WO2024191200A1 - Method, program and device for identifying potential atrial fibrillation on basis of artificial intelligence model - Google Patents

Abstract

The present disclosure provides a method, a program and a device for identifying and diagnosing potential atrial fibrillation on the basis of an artificial intelligence model. The method for identifying potential atrial fibrillation on the basis of an artificial intelligence model, according to one embodiment of the present disclosure, comprises the steps of: acquiring an electrocardiogram signal from a patient with embolic stroke of undetermined source; and inputting the acquired electrocardiogram signal into a pre-trained neural network model to diagnose the patient with embolic stroke of undetermined source.

Description

Method, program and device for identifying potential atrial fibrillation based on artificial intelligence model

The present disclosure relates to deep learning technology in the medical field, and more particularly to a method, program and device for identifying potential atrial fibrillation in a patient based on an artificial intelligence model.

Paroxysmal atrial fibrillation (AF) is a potential cause of embolic stroke of undetermined source (ESUS). Paroxysmal atrial fibrillation can be treated with oral anticoagulant therapy, and treating paroxysmal atrial fibrillation can reduce the risk of recurrent stroke.

However, since the patient's paroxysmal atrial fibrillation does not occur periodically, it is difficult to detect. For this purpose, an insertable cardiac monitor (ICM) is used, but the insertable cardiac monitor is expensive and its use is limited, especially because it must be inserted into the patient.

Recently, with the development of information and communication technology, a method using a neural network model has been proposed to detect paroxysmal atrial fibrillation in patients with idiopathic embolic stroke. However, there is a problem of low accuracy when using a neural network model.

The present disclosure has been made in response to the aforementioned background technology, and aims to provide a method, program and device for identifying potential atrial fibrillation based on an artificial intelligence model.

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

According to one embodiment of the present disclosure for achieving the above-described task, a method for identifying atrial fibrillation based on an artificial intelligence model, performed by a computing device including at least one processor, comprises the steps of obtaining an electrocardiogram signal of a subject of diagnosis, and the step of inputting the obtained electrocardiogram signal into a pre-trained neural network model to identify atrial fibrillation of the subject of diagnosis.

Alternatively, the obtaining step may further include a step of obtaining an electrocardiogram signal of the subject of diagnosis and information on the subject of diagnosis, and the identifying step may further include a step of inputting the obtained electrocardiogram signal of the subject of diagnosis and information on the subject of diagnosis into a pre-trained neural network model to identify atrial fibrillation of the subject of diagnosis.

Alternatively, the subject of diagnosis may be a patient with an embolic stroke of unknown etiology, the potential atrial fibrillation includes paroxysmal atrial fibrillation, and the identifying step may include a step of inputting the acquired electrocardiogram signal and information on the subject of diagnosis into the neural network model to identify whether the patient with an embolic stroke of unknown etiology has paroxysmal atrial fibrillation, and diagnosing based on the identification result.

Alternatively, the neural network model may include a first neural network that extracts features of the electrocardiogram signal, a second neural network that extracts demographic features based on the information of the subject of diagnosis, and a third neural network that is connected to the first and second neural networks and determines whether the patient with an embolic stroke of unknown cause has paroxysmal atrial fibrillation based on the features of the electrocardiogram signal and the demographic features.

Alternatively, the identifying step may include a step of inputting the electrocardiogram signal into the first neural network, inputting the diagnosis subject information into the second neural network, obtaining a probability value of the paroxysmal atrial fibrillation through the third neural network, and diagnosing the patient with an embolic stroke of unknown cause based on the obtained probability value.

Alternatively, the first neural network may include a plurality of residual blocks, the residual blocks including a first sub-block including a skip connection structure connected with a convolutional layer and a max pooling layer, and a second sub-block connected to the first sub-block and including a skip connection structure.

Alternatively, the first neural network can extract features of the electrocardiogram signal based on at least one of the PR interval, QRS amplitude, and QT interval of the electrocardiogram signal.

Alternatively, the third neural network may output a higher value for the probability of paroxysmal atrial fibrillation as the PR interval of the electrocardiogram signal is longer, the QRS amplitude is larger, and the QT interval is longer.

Alternatively, the identifying step may include a step of inputting the acquired electrocardiogram signal into the neural network model to acquire a first probability value of the atrial fibrillation, and a step of identifying the atrial fibrillation of the subject of the diagnosis based on the acquired first probability value, the left atrial diameter (LAD) of the subject of the diagnosis, and the atrial ectopic load value of the subject of the diagnosis.

Alternatively, the identifying step may include a step of inputting the obtained first probability value, the left atrial diameter of the subject of diagnosis, and the atrial ectopic load value of the subject of diagnosis into a logistic regression model to obtain a second probability value of whether the subject of diagnosis has atrial fibrillation, and a step of determining whether the subject of diagnosis has paroxysmal atrial fibrillation based on the obtained second probability value.

Alternatively, the subject information may include at least one of information about age, gender, height, weight, or brain imaging test results.

Alternatively, the obtaining step may further include a step of obtaining an electrocardiogram signal and an electroencephalogram signal of the subject of diagnosis, and the identifying step may further include a step of inputting the obtained electrocardiogram signal and the electroencephalogram signal of the subject of diagnosis into a pre-trained neural network model to identify atrial fibrillation of the subject of diagnosis.

