Method, program, and device for generating artificial intelligence for diagnosing diseases on basis of electrocardiograms

Abstract

Disclosed are a method, program, and device for generating artificial intelligence for diagnosing diseases on the basis of electrocardiograms, which are performed by a computing device, according to an embodiment of the present disclosure. The method may comprise the steps of: acquiring electrocardiogram data and medical treatment data regarding a subject from whom the electrocardiogram data was measured; and analyzing record information present in the medical treatment data to classify the electrocardiogram data as either training data or non-training data.

Description

Method, program, and device for creating an artificial intelligence that diagnoses diseases based on electrocardiograms

[0001] The present disclosure relates to artificial intelligence technology in the medical field, and more specifically, to a method for configuring learning data necessary for creating an artificial intelligence that diagnoses a disease based on an electrocardiogram, and a method for training the artificial intelligence using such learning data.

[0002]

[0003] An electrocardiogram (ECG) is a medical test that records the electrical activity of the heart and is an essential tool for diagnosing heart disease. It can diagnose various heart conditions, including myocardial infarction and arrhythmia, and its accuracy plays a crucial role in saving patients' lives. However, human experts analyzing ECG data can be prone to errors due to subjective judgment, and interpreting complex data can sometimes be time-consuming. To overcome these limitations, artificial intelligence (AI) technology is increasingly playing a crucial role.

[0004] Recent AI models have demonstrated remarkable performance in the medical field, particularly in the analysis of electrocardiogram data. However, the development of AI models based on real-world clinical data faces numerous challenges. For example, data on specific diseases or conditions can be relatively rare, making it difficult to secure sufficient data for AI models to learn from. Furthermore, while accurately labeling medical data is paramount, this process requires a high level of expertise, inevitably requiring significant time and effort. These challenges are major factors hindering the generalization ability of models and the efficiency of clinical application.

[0005]

[0006] The purpose of this disclosure is to provide a method for constructing customized data reflecting the characteristics of a specific disease and training an artificial intelligence model based on the constructed data.

[0007] However, the problems to be solved in this disclosure are not limited to the problems mentioned above, and other problems not mentioned can be clearly understood based on the description below.

[0008]

[0009] According to one embodiment of the present disclosure, a method for generating artificial intelligence for diagnosing diseases based on electrocardiograms, performed by a computing device to achieve the aforementioned task, is disclosed. The method may include the steps of: acquiring electrocardiogram data and clinical data regarding a subject of measurement of the electrocardiogram data; and analyzing record information present in the clinical data to classify the electrocardiogram data as learning data or non-learning data.

[0010] Alternatively, the step of analyzing the record information present in the medical data to classify the electrocardiogram data into learning data or non-learning data may include the steps of: identifying key-value values ​​corresponding to the record information or text corresponding to the record information; analyzing whether a value matching a preset reference value exists among the identified key-value values ​​or analyzing whether a text matching a preset reference keyword exists among the identified texts; and classifying the electrocardiogram data into learning data or non-learning data depending on whether a value matching the preset reference value or a text matching the preset reference keyword exists.

[0011] Alternatively, the step of classifying the electrocardiogram data into learning data or non-learning data depending on whether there is a value matching the preset reference value or a text matching the preset reference keyword, may be such that if there is a value matching the preset reference value or a text matching the preset reference keyword, the electrocardiogram data is classified as non-learning data, and if there is no value matching the preset reference value or a text matching the preset reference keyword, the electrocardiogram data is classified as learning data.

[0012] Alternatively, the text matching the preset reference value or the preset reference keyword may be a value corresponding to at least one of treatment history information for a specific disease, surgical history information for the specific disease, or medical expert opinion information for the specific disease.

[0013] Alternatively, the findings information may include a statement that may suggest that the subject has undergone a procedure or surgery for a particular condition.

[0014] Alternatively, the method may further include a step of training the neural network model by inputting the electrocardiogram data classified as the training data into the neural network model so that the neural network model analyzes the possibility of occurrence of a specific disease based on the electrocardiogram data.

[0015] Alternatively, the specific condition may be at least one of aortic valve stenosis or left ventricular diastolic dysfunction.

[0016] According to one embodiment of the present disclosure for achieving the aforementioned task, a computer program stored on a computer-readable storage medium is disclosed. When the computer program is executed on one or more processors, it performs operations for generating artificial intelligence that diagnoses a disease based on an electrocardiogram. In this case, the operations may include: an operation for acquiring electrocardiogram data and clinical data regarding a subject of measurement of the electrocardiogram data; and an operation for analyzing record information present in the clinical data to classify the electrocardiogram data into learning data or non-learning data.

