting reading of an electrocardiogram signal may mean extracting various characteristic information included in an electrocardiogram signal and analyzing the same to diagnose whether a subject has a pathological abnormality or to support diagnosis by improving the reading accuracy of an electrocardiogram signal.

[0044] Specifically, a personalized interpretation model according to an embodiment of the present invention can be used to extract features (e.g., shape, amplitude, width of waveform, etc.) and features regarding the RR interval (RRI) of one or more QRS waves (or QRS complexes) included in an electrocardiogram signal measured for a subject, compare the features with corresponding features extracted from a guide electrocardiogram signal of the subject, and diagnose a pathological abnormality based on a difference. More specifically, the personalized interpretation model according to an embodiment of the present invention can be used to diagnose a subject as having a specific disease (or as having a high possibility of developing a specific disease) by linking a specific electrocardiogram signal with a specific abnormality (e.g., diagnosing an arrhythmia based on an irregular RR interval, or diagnosing a ventricular premature beat based on an abnormally wide QRS wave width, etc.) in addition to diagnosing the presence of an abnormality based on a difference derived from the comparison with the features extracted from the guide electrocardiogram signal.

[0045] A feature information extraction unit (210) according to an embodiment of the present invention can extract first feature information from a measured electrocardiogram signal of a subject. The measured electrocardiogram signal according to an embodiment of the present invention may mean an electrocardiogram signal actually measured from one or more entities including the subject. Specifically, (1) "measured electrocardiogram signal of the subject" may mean an electrocardiogram signal measured from the subject who is the target of interpretation according to an embodiment of the present invention as described above, and (2) "measured electrocardiogram signal for another subject" may mean one or more (or multiple) electrocardiogram signals measured from entities other than the subject who is the target of interpretation and used as learning data in learning a personalized interpretation model according to an embodiment of the present invention.

[0046] Continuing, the feature information extraction unit (210) according to one embodiment of the present invention can extract second feature information from the guide electrocardiogram signal of the subject. The guide electrocardiogram signal according to one embodiment of the present invention is a part of the measured electrocardiogram signal of the subject, and may refer to an electrocardiogram signal already read by a reader (manager).

[0047] Specifically, (1) "guide electrocardiogram signal of the subject" may mean an electrocardiogram signal already read by a reader (manager) among electrocardiogram signals measured from the subject who is the subject of reading according to one embodiment of the present invention as described above, and (2) "guide electrocardiogram signal for another subject" may mean one or more (or multiple) electrocardiogram signals used as learning data as electrocardiogram signals already read by a reader (manager) among electrocardiogram signals measured from another entity other than the subject who is the subject of reading as described above in learning a personalized reading model according to one embodiment of the present invention.

[0048] Meanwhile, the meaning of the already read electrocardiogram signal described above may mean at least a part of the measured electrocardiogram signal for which the reading result is generated in advance by the reader (administrator) before reading the measured electrocardiogram signal using the personalized reading model according to one embodiment of the present invention.

[0049] According to one embodiment of the present invention, even when a measured electrocardiogram signal of a (new) subject, which is completely different from the electrocardiogram signal data used in the training process of the model, is input, the personalized reading model can generate a reading result from the measured electrocardiogram signal by taking into account the individual characteristics of the subject, as long as only the reading result (i.e., guide electrocardiogram signal) of a reader (manager) for at least one waveform is given.

[0050] That is, it should be understood that the guide ECG signal according to one embodiment of the present invention also includes a case where, in an ECG signal measured over a long period of time, a reader (manager) reads only one waveform (e.g., one QRS complex included in one specific bit sequence).

[0051] A measured electrocardiogram signal according to one embodiment of the present invention may include the aforementioned guide electrocardiogram signal. Specifically, the guide electrocardiogram signal may refer to a portion of the measured electrocardiogram signal that has already been read by a reader (administrator), and the aforementioned portion may refer to a portion or section including at least one bit sequence.