Alternatively, the neural network model may include a fourth neural network and a fifth neural network that extract features corresponding to the electrocardiogram signal and the brainwave signal, respectively, and a sixth neural network that performs a classification task regarding potential atrial fibrillation based on the extracted features.

According to one embodiment of the present disclosure for achieving the above-described task, a computer program stored in a computer-readable storage medium, wherein the computer program, when executed on one or more processors, performs an operation of identifying potential atrial fibrillation based on an artificial intelligence model, wherein the operation may include an operation of acquiring an electrocardiogram signal and information on the subject of diagnosis, and an operation of inputting the acquired electrocardiogram signal and information on the subject of diagnosis into a pre-learned neural network model, thereby identifying atrial fibrillation in the subject of diagnosis. /

According to one embodiment of the present disclosure for achieving the above-described task, a computing device for identifying potential atrial fibrillation based on an artificial intelligence model, comprising: a processor including at least one core; a memory including program codes executable by the processor; and a network unit, wherein the processor acquires an electrocardiogram signal and information on the subject of diagnosis, and inputs the acquired electrocardiogram signal and information on the subject of diagnosis into a pre-trained neural network model to identify atrial fibrillation in the subject of diagnosis.

According to a method for identifying potential atrial fibrillation based on an artificial intelligence model according to one embodiment of the present disclosure, the cause of an unexplained embolic stroke can be identified, thereby providing a treatment method for a patient with an unexplained embolic stroke.

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

FIG. 2 is a flowchart schematically illustrating a method for identifying potential atrial fibrillation based on an artificial intelligence model according to one embodiment of the present disclosure.

FIG. 3 is an exemplary diagram showing the configuration of a pre-learned neural network model based on an electrocardiogram signal and information on a subject of diagnosis according to one embodiment of the present disclosure.

FIG. 4 is a diagram showing the performance of a pre-learned neural network model according to an embodiment of the present disclosure.

FIG. 5 is an exemplary diagram showing the configuration of a pre-learned neural network model based on electrocardiogram signals and brainwave signals according to one embodiment of the present disclosure.

FIG. 6 is a detailed block diagram of a computing device according to one embodiment of the present disclosure.

Hereinafter, embodiments of the present disclosure will be 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 the present disclosure are provided so that those skilled in the art can 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 below.

Throughout the specification of the present disclosure, the same or similar drawing numbers refer to the same or similar components. In addition, in order to clearly describe the present disclosure, drawing numbers of parts in the drawings that are not related to the description of the present disclosure may be omitted.

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 the context makes clear, "X employs either A or B" should be understood to mean either one of the natural inclusive permutations. For example, unless otherwise specified herein or the context makes clear, "X employs A or B" can be interpreted to mean either X employs A, X employs B, or X employs both A and B.

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

The terms "comprises" and/or "comprising" as used herein should be understood to mean the presence of particular 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.

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.”

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 according to a predetermined standard such as a functional viewpoint, a structural viewpoint, 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 idea of the present disclosure but should be distinguished for convenience of explanation may also be distinguished as a first component or a second component.

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

Meanwhile, the term "module" or "unit" used in the present disclosure may 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, a combination of software and hardware, etc. In this case, the "module" or "unit" may be a unit composed of a single element, or may be 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 a computing device itself that constitutes a system, or an application that is executed on a computing device, etc. However, the above-described concept is only an example, and the concept of “module” or “part” may be variously defined within a category understandable to those skilled in the art based on the contents of the present disclosure.

The term "model" used in the present disclosure 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 regarding 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 a problem-solving ability through learning. In this case, the neural network may have a problem-solving ability by optimizing parameters connecting nodes or neurons through learning. A neural network "model" may include a single neural network, or may include a neural network set in which multiple neural networks are combined.

The term "data" used in the present disclosure may include "images", signals, and the like. The term "image" used in the present disclosure may refer to multidimensional data composed of discrete image elements. In other words, "image" may be understood as a term referring to a digital representation of an object that can be seen with the human eye. For example, "image" may refer to multidimensional data composed of elements corresponding to pixels in a two-dimensional image. "Image" may refer to multidimensional data composed of elements corresponding to voxels in a three-dimensional image.

The explanation of the terms set forth above is intended to aid in understanding the present disclosure. Therefore, if the terms set forth above are not explicitly stated as matters limiting the contents of the present disclosure, it should be noted that they are not used to limit the technical ideas of the contents of the present disclosure.

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

The 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 a plurality of servers and clients interact to comprehensively process data. Since the above description is only one example related to the type of the computing device (100), the type of the computing device (100) may be configured in various ways within a category understandable to a person skilled in the art based on the contents of the present disclosure. Meanwhile, the computing device (100) may be implemented as various electronic devices such as a desktop, a laptop, a smartphone, and a server.

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

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 of 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 type of the processor (110) described above 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.

The processor (110) is connected to other components of the computing device (100) (i.e., memory (120) and network unit (130)) and controls the overall operation of the computing device (100).

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 RAM (random access memory), a SRAM (static random access memory), a ROM (read-only memory), an EEPROM (electrically erasable programmable read-only memory), a PROM (programmable read-only memory), 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 an 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.