[0017] According to one embodiment of the present disclosure for achieving the aforementioned task, a computing device for generating artificial intelligence for diagnosing a disease based on an electrocardiogram is disclosed. The device may include a processor including at least one core; a memory including program codes executable by the processor; and a network unit for acquiring electrocardiogram data and medical data regarding a subject of measurement of the electrocardiogram data. In this case, the processor may analyze record information present in the medical data to classify the electrocardiogram data into learning data or non-learning data.

[0018]

[0019] The present disclosure enables accurate and rapid analysis of electrocardiogram data by constructing customized learning data reflecting the characteristics of specific diseases such as aortic valve stenosis and left ventricular diastolic dysfunction, thereby maximizing the utility of artificial intelligence in actual medical settings.

[0020]

[0021] FIG. 1 is a block diagram of a computing device according to one embodiment of the present disclosure.

[0022] FIG. 2 is a block diagram illustrating a process for generating learning data according to one embodiment of the present disclosure.

[0023] FIG. 3 is a block diagram showing the structure of a neural network model according to one embodiment of the present disclosure.

[0024] FIG. 4 is a flowchart illustrating a method for generating artificial intelligence according to one embodiment of the present disclosure.

[0025] FIG. 5 is a flowchart illustrating a method for diagnosing a disease using artificial intelligence according to one embodiment of the present disclosure.

[0026]

[0027] Below, embodiments of the present disclosure are described in detail with reference to the attached drawings so that those skilled in the art can easily implement the present disclosure. The embodiments presented in this disclosure are provided to enable those skilled in the art to utilize or implement the contents of the present disclosure. Accordingly, various modifications to the embodiments of the present disclosure will be apparent to those skilled in the art. That is, the present disclosure may be implemented in various different forms and is not limited to the embodiments described below.

[0028] Throughout the specification of this disclosure, identical or similar drawing numbers refer to identical or similar components. Furthermore, for clarity in the description of this disclosure, drawing numbers for parts unrelated to the description of this disclosure may be omitted in the drawings.

[0029] The term "or" as used herein is intended to mean an inclusive "or" rather than an exclusive "or." That is, unless otherwise specified herein or clear from context, "X employs A or B" should be understood to mean either of the natural inclusive permutations. For example, unless otherwise specified herein or clear from context, "X employs A or B" can be interpreted to mean either X employs A, X employs B, or X employs both A and B.

[0030] The term "and / or" as used herein should be understood to refer to and include all possible combinations of one or more of the related concepts listed.

[0031] The terms "comprises" and / or "comprising" as used herein should be understood to mean the presence of certain features and / or components. However, it should be understood that the terms "comprises" and / or "comprising" do not exclude the presence or addition of one or more other features, other components, and / or combinations thereof.

[0032] Unless otherwise specified in this disclosure or unless the context makes it clear that the singular form is intended to be referred to, the singular should generally be construed to include “one or more.”

[0033] The term "Nth (N is a natural number)" used in the present disclosure can be understood as an expression used to mutually distinguish components of the present disclosure based on a predetermined standard such as a functional perspective, a structural perspective, or convenience of explanation. For example, components performing different functional roles in the present disclosure can be distinguished as a first component or a second component. However, components that are substantially the same within the technical spirit of the present disclosure but must be distinguished for convenience of explanation may also be distinguished as a first component or a second component.

[0034] The term "acquisition" as used in this disclosure may be understood to mean not only receiving data through a wired or wireless communication network with an external device or system, but also generating data in an on-device form.

[0035] Meanwhile, the term "module" or "unit" used in the present disclosure can be understood as a term referring to an independent functional unit that processes computing resources, such as a computer-related entity, firmware, software or a part thereof, hardware or a part thereof, or a combination of software and hardware. At this time, the "module" or "unit" may be a unit composed of a single element, or a unit expressed as a combination or set of multiple elements. For example, as a narrow concept, a "module" or "unit" may refer to a hardware element of a computing device or a set thereof, an application program that performs a specific function of software, a processing process implemented through software execution, or a set of instructions for program execution, etc. In addition, as a broad concept, a "module" or "unit" may refer to the computing device itself that constitutes the system, or an application running on the computing device, etc. However, since the above-described concept is only an example, the concept of “module” or “part” may be defined in various ways within a range understandable to those skilled in the art based on the contents of the present disclosure.