[0052] Meanwhile, even if it is the same normal electrocardiogram signal, there may be differences in detailed characteristics depending on the individual. For example, even if it is the same normal QRS wave, the detailed shape of the wavelength may be different depending on the individual. Since the support system (200) according to one embodiment of the present invention reads the measured electrocardiogram signal of the subject by using the guide electrocardiogram signal already read by the reader (manager) among the measured electrocardiogram signals of the subject as a reference, it is possible to perform electrocardiogram reading by taking into account even the differences in waveforms depending on the individual (i.e., it is possible to perform personalized electrocardiogram reading), thereby having the effect of maximizing the accuracy of the electrocardiogram reading.

[0053] The first characteristic information according to one embodiment of the present invention may refer to information extracted from a measured electrocardiogram signal of a subject using a personalized reading model according to one embodiment of the present invention. Specifically, the first characteristic information includes, but is not limited to, the presence, shape, amplitude, and width of each waveform (P wave, QRS wave, T wave) included in the measured electrocardiogram signal, and includes all characteristics that can be utilized to diagnose an abnormal condition of a subject by classifying or distinguishing the electrocardiogram signal by characteristic.

[0054] The second characteristic information according to one embodiment of the present invention may refer to information extracted from the guide electrocardiogram signal of the subject using the personalized reading model according to one embodiment of the present invention. Specifically, the second characteristic information includes, but is not limited to, the presence, shape, amplitude, and width of each waveform (P wave, QRS wave, T wave) included in the guide electrocardiogram signal, and includes all characteristics that can be used as a comparison target (i.e., a standard indicating the normal electrocardiogram signal of the subject) in diagnosing an abnormal state of the subject. More specifically, the second characteristic information may include the same type of information as the first characteristic information (for example, corresponding to the first characteristic information including the shape, amplitude, and width of the QRS wave, the second characteristic information also includes the same three characteristics, but only the specific contents of the characteristic information are different, etc.).

[0055] Specifically, the support system (200) according to one embodiment of the present invention can (1) receive a guide electrocardiogram signal of a subject (i.e., a guide electrocardiogram signal of which a personalized reading model according to one embodiment of the present invention has not learned the characteristics), (2) infer (or identify) the personal characteristics of the subject from the guide electrocardiogram signal of the subject using the personalized reading model, and (3) generate a reading result by comparing the guide electrocardiogram signal of the subject and the measured electrocardiogram signal of the subject based on the inferred personal characteristics of the subject.

[0056] A personalized reading model according to one embodiment of the present invention may be a model trained using one or more pairs of measured electrocardiogram signals and guide electrocardiogram signals as training data. Specifically, the personalized reading model according to one embodiment of the present invention may be trained by (1) collecting (long-term) electrocardiogram signals from one or more (or multiple) individuals (i.e., individuals other than the subject of the above-described reading), (2) having a reader (manager) read at least one waveform among multiple waveforms included in each of the collected (long-term) electrocardiogram signals, (3) pairing the above-described (long-term) electrocardiogram signals and the guide electrocardiogram signals generated as a result of the reading in (2) described above, and (4) using the above-described (long-term) electrocardiogram signal and guide electrocardiogram signal pair(s) as training data. By the above-described process, the personalized reading model according to one embodiment of the present invention can simultaneously learn about individual electrocardiogram characteristics and general electrocardiogram patterns, thereby performing readings quickly and efficiently even for electrocardiogram signals measured from (new) subjects who have not been learned before, and can further generate reading results by considering the individual characteristics of the (new) subjects. Therefore, the personalized reading model according to one embodiment of the present invention can be applied to actual clinical settings and used for personalized electrocardiogram signal analysis for each patient, thereby providing reading results with very high accuracy that take individual characteristics into account, and thus has high practicality.

[0057] Continuing, the personalized reading model according to one embodiment of the present invention may be a model that is contrastively learned by using pairs of measured electrocardiogram signals and guide electrocardiogram signals for at least one other subject as learning data. That is, the personalized reading model according to one embodiment of the present invention may be contrastively learned by using one or more pairs (i.e., multiple pairs are also possible) of measured electrocardiogram signals and guide electrocardiogram pairs for other subjects (or collected from other subjects) in addition to the subject being read as learning data.

[0058] Contrastive learning according to one embodiment of the present invention is one of the self-supervised learning methods without a label, and means learning a method of extracting features such that the distances between embedding vectors derived from data with high correlation (or vector similarity) are close to each other in the embedding space, and the distances between embedding vectors derived from data with low correlation (or vector similarity) are far from each other in the embedding space. Accordingly, the feature information extraction unit (210) according to one embodiment of the present invention can classify or cluster signals based on information about the correlation of electrocardiogram signals by using a personalized reading model, and thus can perform a function of classifying normal electrocardiogram signals and signals that are significantly different from normal electrocardiogram signals.