The memory (120) can structure and organize and manage data, a combination 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 electrocardiogram signals and information on a subject of diagnosis received through the network unit (130) described below. In addition, the memory (120) can store a neural network model learned to identify potential atrial fibrillation based on the electrocardiogram signals and information on a subject of diagnosis, and can store program codes that operate to perform learning of the neural network model, program codes that operate the neural network model to receive electrocardiogram signals and information on a subject of diagnosis and perform inference according to the intended use of the computing device (100), and processed data generated as the program codes are executed. In addition, the memory (120) may include a learning data set for learning the neural network model. Here, the learning data set may include electrocardiogram signals and information on a subject of diagnosis of a plurality of subjects of diagnosis.

The network unit (130) according to one embodiment of the present disclosure may be understood as a configuration unit that transmits and receives data via any type of known wired and wireless communication system. For example, the network unit (130) may perform data transmission and reception using a wired and 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), ultrawide-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.

The network unit (130) can receive data required 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 the calculation 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 standardization of databases and medical data in a hospital environment, or a computing device, etc. The network unit (130) can transmit output data of a neural network model, intermediate data, processed data, etc. derived from the calculation process of the processor (110), etc. through communication with the aforementioned database, server, or computing device, etc.

FIG. 2 is a flowchart schematically illustrating a method for diagnosing potential atrial fibrillation based on an artificial intelligence model according to one embodiment of the present disclosure.

Referring to FIG. 2, the processor (110) can obtain an electrocardiogram signal of a subject of diagnosis (S210).

Here, the subject of diagnosis includes a patient who is diagnosed with a disease through a computing device. In particular, the patient may be a patient suspected of having atrial fibrillation. As an example, the patient may include a patient with an Embolic Stroke of Undetermined Source (ESUS).

Meanwhile, according to one embodiment of the present disclosure, the processor (110) can obtain patient information (hereinafter, diagnosis subject information) together. That is, the processor (110) can obtain the electrocardiogram signal of the diagnosis subject and the diagnosis subject information. Here, the diagnosis subject information may be information that can identify the diagnosis subject. Or, the diagnosis subject information may be another diagnosis information of the patient.

For example, the information on the subject of diagnosis may be information on a patient with an idiopathic embolic stroke. Accordingly, the processor (110) may obtain an electrocardiogram (ECG) signal and information on the subject of diagnosis of a patient with an idiopathic embolic stroke.

Specifically, the processor (110) can obtain at least one of the patient's electrocardiogram signal and the diagnosis subject information from an external electronic device through the network unit (130).

An electrocardiogram signal is a wave-shaped signal generated by analyzing the electrical activity of the heart, and may also be referred to as electrocardiogram data.

The information on the subject of diagnosis may include at least one of the patient's age, gender, weight, height, or brain imaging test results. Specifically, the information on the subject of diagnosis may include the patient's age, gender, weight, or height as information that can identify the patient. In addition, the information on the subject of diagnosis may include another diagnostic information of the patient, the result of a brain imaging test. The result of the brain imaging test may be a result of a computed tomography scan or a magnetic resonance imaging scan, and may be acquired from an external imaging device through the network unit (130). In this case, the result of the brain imaging test may include information on brain anatomical structure, stroke, brain tumor, cerebral hemorrhage, inflammation, cerebral infarction, etc.

The external electronic device may be a signal measuring device that acquires an electrocardiogram signal of a patient. For example, the external electronic device includes at least one electrode, and can detect an electrical signal generated from the heart of the patient by attaching the at least one electrode to the patient's body to acquire an electrocardiogram signal. In addition, the processor (110) can receive the electrocardiogram signal acquired by the external electronic device through the network unit (130).

In addition, the processor (110) may directly obtain an electrocardiogram signal through a sensing unit (not shown) included in the computing device (100). The sensing unit may include at least one electrode. When at least one electrode of the sensing unit is attached to the patient's body, the processor (110) may detect an electrical signal generated from the patient's heart through the at least one electrode to obtain an electrocardiogram signal. As an example, the obtained electrocardiogram signal may be electrocardiogram data of 12 leads, and the sensing unit may include 12 multiple leads.

In addition, the electrocardiogram signal can be measured at a sampling rate of 500 Hz for 10 seconds. At this time, the processor (110) can process the electrocardiogram signal by removing a 1-second region where the electrocardiogram signal begins and a last 1-second region where the electrocardiogram signal ends in order to remove noise or artifacts included in the electrocardiogram signal.

The processor (110) can obtain information on a subject of diagnosis corresponding to a patient through an input interface (not shown) of the computing device (100). For example, the input interface can be implemented as a touch screen or a keyboard, and the processor (110) can receive information on a subject of diagnosis from a user through the input interface. However, the present invention is not limited thereto, and the processor (110) can also obtain information on a subject of diagnosis by receiving it from an external electronic device through the network unit (130). Here, the external electronic device can be a signal measuring device that obtains an electrocardiogram signal of the patient described above, or can be another device (for example, a user terminal, etc.) connected to the computing device (100) through the network unit (130).

Then, the processor (110) inputs the acquired electrocardiogram signal into a pre-trained neural network model to identify potential atrial fibrillation of the subject of diagnosis (S220).