[0036] The term "model" as used herein may be understood as a system implemented using mathematical concepts and language to solve a specific problem, a set of software units to solve a specific problem, or an abstract model of a processing process to solve a specific problem. For example, a neural network "model" may refer to the entire system implemented as a neural network that has problem-solving capabilities through learning. In this case, the neural network can have problem-solving capabilities by optimizing the parameters connecting nodes or neurons through learning. A neural network "model" may include a single neural network or a set of neural networks that are a combination of multiple neural networks.

[0037] The explanation of the above terms is intended to aid understanding of the present disclosure. Therefore, unless explicitly stated as limiting the contents of the present disclosure, it should be noted that the above terms are not intended to limit the technical ideas of the present disclosure.

[0038] FIG. 1 is a block diagram of a computing device according to one embodiment of the present disclosure.

[0039] A computing device (100) according to one embodiment of the present disclosure may be a hardware device or a part of a hardware device that performs comprehensive processing and calculation of data, or may be a software-based computing environment connected to a communication network. For example, the computing device (100) may be a server that performs intensive data processing functions and shares resources, or may be a client that shares resources through interaction with a server. In addition, the computing device (100) may be a cloud system in which multiple servers and clients interact to comprehensively process data. Since the above description is only one example related to the type of computing device (100), the type of computing device (100) may be configured in various ways within a range understandable to those skilled in the art based on the contents of the present disclosure.

[0040] Referring to FIG. 1, a computing device (100) according to one embodiment of the present disclosure may include a processor (110), a memory (120), and a network unit (130). However, FIG. 1 is merely an example, and the computing device (100) may include other components for implementing a computing environment. In addition, only some of the disclosed components may be included in the computing device (100).

[0041] The processor (110) according to one embodiment of the present disclosure may be understood as a configuration unit including hardware and / or software for performing computing operations. For example, the processor (110) may read a computer program to perform data processing for machine learning. The processor (110) may process computational processes such as processing input data for machine learning, feature extraction for machine learning, and error calculation based on backpropagation. The processor (110) for performing such data processing may include a central processing unit (CPU), a general purpose graphics processing unit (GPGPU), a tensor processing unit (TPU), an application specific integrated circuit (ASIC), or a field programmable gate array (FPGA). The above-described type of the processor (110) is only one example, and thus, the type of the processor (110) may be configured in various ways within a range understandable to those skilled in the art based on the contents of the present disclosure.

[0042] The processor (110) can build learning data for generating an artificial intelligence model for diagnosing a disease based on electrocardiogram data. The processor (110) can analyze what diagnosis and treatment the subject of the electrocardiogram data received during the medical treatment process, and determine whether the electrocardiogram data obtained from the subject is suitable for training an artificial intelligence model for diagnosing a specific disease. Based on the determination result, the processor (110) can determine whether to include the electrocardiogram data in the learning data of the artificial intelligence model for diagnosing a specific disease. For example, in order to generate learning data for an artificial intelligence model for diagnosing aortic stenosis, the processor (110) can analyze the medical data of the subject of the measurement to identify whether the subject of the candidate data received treatment related to aortic stenosis. If the subject of the candidate data is confirmed to have received treatment related to aortic valve stenosis, the processor (110) may determine that the cause of the aortic valve stenosis has disappeared and decide not to include the candidate data in the training data. Conversely, if the subject of the candidate data is confirmed to have not received treatment related to aortic valve stenosis, the processor (110) may determine that the cause of the aortic valve stenosis exists and may include the candidate data in the training data. Through this process, the processor (110) can construct training data that emphasizes the characteristics of aortic valve stenosis.

[0043] The processor (110) can generate an artificial intelligence model for diagnosing a specific disease based on the learning data constructed as described above. The learning data constructed by the processor (110) is data selected by analyzing the diagnosis history of a specific disease that the artificial intelligence model aims to predict. In other words, the learning data constructed by the processor (110) is learning data specialized for the diagnostic area that the artificial intelligence model aims to predict, and therefore, the processor (110) can generate an artificial intelligence model optimized for diagnosing a specific disease based on such learning data. For example, the processor (110) can input the constructed learning data by filtering candidate data based on whether or not treatment related to aortic valve stenosis has been received into an artificial intelligence model for predicting the onset of aortic valve stenosis. The processor (110) can compare the onset probability data output by the artificial intelligence model with the labels included in the learning data to calculate an error. In addition, the processor (110) can perform a backpropagation operation on the artificial intelligence model based on the error to update the artificial intelligence model. The processor (110) can train an artificial intelligence model by repeating this process a preset number of times. By training the artificial intelligence model using training data that emphasizes the characteristics of aortic valve stenosis in this way, the processor (110) can create a high-performance model specialized for the diagnosis of aortic valve stenosis.