[0059] Next, according to one embodiment of the present invention, the reading result generation unit (220) can perform a function of generating a reading result for a measured electrocardiogram signal by referring to information regarding the correlation between the first characteristic information and the second characteristic information described above by using a personalized reading model.

[0060] Specifically, information on correlation according to one embodiment of the present invention is all information on a state in which first feature information extracted from a measured electrocardiogram signal of a subject and second feature information extracted from a guide electrocardiogram signal of the subject are related to or connected to each other, and may mean commonalities or differences between the first feature information and the second feature information.

[0061] More specifically, information on relevance according to one embodiment of the present invention may mean vector similarity calculated from vectors embedded from first feature information and second feature information extracted by the feature information extraction unit (210).

[0062] Continuing, according to one embodiment of the present invention, the reading result generation unit (220) can generate a reading result including whether there is at least one readable event in the measured electrocardiogram waveform of the subject by using information regarding the correlation between the measured electrocardiogram signal (or first characteristic information) of the subject and the guide electrocardiogram signal (or second characteristic information) of the subject.

[0063] The meaning of a readable event according to one embodiment of the present invention is a concept that includes all physiological phenomena that a reader or a reading system can identify and analyze from an electrocardiogram signal, and should be understood to include all kinds of processed or unprocessed data that can be obtained from an electrocardiogram signal, in addition to whether the electrocardiogram signal (or waveform) is normal or indicates a specific pathological condition (such as arrhythmia).

[0064] Specifically, the readable event described above may be (1) a readable event for one or more waveforms (P wave, QRS wave, T wave), and (2) a readable event for one or more bit sequences.

[0065] More specifically, a readable event according to one embodiment of the present invention includes whether a waveform of a measured electrocardiogram signal is a normal waveform, and determining whether a waveform is a normal waveform means, but is not limited to, determining (1) whether all or part of a bit sequence included in the measured electrocardiogram signal is a waveform (normal, N) due to normal contraction of the heart (or ventricle), or (2) whether a bit sequence indicating that a waveform is not due to normal contraction, for example, a waveform such as a supraventricular premature beat (S) or a ventricular premature beat (V).

[0066] Continuing, according to one embodiment of the present invention, the reading result generation unit (220) may generate the reading result by further referring to third feature information extracted from the measured electrocardiogram signal of the subject, together with information regarding the correlation between the first feature information and the second feature information.

[0067] The third characteristic information according to one embodiment of the present invention may refer to information regarding the RR interval (RR interval, RRI). For example, the third characteristic information may be information regarding the RR interval between five consecutive bit sequences included in the training data of the personalized reading model of the present invention.

[0068] Continuing, this third feature information can be extracted through a feature information extraction unit (210) according to an embodiment of the present invention and a different encoder, and since the third feature information, unlike the first feature information or the second feature information described above, does not have individual differences for each subject or is not large enough to cause a significant difference in the reading result of the electrocardiogram signal, the third feature information is extracted through a separate model (or encoder) that has not gone through a process of learning about individual differences (i.e., a separate model different from the personalized reading model can be used), and even if this is used for reading, the accuracy of reading the electrocardiogram signal does not decrease.

[0069] Subsequently, the reading result generation unit (220) according to one embodiment of the present invention compares only the features of the electrocardiogram waveform by referring to information about the correlation between the first feature information and the second feature information, calculates a correlation score, and fuses the correlation score derived through the above calculation with the third feature information to generate a reading result for the measured electrocardiogram signal, thereby obtaining additional information from the measured electrocardiogram signal of the subject (for example, determining that the subject may have atrial fibrillation based on the fact that there is no abnormality in the QRS complex but the RR interval is irregular), thereby increasing the accuracy of the reading result of the electrocardiogram signal of the subject.

[0070] Next, according to one embodiment of the present invention, the update unit (230) can update the guide electrocardiogram signal of the subject.