Potential atrial fibrillation may be an atrial fibrillation that is not observed periodically from a subject of diagnosis and may occur intermittently or irregularly. Potential atrial fibrillation may not be observed in an electrocardiogram signal because it occurs irregularly. However, according to an embodiment of the present disclosure, the processor (110) may extract a factor (or factors) of potential atrial fibrillation from an electrocardiogram signal using a pre-learned neural network model to determine whether a patient has potential atrial fibrillation. Hereinafter, an embodiment of the present disclosure related thereto will be described in detail.

The processor (110) can identify potential atrial fibrillation in the subject of diagnosis and diagnose the subject of diagnosis based on the identification result. Diagnosing the subject of diagnosis may provide the diagnosis, treatment, and prescription method of the subject of diagnosis according to the presence or absence of potential atrial fibrillation.

In particular, the processor (110) can input at least one of an electrocardiogram signal and patient information acquired from a patient with an embolic stroke of unknown cause into a pre-learned model to determine whether the patient with an embolic stroke of unknown cause has potential atrial fibrillation. Determining whether there is potential atrial fibrillation may be determining whether atrial fibrillation can occur irregularly from a patient with an embolic stroke of unknown cause, although it is not observed in the acquired electrocardiogram signal. The processor (110) can diagnose a patient with an embolic stroke of unknown cause based on the determination result.

Diagnosing patients with cryptogenic embolic stroke may involve identifying the cause of cryptogenic embolic stroke and generating prescribing information for patients with cryptogenic embolic stroke.

Hereinafter, for the convenience of description of the present disclosure, patients with idiopathic thrombotic stroke are referred to as patients.

For example, the pre-learned neural network model may be a model learned to determine potential atrial fibrillation of a patient based on an electrocardiogram signal. Specifically, the pre-learned neural network model may be a model learned to extract feature information from an electrocardiogram signal when an electrocardiogram signal is input, and to determine whether the patient has potential atrial fibrillation based on the extracted feature information. Here, the feature information may be information related to potential atrial fibrillation included in the electrocardiogram signal.

Meanwhile, the pre-learned model may be a model learned to identify the cause of the patient's embolic stroke of unknown cause based on at least one of the information of the diagnosis subject. In particular, the pre-learned neural network model may be a model learned to identify whether the patient has paroxysmal atrial fibrillation based on the input electrocardiogram signal and the information of the diagnosis subject when the electrocardiogram signal and the information of the diagnosis subject are input. That is, the atrial fibrillation may be paroxysmal atrial fibrillation, and the pre-learned neural network model may identify whether the patient has paroxysmal atrial fibrillation. The processor (110) may input at least one of the acquired electrocardiogram signal or the information of the diagnosis subject into the neural network model to determine whether the patient has paroxysmal atrial fibrillation. Then, the processor (110) may diagnose the patient with an embolic stroke of unknown cause based on the determination result. The processor (110) identifies whether the patient has paroxysmal atrial fibrillation based on the result value of the neural network model, and if the patient has paroxysmal atrial fibrillation, it can determine that the cause of the patient's unexplained embolic stroke is paroxysmal atrial fibrillation.

Meanwhile, the neural network model can be learned in advance based on a learning data set consisting of electrocardiogram signals measured for each of multiple patients and patient data of multiple patients. At this time, the electrocardiogram signal and the patient data can be matched for the same patient. In particular, the electrocardiogram signal can include an electrocardiogram signal of a patient with potential atrial fibrillation. Meanwhile, the learning data set can be stored in the memory (120), and the processor (110) can learn the neural network model in advance based on the learning data set stored in the memory (120) and then store it in the memory (120).

At this time, the pre-learned neural network model can be pre-learned to calculate the probability value of the patient's paroxysmal atrial fibrillation when at least one of an electrocardiogram signal or patient information is input. The probability value of the patient's paroxysmal atrial fibrillation can be a probability value that the patient will experience paroxysmal atrial fibrillation.

The trained neural network model can determine whether a patient has paroxysmal atrial fibrillation based on the generated probability values.

Alternatively, if a probability value is derived from a pre-learned neural network model, the processor (110) can determine whether the patient has paroxysmal atrial fibrillation based on the derived probability value. If the derived probability value is greater than or equal to a preset value, the processor (110) can identify paroxysmal atrial fibrillation in the patient. In other words, the processor (110) can identify that paroxysmal atrial fibrillation, which is not observed in the electrocardiogram signal, is occurring in the patient.

FIG. 3 is an exemplary diagram showing the configuration of a pre-learned neural network model (200) based on an electrocardiogram signal and information on a subject of diagnosis according to one embodiment of the present disclosure.

A neural network model (200) according to an embodiment of the present disclosure may include a neural network (hereinafter, a first neural network) (210) that extracts features of an electrocardiogram signal, a second neural network (hereinafter, a second neural network) (220) that extracts demographic features based on patient information, and a neural network (hereinafter, a third neural network) (230) that is connected to the first and second neural networks and determines whether a patient with an embolic stroke of unknown etiology has paroxysmal atrial fibrillation based on the features of the electrocardiogram signal and demographic features. Here, the first to third neural networks (210, 220, and 230) may each be referred to as a sub-model in that they constitute a part of the neural network model (200), or may also be referred to as an architecture, a block, etc.