[0044] The processor (110) can predict the onset of a specific disease based on electrocardiogram data using an artificial intelligence model for which learning has been completed. The processor (110) can determine the risk of a specific disease based on the onset prediction data output by the artificial intelligence model. For example, the processor (110) can input electrocardiogram data into an artificial intelligence model for diagnosing aortic valve stenosis, and calculate a score indicating the possibility of onset of aortic valve stenosis. The processor (110) can classify a risk group for aortic valve stenosis by comparing the score calculated by the artificial intelligence model with preset cutoff values. In other words, the processor (110) can classify the risk of aortic valve stenosis of a person who measured the input data into one of three levels: low risk, medium risk, and high risk by comparing the score calculated by the artificial intelligence model with the cutoff values. Through this classification, the processor (110) can provide an environment in which a medical professional can quickly identify a person who measured the input data and provide necessary treatment.

[0045] The memory (120) according to one embodiment of the present disclosure may be understood as a configuration unit including hardware and / or software for storing and managing data processed in the computing device (100). That is, the memory (120) may store any type of data generated or determined by the processor (110) and any type of data received by the network unit (130). For example, the memory (120) may include at least one type of storage medium among a flash memory type, a hard disk type, a multimedia card micro type, a card type memory, a random access memory (RAM), a static random access memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a magnetic disk, and an optical disk. In addition, the memory (120) may also include a database system that controls and manages data in a predetermined system. The type of memory (120) described above is only one example, and thus the type of memory (120) can be configured in various ways within a range understandable to those skilled in the art based on the contents of the present disclosure.

[0046] The memory (120) can structure and organize and manage data, combinations of data, and program codes executable by the processor (110) required for the processor (110) to perform operations. For example, the memory (120) can store medical data received through the network unit (130) described below. The memory (120) can store program codes that operate a neural network model to receive medical data as input and perform learning, program codes that operate a neural network model to receive medical data as input and perform inference according to the purpose of use of the computing device (100), and processed data generated as the program codes are executed.

[0047] The network unit (130) according to one embodiment of the present disclosure may be understood as a component that transmits and receives data through any type of known wired or wireless communication system. For example, the network unit (130) may perform data transmission and reception using a wired or wireless communication system such as a local area network (LAN), wideband code division multiple access (WCDMA), long term evolution (LTE), wireless broadband internet (WiBro), fifth generation mobile communication (5G), ultra wide-band, ZigBee, radio frequency (RF) communication, wireless LAN, wireless fidelity, near field communication (NFC), or Bluetooth. Since the above-described communication systems are only examples, the wired and wireless communication system for data transmission and reception of the network unit (130) may be applied in various ways other than the above-described examples.

[0048] The network unit (130) can receive data necessary for the processor (110) to perform calculations through wired or wireless communication with any system or any client, etc. In addition, the network unit (130) can transmit data generated through calculations of the processor (110) through wired or wireless communication with any system or any client, etc. For example, the network unit (130) can receive medical data through communication with a cloud server that performs tasks such as databases in a hospital environment, standardization of medical data, clients such as smart watches, or medical computing devices, etc. The network unit (130) can transmit output data of a neural network model, intermediate data derived from the calculation process of the processor (110), processed data, etc. through communication with the aforementioned database, server, client, or computing device, etc.

[0049] FIG. 2 is a block diagram illustrating a process for generating learning data according to one embodiment of the present disclosure.

[0050] A computing device (100) according to one embodiment of the present disclosure can acquire electrocardiogram data (10) to construct learning data. In addition, the computing device (100) can also acquire medical data (20) of a person who measured the electrocardiogram data (10). For example, the computing device (100) can collect a candidate data set including electrocardiogram data (10) and medical data (20) related to a subject of measurement of the electrocardiogram data (10) through communication with a database established in a medical environment.