[0071] Specifically, the guide ECG signal of the subject may change due to various reasons, including reattachment of the ECG measuring device, the subject's activities (e.g., the subject moving or talking during the measurement), and smoking, drinking, or excessive exercise before the test. Accordingly, the update unit (230) according to one embodiment of the present invention performs a function of updating the guide ECG signal as needed, thereby maintaining the reliability of the guide ECG signal as a comparison standard. Specifically, the update unit (230) according to one embodiment of the present invention may (1) update the guide ECG signal by newly determining the guide ECG signal at a predetermined cycle, or (2) update the guide ECG signal in response to the correlation between the measured ECG signal and the guide ECG signal deviating from a predetermined level.

[0072] 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 feature information extraction unit (210), the reading result generation unit (220), and the update unit (230).

[0073] Next, the control unit (250) according to one embodiment of the present invention can perform a function of controlling the flow of data between the feature information extraction unit (210), the reading result generation unit (220), the update 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 support system (200) or the flow of data between each component of the support system (200), thereby controlling the feature information extraction unit (210), the reading result generation unit (220), the update unit (230), and the communication unit (240) to perform their own functions.

[0074] Finally, a personalized reading model according to one embodiment of the present invention may include (1) a first encoder that extracts first feature information and second feature information from a measured electrocardiogram signal of a subject and a guide electrocardiogram signal of the subject, respectively, (2) a second encoder that extracts third feature information from the measured electrocardiogram signal of the subject, and (3) an analysis unit (not shown) that comprehensively interprets the extracted first to third feature information (e.g., calculates a correlation (or correlation score) of the first to third feature information and classifies the electrocardiogram signal based on the calculated correlation (or correlation score)) to derive a reading result (e.g., a diagnostic result such as that the measured electrocardiogram signal has an abnormality).

[0075] According to one embodiment of the present invention, the first encoder and the second encoder described above may each mean different types of neural network layers, and each encoder may transmit the extracted first to third feature information to the feature information extraction unit (210) according to one embodiment of the present invention. In addition, according to one embodiment of the present invention, the above-mentioned analysis unit may transmit the derived reading result to the reading result generation unit (220) according to one embodiment of the present invention.

[0076] However, the structure of the personalized reading model described above is exemplary and is not limited thereto, and may be modified in any way within the scope necessary to achieve the purpose of the present invention.

[0077] More specifically, the structure of a personalized reading model according to one embodiment of the present invention may be as illustrated in FIG. 3. However, the structure of the personalized reading model illustrated in FIG. 3 is merely an example, and it should be understood that the structure of the personalized reading model according to one embodiment of the present invention is not limited to the contents described below.

[0078] First, a personalized reading model according to an embodiment of the present invention may include an encoder (400) capable of extracting features from an electrocardiogram waveform. This encoder (400) may be a model that has completed learning to extract features of waveforms included in an electrocardiogram signal by using a plurality of pairs of measured electrocardiogram signals and guide electrocardiogram signals as input data. Specifically, since the encoder (400) has been learned to classify a plurality of waveforms included in an electrocardiogram signal based on their similarity to each other through a contrastive learning method, it can identify the characteristics of an electrocardiogram signal unique to each individual from an unknown guide electrocardiogram signal, and can read an electrocardiogram signal by classifying whether the input unknown measured electrocardiogram signals are normal electrocardiogram waveforms or electrocardiogram waveforms showing abnormal symptoms.

[0079] Meanwhile, in relation to contrastive learning, it is only one example that multiple waveforms are learned to be classified based on the similarity between them. It should be understood that the basis for the above-described classification is not limited to the similarity between multiple waveforms, but includes all correlations between multiple waveforms.

[0080] Continuing, appropriate feature information can be extracted from the electrocardiogram signal using the encoder (400) described above so as to enable high-accuracy classification of the electrocardiogram signal. Specifically, the feature information extracted from the measured electrocardiogram signal of the subject through the encoder (400) described above may be first feature information, and the feature information extracted from the guide electrocardiogram signal of the subject may be second feature information. Accordingly, the first feature information extraction encoder (400) and the second feature information extraction encoder (400) illustrated in FIG. 3 may be the same encoder.