For example, referring to FIG. 3, when the patient's characteristic information is input, the second neural network (220) extracts demographic potential features. At this time, when the patient's characteristic information is input, the second neural network (220) linearly transforms the patient's characteristic information through a fully connected layer, transforms it into one-dimensional data, batch normalizes the transformed one-dimensional data, and extracts demographic potential features from the patient's characteristic information using the ReLU (Rectified Linear Unit) function.

When an electrocardiogram signal is input, the first neural network (210) extracts potential features of the electrocardiogram signal. At this time, the first neural network (210) may include a plurality of residual blocks. For example, the first neural network (210) may include five residual blocks (211). The residual blocks may include a first sub-block (211-1) that receives an electrocardiogram signal (or, when a plurality of residual blocks are sequentially connected, output data of a previously arranged residual block) and a second sub-block (212-1) connected to the first sub-block (211-1). At this time, the first and second sub-blocks (211-1 and 211-2) may each have a skip-connection structure that connects an input terminal and an output terminal. Through the skip connection structure, the neural network model can be trained through a residual learning process that adds the input value to the output value output through multiple layers, thereby eliminating the gradient loss that occurs as the layers of the neural network model become deeper.

Meanwhile, the skip connection structure of the first sub-block (211-1) can connect the input terminal of the first sub-block (211-1) and the output terminal of the second batch normalization of the first sub-block (211-1). In addition, the skip connection structure of the first sub-block (211-1) can be connected to a one-dimensional convolutional layer and a maximum pooling layer. That is, in the case of the first sub-block (211-1), the input value (or input data) of the first sub-block (211-1) is added to the output value (or output data) output through the second batch normalization after passing through the one-dimensional convolutional layer and the maximum pooling layer. On the other hand, the skip connection structure of the second sub-block (211-2) connects the input terminal of the second sub-block (211-2) and the output terminal of the second batch normalization of the second sub-block (211-2), and the input value (or input data) of the second sub-block (211-2) is added as is to the output value (or output data) output through the second batch normalization.

Meanwhile, when the potential features of the electrocardiogram signal and the potential demographic features are output from the first and second neural networks (210 and 220), respectively, the potential features of the electrocardiogram signal and the potential demographic features are concatenated and input to the third neural network (230) connected to the first and second neural networks (210 and 220). At this time, the third neural network (230) can perform the input classification task. Specifically, when the potential demographic features and the potential features of the electrocardiogram signal are input, the third neural network (230) can determine whether the patient has paroxysmal atrial fibrillation. The third neural network (230) can calculate a probability value that the patient will have paroxysmal atrial fibrillation.

According to one embodiment of the present disclosure, the first neural network (210) can extract features of an electrocardiogram signal based on at least one of the PR interval, the QRS amplitude, and the QT interval of the electrocardiogram signal. In particular, in the case of an electrocardiogram signal of atrial fibrillation, a long PR interval, a high QRS amplitude, and a long QT interval are observed. Therefore, the neural network model can predict whether a patient has paroxysmal atrial fibrillation by extracting features of the electrocardiogram signal in relation to at least one of the PR interval, the QRS amplitude, and the QT interval of the electrocardiogram signal through the first neural network (210). At this time, the neural network model (200) (or the third neural network (230)) can calculate a higher probability value when the PR interval is long, the QRS amplitude is high, and the QT interval is long.

When the processor (110) identifies the patient's paroxysmal atrial fibrillation based on the probability value acquired through the pre-learned neural network model (200), the processor (110) can determine that the patient's unexplained embolic stroke is unusual for paroxysmal atrial fibrillation and generate diagnostic information. For example, the processor (110) can generate diagnostic information that the patient's unexplained embolic stroke is caused by paroxysmal atrial fibrillation and suggest oral anticoagulation therapy to the patient.

FIG. 4 is a diagram showing the performance of a pre-learned neural network model (200) according to one embodiment of the present disclosure.

Referring to FIG. 4, the performance of the neural network model (200) according to an embodiment of the present disclosure was compared with the performance of the patient's left atrial diameter (LAD), which is one of the factors of paroxysmal atrial fibrillation, the patient's atrial ectopic burden value, the known CHARGE-AF model, the C2HEST model, and the HATCH model. To this end, the electrocardiogram signals and patient information of multiple patients with insertable cardiac monitors (ICMs) attached were input into the neural network model (200), and the results of identifying the patients' paroxysmal atrial fibrillation were compared to the results of identifying the patients' paroxysmal atrial fibrillation based on the ICM data to determine whether they matched. At this time, the neural network model (200) according to an embodiment of the present disclosure is superior to other models in terms of the area under the curve (AUC) (0.827), sensitivity (0.824), and specificity (0.807).

Meanwhile, according to an embodiment of the present disclosure, the processor (110) may determine whether the patient has paroxysmal atrial fibrillation based on the patient's left atrial diameter and the patient's atrial ectopic load value in addition to the probability value obtained from the pre-learned neural network model (200). Specifically, the processor (110) may obtain the probability value, the patient's left atrial diameter, and the patient's atrial ectopic load value, calculate a score based on the probability value, the patient's left atrial diameter, and the patient's atrial ectopic load value, and determine whether the patient has paroxysmal atrial fibrillation based on the calculated score. The patient's left atrial diameter may be the length between the anterior and posterior walls of the left atrium measured in the diaphragmatic long axis plane at the end of ventricular systole using ultrasound. In addition, the atrial ectopic load may be a value obtained by multiplying the number of QRS complexes performed in the atrial envelope by 100 compared to the total number of QRS complexes of the patient monitored for 24 hours. At this time, the score can be calculated as a higher value as the probability value increases, the patient's left atrial diameter increases, and the patient's atrial ectopic load value increases. Meanwhile, the processor (110) can calculate the score by applying weights to the probability value, the patient's left atrial diameter, and the patient's atrial ectopic load value, respectively.