[0051] The computing device (100) can analyze key-value values ​​corresponding to the record information included in the medical data (20) to determine whether the candidate data, electrocardiogram data (10), can be used as learning data (30) of an artificial intelligence model for diagnosing a specific disease. Depending on the determination, the computing device (100) can classify the electrocardiogram data (10) as learning data (30) or as non-learning data (40) that is not used for learning. For example, the medical data (20) may include record information on diagnosis, examination, or treatment generated when a subject of the electrocardiogram data (10) visits a medical institution in the form of key-value values. Accordingly, the computing device (100) can identify the record information included in the medical data (20) as key-value values ​​(S110). In addition, the computing device (100) can determine whether there is a value matching a preset reference value among the identified key-value values ​​(S120). At this time, the reference value may be a preset value to select information specialized for a specific disease that the neural network model is targeting for prediction. In other words, the reference value may be preset as a key-value value corresponding to information that may influence the diagnosis of a specific disease among the record information related to the specific disease that the neural network model is targeting for prediction. Specifically, the reference value may be a value corresponding to at least one of procedure history information performed in relation to a specific disease, surgery history information, or opinion information on a specific disease from a medical professional such as a cardiac ultrasound specialist or attending physician. In addition, the opinion information may be information that includes a phrase that can be assumed to have received a procedure or surgery for a specific disease even though the subject of the electrocardiogram data (10) has not received a procedure or surgery for the specific disease. All information corresponding to the preset reference value in this way indicates a diagnosis result that the cause of the specific disease has been eliminated through treatment or that the cause does not exist.

[0052] If there is no value matching the preset reference value among the key-value values ​​identified through step S110, the computing device (100) can determine that the electrocardiogram data (10) is data measured in a state where the cause of a specific disease exists. If the electrocardiogram data (10) was measured in a state where the cause of a specific disease exists, it can be seen that the electrocardiogram data (10) well represents the characteristics necessary for artificial intelligence to diagnose the specific disease. Therefore, if there is no value matching the preset reference value among the key-value values ​​identified through step S110, the computing device (100) can classify the electrocardiogram data (10) as learning data (30) (S130). Conversely, if there is a value matching the preset reference value among the key-value values ​​identified through step S110, the computing device (100) can determine that the electrocardiogram data (10) is data measured in a state where the cause of a specific disease does not exist. If the electrocardiogram data (10) was measured in a state where the cause of a specific disease does not exist, it is difficult to say that the electrocardiogram data (10) properly expresses the characteristics necessary for artificial intelligence to diagnose a specific disease. Therefore, if there is a value that matches a preset reference value among the key-value values ​​identified through step S110, the computing device (100) can classify the electrocardiogram data (10) as non-learning data (40) (S140).

[0053] The computing device (100) can analyze the text corresponding to the record information included in the medical data (20) to determine whether the candidate data, the electrocardiogram data (10), can be used as learning data (30) of an artificial intelligence model for diagnosing a specific disease. Depending on the determination, the computing device (100) can classify the electrocardiogram data (10) as learning data (30) or as non-learning data (40) that is not used for learning. For example, the medical data (20) may store record information on diagnosis, examination, or treatment generated when a subject of the electrocardiogram data (10) visits a medical institution in a CSV (comma-separated value) format. Accordingly, the computing device (100) can identify the text-type record information included in the medical data (20) (S110). In addition, the computing device (100) can determine whether there is a text matching a preset reference keyword among the identified texts (S120). At this time, the reference keyword may be a keyword preset to select information specialized for a specific disease that the neural network model aims to predict. In other words, the reference keyword may be preset to a value corresponding to information that may influence the diagnosis of a specific disease among the record information related to the specific disease that the neural network model aims to predict. Specifically, the reference keyword may be a value corresponding to at least one of information on the history of procedures performed in relation to a specific disease, information on the history of surgery, or information on the opinion of a medical professional such as a cardiac ultrasound specialist or attending physician regarding a specific disease. In addition, the opinion information may be information that includes a phrase that can be assumed to have received a procedure or surgery for a specific disease, even though the subject of the electrocardiogram data (10) did not receive the procedure or surgery. All information corresponding to the preset reference keyword in this way indicates a diagnosis result that the cause of the specific disease has been eliminated through treatment or that the cause does not exist.

[0054] If a preset reference keyword does not exist in the text identified through step S110, the computing device (100) can determine that the electrocardiogram data (10) is data measured in a state where the cause of a specific disease exists. If the electrocardiogram data (10) was measured in a state where the cause of a specific disease exists, it can be seen that the electrocardiogram data (10) well represents the characteristics necessary for artificial intelligence to diagnose the specific disease. Therefore, if a preset keyword does not exist in the text identified through step S110, the computing device (100) can classify the electrocardiogram data (10) as learning data (30) (S130). Conversely, if a preset reference keyword exists in the text identified through step S110, the computing device (100) can determine that the electrocardiogram data (10) is data measured in a state where the cause of a specific disease does not exist. If the electrocardiogram data (10) was measured in a state where the cause of a specific disease does not exist, it is difficult to say that the electrocardiogram data (10) properly represents the characteristics necessary for artificial intelligence to diagnose a specific disease. Therefore, if a preset reference keyword exists in the text identified through step S110, the computing device (100) can classify the electrocardiogram data (10) as non-learning data (40) (S140).