[0081] Continuing, the similarity between the extracted first feature information and the second feature information can be calculated (420), and a similarity score can be obtained based on this. Specifically, this similarity score can be a value calculated using the vector similarity of the two vectors, corresponding to the first feature information and the second feature information being extracted as vectors.

[0082] Continuing, third feature information is extracted from the measured electrocardiogram signal using a third feature information extraction encoder (410), and the extracted third feature information can be used as input data for reading the electrocardiogram signal together with the similarity score between the first feature information and the second feature information. Meanwhile, since the third feature information does not differ in its characteristics from person to person (or, even if there is a slight difference, the difference is not large enough to affect the electrocardiogram signal reading result), the process of comparing it with feature information extracted from a separate guide electrocardiogram signal can be omitted, and the third feature information can be extracted using a third feature information extraction encoder (410) different from the encoder (400) used to extract the first feature information or the second feature information described above.

[0083] Meanwhile, according to one embodiment of the present invention, the encoder (400) and the third feature information extraction encoder (410) may include a convolutional neural network as a neural network layer. Specifically, the encoder (400) used to extract the first feature information or the second feature information may include a residual neural network (ResNet) as a neural network layer, and the third feature information extraction encoder may include a DenseNet as a neural network layer.

[0084] Continuing, the third feature information can be synthesized (430) along with the first feature information, second feature information, and similarity score above, and used to derive the final interpretation results for the measured electrocardiogram waveform. Specifically, by processing the first feature information, second feature information, their similarity scores, and third feature information using an attention mechanism, the important portions of the aforementioned multiple pieces of information can be highlighted, thereby improving the model's prediction accuracy.

[0085] Continuing, the input value processed through the above attention mechanism can be converted into a one-dimensional vector through a subsequent processing step, and by inputting this into the pre-connection layer (440), the learned weights and biases can be used to derive the final classification result (e.g., the reading result for the measured electrocardiogram signal of the subject).

[0086] 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.

[0087] 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.

[0088] 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 supporting the interpretation of the electrocardiogram signal of a subject, A step of extracting first feature information and second feature information from the measured electrocardiogram signal of the subject and the guide electrocardiogram signal of the subject, respectively, using a personalized reading model, and A step of generating a reading result for the measured electrocardiogram signal by referring to information regarding a correlation between the first characteristic information and the second characteristic information, The above personalized reading model is a model that is learned by using pairs of measured electrocardiogram signals and guide electrocardiogram signals for at least one other subject as learning data, The above guide ECG signal is an ECG signal that has already been read. method.

2. In paragraph 1, The above personalized reading model is a model that is contrastively learned using pairs of measured electrocardiogram signals and guide electrocardiogram signals for at least one other subject as learning data. method.

3. In paragraph 1, The above reading result relates to at least one event that can be read from the measured electrocardiogram signal of the subject. method.

4. In paragraph 1, In the step of generating the reading result, the reading result is generated by further referencing third feature information extracted from the measured electrocardiogram signal of the subject, together with information regarding the correlation between the first feature information and the second feature information. method.

5. In paragraph 1, Further comprising a step of updating the guide electrocardiogram signal of the subject. method.

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

7. A system for supporting the interpretation of the electrocardiogram signal of the subject, A feature information extraction unit that extracts first feature information and second feature information from the subject's measured electrocardiogram signal and the subject's guide electrocardiogram signal, respectively, using a personalized reading model, and Including a reading result generation unit that generates a reading result for the measured electrocardiogram signal by referring to information regarding a correlation between the first characteristic information and the second characteristic information, The above personalized reading model is a model that is learned by using pairs of measured electrocardiogram signals and guide electrocardiogram signals for at least one other subject as learning data, The above guide ECG signal is an ECG signal that has already been read. System.

8. In paragraph 7, The above personalized reading model is a model that is contrastively learned using pairs of measured electrocardiogram signals and guide electrocardiogram signals for at least one other subject as learning data. System.

9. In paragraph 7, The above reading result relates to at least one event that can be read from the measured electrocardiogram signal of the subject. System.

10. In paragraph 7, The above reading result generation unit generates a reading result by further referencing third feature information extracted from the measured electrocardiogram signal of the subject, along with information regarding the correlation between the first feature information and the second feature information. System.

11. In paragraph 7, Further comprising an update unit for updating the guide electrocardiogram signal of the subject. System.

Images