The processor (110) determines that the patient is suffering from paroxysmal atrial fibrillation if the probability value obtained from the pre-learned neural network model (200) is greater than or equal to a first value, and if the probability value is less than the first value and greater than or equal to a second value, the processor (110) may determine whether the patient is suffering from paroxysmal atrial fibrillation based on the patient's left atrial diameter and the patient's atrial ectopic load value in addition to the probability value.

Meanwhile, the processor (110) inputs the probability value (hereinafter, the first probability value) obtained from the acquired pre-learned neural network model (200), the patient's left atrial diameter, and the patient's atrial ectopic load value into a logistic regression model, thereby obtaining a final probability value (hereinafter, the second probability value) of whether the patient with an embolic stroke of unknown cause has paroxysmal atrial fibrillation, and may determine whether the patient with an embolic stroke of unknown cause has paroxysmal atrial fibrillation based on the obtained second probability value. At this time, the logistic regression model may be pre-learned based on the first probability value, the patient's left atrial diameter, and the patient's ectopic load value. Meanwhile, the processor (110) may obtain the second probability value by using a multi-layer perceptron (MLP), a convolutional neural network (CNN), a recurrent neural network (RNN), etc., learned based on the first probability value, the patient's left atrial diameter, and the patient's ectopic load value, without being limited thereto.

At this time, the processor (110) can determine that the patient is experiencing paroxysmal atrial fibrillation if the second probability value is greater than or equal to a preset third value. At this time, the third value can be set to a value higher than the first value described above, which is a judgment criterion for the first probability value obtained from the neural network model (200).

FIG. 5 is an exemplary diagram showing the configuration of a pre-learned neural network model based on electrocardiogram signals and brainwave signals according to one embodiment of the present disclosure.

Meanwhile, according to one embodiment of the present disclosure, the processor (110) can identify whether the patient has potential atrial fibrillation based on an electroencephalogram (EEG) signal in addition to the patient's electrocardiogram signal. To this end, the learned model may be a model learned to identify whether the patient has potential atrial fibrillation based on an electrocardiogram signal and an EEG signal.

Specifically, the processor (110) can obtain the patient's electrocardiogram signal and brain wave signal. Here, the brain wave signal is a signal obtained by measuring the electrical activity of the patient's brain, and may be obtained from an external measuring device capable of measuring brain waves through the network unit (130), or may be obtained through a sensing unit of a computing device (for example, at least one electrode included in the sensing unit).

The processor (110) inputs the acquired patient's electrocardiogram signal and the brainwave signal into a pre-learned neural network model, thereby identifying atrial fibrillation of the subject of diagnosis. At this time, the pre-learned neural network model may be a model learned in advance based on a learning data set including electrocardiogram signals and brainwave signals of a plurality of patients, and the plurality of patients may be patients with potential atrial fibrillation.

Meanwhile, according to one embodiment of the present disclosure, the pre-learned neural network model may include a plurality of neural networks (fourth and fifth neural networks) that extract features corresponding to electrocardiogram signals and brainwave signals, respectively, and a neural network (sixth neural network) that performs a classification task regarding potential atrial fibrillation based on the extracted features.

For example, referring to FIG. 5, when an EEG signal is input, the fourth neural network (310) extracts potential features of the EEG signal. At this time, the fourth neural network (310) may include a plurality of residual blocks. For example, the fourth neural network (310) may include five residual blocks (311). The residual blocks may include a third sub-block (311-1) that receives an EEG signal (or, when a plurality of residual blocks are sequentially connected, output data of a previously arranged residual block) and a fourth sub-block (312-1) connected to the third sub-block (311-1). At this time, the third and fourth sub-blocks (311-1 and 311-2) may each have a skip-connection structure that connects an input terminal and an output terminal. Through the skip connection structure, the neural network model can be trained through a residual learning process that adds the input value to the output value output through multiple layers, thereby eliminating the gradient loss that occurs as the layers of the neural network model become deeper.

Meanwhile, the skip connection structure of the third sub-block (311-1) can connect the input terminal of the third sub-block (211-1) and the output terminal of the second batch normalization of the third sub-block (311-1). In addition, the skip connection structure of the third sub-block (311-1) can be connected to a one-dimensional convolutional layer and a max pooling layer. That is, in the case of the third sub-block (311-1), the input value (or input data) of the third sub-block (311-1) is added to the output value (or output data) output through the second batch normalization after passing through the one-dimensional convolutional layer and the max pooling layer. On the other hand, the skip connection structure of the fourth sub-block (311-2) connects the input terminal of the fourth sub-block (311-2) and the output terminal of the second batch normalization of the fourth sub-block (311-2), and the input value (or input data) of the fourth sub-block (411-2) is added as is to the output value (or output data) output through the second batch normalization.