[0055] The computing device (100) can generate high-quality learning data by reflecting the diagnostic characteristics of a specific disease through this classification process. The learning data generated through the classification performed by the computing device (100) will contain a high purity of the features necessary for artificial intelligence to predict a specific disease, and thus can be effectively utilized to enhance the learning efficiency and prediction performance of the artificial intelligence model.

[0056] Meanwhile, in the classification process described above, the specific disease targeted by the neural network model may be at least one of aortic valve stenosis or left ventricular diastolic dysfunction. Aortic valve stenosis occurs when the aortic valve thickens or calcifies, preventing it from opening and closing properly, hindering blood flow from the left ventricle to the aorta. Valve disease can be improved through valve replacement or surgery, and there are three main treatment options. First, transcatheter aortic valve implantation (TAVI) is a non-invasive procedure that replaces a damaged aortic valve with an artificial valve using a catheter. Second, valve repair is a procedure that restores function by suturing or reinforcing the damaged valve without removing it. Third, valve replacement involves removing the damaged valve and replacing it with a mechanical or bioprosthetic valve. Mechanical valves can be used semi-permanently, but require lifelong anticoagulant therapy to prevent blood clots. Bioprosthetic valves do not require anticoagulants, but their durability is limited and may require revision surgery after 10 to 15 years. When symptoms develop, aortic valve replacement is the primary treatment. If this procedure is not performed promptly, severe cases can lead to heart failure or death.

[0057] Unlike heart diseases such as heart failure or acute myocardial infarction, aortic valve stenosis can be considered cured by performing the aforementioned procedures or surgeries. Conversely, heart failure can be caused by a variety of factors, including damage or weakening of the heart muscle. Because these underlying causes cannot be fundamentally eliminated, even if heart failure-related procedures or surgeries are performed, the underlying cause cannot be considered eliminated. Acute myocardial infarction is a disease caused by blockage in the blood vessels supplying blood to the heart. Even if percutaneous coronary intervention (PCI) is performed on a specific blood vessel to treat this condition, there is a risk of further blockage elsewhere in the blood vessel. Therefore, even in the case of acute myocardial infarction, performing PCI does not necessarily eliminate the underlying cause.

[0058] In other words, unlike heart failure or acute myocardial infarction, patients with aortic valve stenosis who have undergone a procedure or surgery can be considered cured because the cause of the disease has been removed. Therefore, if the diagnostic data of a subject whose electrocardiogram data was measured includes information about a history of aortic valve surgery or surgery, or information about findings that suggest such a procedure or surgery (e.g., echocardiography results with phrases such as "well-functioning bioprosthetic avbioprosthetic aortic valve"), the subject can be considered not to have aortic valve stenosis. The absence of aortic valve stenosis in a subject's electrocardiogram data can be interpreted as a lack or absence of features necessary for the neural network model to diagnose aortic valve stenosis. Therefore, classifying such electrocardiogram data as not included in the training data can contribute to improving the learning efficiency and predictive performance of a neural network model for diagnosing aortic valve stenosis.

[0059] Meanwhile, even in the case of left ventricular diastolic dysfunction, if there is a record of a procedure or surgery to replace the aortic valve, the burden on the left ventricle is reduced, improving diastolic function, so it can be considered that the cause has disappeared. Therefore, when configuring the training data for a neural network model to diagnose left ventricular diastolic dysfunction, if the diagnostic data of the subject whose electrocardiogram data was measured as described above includes information about a history of aortic valve procedure or surgery, or information about findings that can be considered to indicate that such a procedure or surgery has been performed, the electrocardiogram data in question can be classified as not being included in the training data.

[0060] Considering the above, the learning data generation process described based on Fig. 2 can be understood as being optimized for building customized learning data for a neural network model specialized in diagnosing aortic valve stenosis or left ventricular diastolic dysfunction, rather than heart failure or acute myocardial infarction.

[0061] FIG. 3 is a block diagram showing the structure of a neural network model according to one embodiment of the present disclosure.

[0062] A neural network model according to one embodiment of the present disclosure may include a residual neural network (310) that receives electrocardiogram data (50) as input and extracts features necessary for diagnosis, an encoder (320) that performs self-attention calculation based on the output of the residual neural network (310), and a fully connected neural network (330) that calculates the likelihood of disease onset (60) based on the output of the encoder (320).