Meanwhile, the fifth neural network (320) extracting potential features of the electrocardiogram signal in FIG. 5 may be identical to the first neural network (210) extracting potential features of the electrocardiogram signal illustrated in FIG. 3, and therefore the description of the first neural network (210) based on FIG. 3 may be applied equally. Therefore, a detailed description is omitted.

Meanwhile, when the potential features of the electrocardiogram signal and the brainwave signal are output from the fourth and fifth neural networks (310 and 320), respectively, the potential features of the electrocardiogram signal and the brainwave signal are concatenated and input to the sixth neural network (330) connected to the fourth and fifth neural networks (310 and 320). At this time, the sixth neural network (330) can perform the input classification task. Specifically, when the potential features of the brainwave signal and the potential features of the electrocardiogram signal are input, the sixth neural network (330) can determine whether the patient has paroxysmal atrial fibrillation. The sixth neural network (330) can calculate a probability value that the patient will have paroxysmal atrial fibrillation.

Meanwhile, although the pre-learned model (the first neural network model) for identifying potential atrial fibrillation (e.g., paroxysmal atrial fibrillation) based on the latent features of the electrocardiogram signal and the demographic features illustrated in FIG. 3 and the pre-learned neural network model (the second neural network model) for identifying potential atrial fibrillation (e.g., paroxysmal atrial fibrillation) based on the latent features of the electrocardiogram signal and the latent features of the EEG signal illustrated in FIG. 5 are described as separate components, the first and second neural network models may be parts of the same pre-learned neural network model (the third neural network model). That is, the pre-learned third neural network model may include a neural network for extracting latent features of the electrocardiogram signal as an input value, a neural network for extracting latent features of the electrocardiogram signal as an input value, a neural network for extracting latent features of the EEG signal as an input value, and a neural network for extracting demographic features of the subject information as an input value. And, the pre-trained third neural network model may include a neural network (e.g., a classifier) that performs a classification task (i.e., a task regarding potential atrial fibrillation) based on latent features of the electrocardiogram signal and latent features of the electrocardiogram signal and a neural network (e.g., a classifier) that performs a classification task (i.e., a task regarding potential atrial fibrillation) based on latent features of the electrocardiogram signal and demographic features. In addition, the pre-trained third neural network model may include a neural network that performs a classification task based on latent features of the electrocardiogram signal, latent features of the electrocardiogram signal and demographic features.

At this time, the processor may select a neural network that extracts features based on the acquired signal (ECG signal or EEG signal) or information (information on the subject of diagnosis) and a neural network that selects a classification task to determine whether there is potential atrial fibrillation. Alternatively, the processor may generate multiple inputs by combining the acquired signal (ECG signal or EEG signal) and information (information on the subject of diagnosis), and input the generated inputs into a pre-trained third neural network model to obtain a probability value regarding potential atrial fibrillation corresponding to each input. In this case, the processor (110) may determine the potential atrial fibrillation of the patient based on the multiple probability values. For example, the processor (110) may determine the potential atrial fibrillation of the patient based on an average value or a highest value of the multiple probability values.

Meanwhile, the pre-learned third neural network model may be a model learned to identify not only the presence or absence of paroxysmal atrial fibrillation based on the input electrocardiogram signal and information on the subject of diagnosis, but also whether the subject of diagnosis is a patient with an embolic stroke of unknown cause. At this time, the pre-learned model can also produce a probability value corresponding to the subject of diagnosis being a patient with an embolic stroke.

FIG. 6 is a detailed block diagram of a computing device (100) according to one embodiment of the present disclosure.

Referring to FIG. 6, a computing device (100) according to an embodiment of the present disclosure includes a processor (110), a memory (120), a network unit (130), a display (140), a user interface (150), and a sensing unit (160). A detailed description of any configurations illustrated in FIG. 6 that overlap with those illustrated in FIG. 2 will be omitted.

The display (140) can display various images. Here, the images include both still images and moving images. The display (140) can also output electrocardiogram signals, results of determining whether a patient has paroxysmal atrial fibrillation, etc. The display (140) can be implemented as various types of displays, such as an LCD (Liquid Crystal Display Panel), an OLED (Organic Light Emitting Diodes), an LCoS (Liquid Crystal on Silicon), a DLP (Digital Light Processing), etc. In addition, the display (140) can also include a driving circuit, a backlight unit, etc., which can be implemented in a form, such as an a-si TFT, an LTPS (low temperature poly silicon) TFT, an OTFT (organic TFT), etc.

Meanwhile, the display (140) may be implemented as a touch screen by being combined with a touch panel, and in this case, the display (140) may perform the function of an input interface that receives a user's touch input as well as an output interface that outputs an image through the touch screen.

The user interface (150) is a configuration used by the computing device (100) to perform interaction with the user, and may include at least one of a touch sensor, a motion sensor, a button, a jog dial, and a switch, but is not limited thereto. The processor (110) may receive patient information through the user interface (150).

The sensing unit (160) senses the patient's bio-information to obtain the patient's bio-signal. For example, the sensing unit (160) may obtain the patient's electrocardiogram signal. To this end, the sensing unit (160) may include at least one electrode, etc. In addition, the sensing unit (160) may measure the patient's weight, height, etc. In addition, the sensing unit (160) may perform an EEG test on the patient and obtain the EEG test results.

The various embodiments of the present disclosure described above can be combined with additional embodiments and can be modified within a scope that can be understood by those skilled in the art in light of the detailed description described 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 form. Accordingly, all changes or modifications derived from the meaning, scope, and equivalent concept of the claims of the present disclosure should be interpreted as being included in the scope of the present disclosure.

Claims (15)

A method for identifying potential atrial fibrillation based on an artificial intelligence model, performed by a computing device comprising at least one processor, A step of acquiring an electrocardiogram signal of a subject of diagnosis; and A step of inputting the acquired electrocardiogram signal into a pre-trained neural network model to identify potential atrial fibrillation of the subject of diagnosis; Including, method. In the first paragraph, The above obtaining steps are: Further comprising a step of obtaining an electrocardiogram signal and information on the subject of diagnosis; The above identifying step is, A step of inputting the acquired electrocardiogram signal of the subject of diagnosis and the information of the subject of diagnosis into a pre-learned neural network model to identify atrial fibrillation of the subject of diagnosis; more inclusive, method. In the second paragraph, The above diagnosis subjects are, Patients with idiopathic cerebral stroke, The above potential atrial fibrillation is, Including paroxysmal atrial fibrillation, The above identifying step is, A step of inputting the acquired electrocardiogram signal and information on the subject of diagnosis into the neural network model to identify whether the patient with an idiopathic embolic stroke has paroxysmal atrial fibrillation, and diagnosing the patient based on the identification result; Including, method. In the third paragraph, The above neural network model is, A first neural network for extracting features of the above electrocardiogram signal; A second neural network that extracts demographic characteristics based on the above diagnostic subject information; and A third neural network, connected to the first and second neural networks, for determining whether the patient with idiopathic embolic stroke has paroxysmal atrial fibrillation based on the characteristics of the electrocardiogram signal and the demographic characteristics; Including, method. In paragraph 4, The above identifying step is, A step of inputting the electrocardiogram signal into the first neural network, inputting the information on the subject of diagnosis into the second neural network, obtaining a probability value of the paroxysmal atrial fibrillation through the third neural network, and diagnosing the patient with an embolic stroke of unknown cause based on the obtained probability value; Including, method. In paragraph 4, The above first neural network, Contains multiple residual blocks, The above remaining blocks are, A first sub-block including a skip connection structure connected to a convolutional layer and a max pooling layer; and A second sub-block connected to the first sub-block and including a skip connection structure; Including, method. In paragraph 4, The above first neural network, Extracting features of the electrocardiogram signal based on at least one of the PR interval, QRS amplitude and QT interval of the electrocardiogram signal. method. In Article 4 The above third neural network is, The longer the PR interval of the above electrocardiogram signal, the larger the QRS amplitude, and the longer the QT interval, the higher the probability value of paroxysmal atrial fibrillation is output. method. In the first paragraph, The above identifying step is, A step of inputting the acquired electrocardiogram signal information into the neural network model to obtain the first probability value of the atrial fibrillation; and A step of identifying atrial fibrillation of the subject of diagnosis based on the acquired first probability value, the left atrial diameter (LAD) of the subject of diagnosis, and the atrial ectopic load value of the subject of diagnosis; Including, method. In Article 9, The above identifying step is, A step of inputting the first probability value obtained above, the left atrial diameter of the subject of diagnosis, and the atrial ectopic load value of the subject of diagnosis into a logistic regression model to obtain a second probability value of whether the subject of diagnosis has atrial fibrillation; and A step of determining whether the subject of diagnosis has paroxysmal atrial fibrillation based on the second probability value obtained above; Including, method. In the second paragraph, The above diagnostic subject information is: Including at least one of the following information: age, sex, height, weight, and brain imaging test results; method. In the first paragraph, The above obtaining steps are: Further comprising a step of obtaining electrocardiogram signals and brain wave signals of the subject of the diagnosis; The above identifying step is, A step of inputting the electrocardiogram signal and the brainwave signal of the subject of diagnosis obtained above into a pre-learned neural network model to identify the atrial fibrillation of the subject of diagnosis; more inclusive, method. In Article 12, The above neural network model is, A fourth neural network for extracting features corresponding to the electrocardiogram signal and the brain wave signal, respectively; and a fifth neural network; and A method comprising: a sixth neural network for performing a classification task regarding potential atrial fibrillation based on the extracted features. A computer program stored in a computer-readable storage medium, wherein the computer program, when executed on one or more processors, performs an operation of identifying potential atrial fibrillation based on an artificial intelligence model; The above actions are, The act of acquiring an electrocardiogram signal of a subject of diagnosis; and An operation of inputting the acquired electrocardiogram signal into a pre-trained neural network model to identify atrial fibrillation of the subject of the diagnosis; Including, Computer program. A computing device for identifying potential atrial fibrillation based on an artificial intelligence model, A processor comprising at least one core; a memory containing program codes executable by the processor; and including a network unit; The above processor, Obtaining an electrocardiogram signal of a subject of diagnosis, and inputting the obtained electrocardiogram signal into a pre-trained neural network model to identify atrial fibrillation of the subject of diagnosis. device.

Images