[0063] For example, the residual neural network (310) can extract morphological features of the electrocardiogram from the electrocardiogram data (50). The encoder (320) can capture features that play an important role in diagnosis by performing a self-attention operation based on the correlation between features extracted through the residual neural network (310). The fully connected neural network (330) can output the probability of disease onset (60) as probability data based on the features captured through the residual neural network (310) and the encoder (320). The probability data output through the fully connected neural network (330) can be converted into a score form.

[0064] Meanwhile, the probability of disease onset (60) converted into a score format can be compared with a plurality of preset cutoff values ​​to be used to classify the disease risk group of the diagnosis subject. Specifically, the probability of disease onset (60) converted into a score format can be compared with a first cutoff value to be used to classify the diagnosis subject into a low-risk group or a medium-risk group for a specific disease. At this time, the first cutoff value can be a value set based on sensitivity N% (N is a natural number) according to clinical research results. In addition, the probability of disease onset (60) converted into a score format can be compared with a second cutoff value to be used to classify the diagnosis subject into a medium-risk group or a high-risk group for a specific disease. At this time, the second cutoff value can be a value set based on specificity M% (M is a natural number) according to clinical research results. Such risk group classification can provide a medical environment that allows for rapid diagnosis and response by allowing for intuitive understanding of the diagnosis subject's condition.

[0065] FIG. 4 is a flowchart illustrating a method for generating artificial intelligence according to one embodiment of the present disclosure.

[0066] Referring to FIG. 4, a computing device (100) according to one embodiment of the present disclosure can obtain electrocardiogram (ECG) data and medical data regarding the subject of the ECG data measurement (S210). For example, the computing device (100) can obtain medical data paired with ECG data through communication with a database established at a medical institution.

[0067] The computing device (100) can analyze the record information present in the medical data to classify the electrocardiogram data as learning data or non-learning data (S220). The computing device (100) can identify key-value values ​​or text corresponding to the record information present in the medical data. The computing device (100) can analyze whether a value matching a preset reference value exists among the identified key-value values. The computing device (100) can analyze whether a preset reference keyword exists in the identified text. The computing device (100) can classify the electrocardiogram data as learning data or non-learning data depending on whether a value matching the preset reference value exists. The computing device (100) can classify the electrocardiogram data as learning data or non-learning data depending on whether a preset reference keyword exists. Specifically, the computing device (100) can classify the electrocardiogram data as non-learning data if a value matching the preset reference value or a text matching the preset reference keyword exists. Conversely, the computing device (100) may classify the electrocardiogram data as learning data if there is no value matching a preset reference value or no text matching the preset reference keyword. At this time, the reference value for classifying the electrocardiogram data as learning data or non-learning data may be a value corresponding to at least one of treatment history information for a specific disease, surgery history information for a specific disease, or medical expert opinion information for a specific disease. In addition, the opinion information may include a phrase that may be assumed to indicate that the subject of measurement has undergone a procedure or surgery for a specific disease. For example, a phrase included in the opinion information related to aortic valve stenosis may be the phrase "well-functioning bioprosthetic avbioprosthetic aortic valve."If that statement is included in a doctor's report, other doctors can determine if the patient has had a procedure or surgery related to aortic valve stenosis.

[0068] In addition, the computing device (100) can input electrocardiogram data classified as learning data into the neural network model to train the neural network model so that the neural network model can analyze the possibility of occurrence of a specific disease based on the electrocardiogram data. At this time, the specific disease may be at least one of aortic valve stenosis and left ventricular diastolic dysfunction. The computing device (100) can build high-quality learning data containing high-purity information necessary for diagnosing aortic valve stenosis or left ventricular diastolic dysfunction through the above-described S220. Therefore, the computing device (100) can train the neural network model using such high-quality learning data, thereby significantly improving the learning efficiency and prediction performance of the neural network model.

[0069] FIG. 5 is a flowchart illustrating a method for diagnosing a disease using artificial intelligence according to one embodiment of the present disclosure.

[0070] Referring to FIG. 5, a computing device (100) according to an embodiment of the present disclosure may input electrocardiogram data into a neural network model trained through the process of FIG. 4, and output a score indicating the likelihood of developing a specific disease (S310). At this time, the score may be a value converted into the likelihood of developing a specific disease. For example, the computing device (100) may input electrocardiogram data into a neural network model trained to predict aortic valve stenosis, and calculate the likelihood of developing aortic valve stenosis in the form of a score. The computing device (100) may input electrocardiogram data into a neural network model trained to predict left ventricular diastolic dysfunction, and calculate the likelihood of developing left ventricular diastolic dysfunction in the form of a score. The model trained to predict aortic valve stenosis and the model trained to predict left ventricular diastolic dysfunction may be the same model, or may be separate models that are built independently.

[0071] The computing device (100) can classify the risk group of a specific disease for the subject of the electrocardiogram data by comparing the score output through S310 with preset cutoff values ​​(S320). For example, if the probability of developing aortic valve stenosis is calculated in the form of a score from a model trained to predict aortic valve stenosis, the computing device (100) can classify the subject of the electrocardiogram data into one of three risk groups of low-risk group, medium-risk group, and high-risk group for aortic valve stenosis by comparing the score with a plurality of cutoff values. At this time, the cutoff value for distinguishing between the low-risk group and the medium-risk group can be set based on the sensitivity, which is one of the indices used to verify the performance of the model. The cutoff value for distinguishing between the medium-risk group and the high-risk group can be set based on the specificity, which is one of the indices used to verify the performance of the model.

[0072] The various embodiments of the present disclosure described above can be combined with additional embodiments and modified within the scope understood by those skilled in the art in light of the detailed description above. It should be understood that the embodiments of the present disclosure are illustrative in all respects and not restrictive. For example, each component described as a single component may be implemented in a distributed manner, and likewise, components described as distributed may be implemented in a combined manner. Accordingly, all changes or modifications derived from the meaning, scope, and equivalent concepts of the claims of the present disclosure should be construed as being included within the scope of the present disclosure.

Claims

1. A method for generating an artificial intelligence for diagnosing a disease based on an electrocardiogram, the method being performed by a computing device including at least one processor, A step of acquiring electrocardiogram data and medical data regarding a subject of measurement of the electrocardiogram data; and A step of analyzing the record information present in the above medical data and classifying the electrocardiogram data into learning data or non-learning data; including, method.

2. In paragraph 1, The step of classifying the electrocardiogram data into learning data or non-learning data by analyzing the record information existing in the above medical data is as follows: A step of identifying key-value values ​​corresponding to the above record information or text corresponding to the above record information; A step of analyzing whether there is a value matching a preset reference value among the identified key-value values, or analyzing whether there is a text matching a preset reference keyword in the identified text; and A step of classifying the electrocardiogram data into learning data or non-learning data depending on whether there is a value matching the preset reference value or a text matching the preset reference keyword; including, method.

3. In paragraph 2, The step of classifying the electrocardiogram data into learning data or non-learning data depending on whether there is a value matching the preset reference value or a text matching the preset reference keyword, If there is a value matching the preset reference value or a text matching the preset reference keyword, the electrocardiogram data is classified as non-learning data, If there is no value matching the preset reference value or no text matching the preset reference keyword, the electrocardiogram data is classified as learning data. method.

4. In paragraph 2, The above preset reference value or the above preset reference keyword, A value corresponding to at least one of the following: treatment history information for a specific disease, surgical history information for the specific disease, or medical expert opinion information for the specific disease. method.

5. In paragraph 4, The above opinion information is, Including phrases that may be assumed to indicate that the subject of the measurement has undergone a procedure or surgery for a specific disease; method.

6. In paragraph 1, A step of training a neural network model by inputting electrocardiogram data classified as the training data into the neural network model so that the neural network model analyzes the possibility of occurrence of a specific disease based on the electrocardiogram data; including more, method.

7. In paragraph 4 or paragraph 6, The above specific disease is, At least one of aortic valve stenosis or left ventricular diastolic dysfunction, method.

8. A computer program stored in a computer-readable storage medium, wherein the computer program, when executed on one or more processors, performs operations for generating artificial intelligence that diagnoses a disease based on an electrocardiogram. The above actions are, An operation of acquiring electrocardiogram data and medical data regarding a subject of measurement of said electrocardiogram data; and An operation of analyzing the record information present in the above medical data and classifying the electrocardiogram data into learning data or non-learning data; including, Computer program.

9. A computing device for creating artificial intelligence that diagnoses diseases based on electrocardiograms. A processor comprising at least one core; A memory containing program codes executable by the processor; and A network unit for obtaining electrocardiogram data and medical data regarding the subject of measurement of the electrocardiogram data; Including, The above processor, By analyzing the record information present in the above medical data, the electrocardiogram data is classified into learning data or non-learning data. device.

Images