Electrocardiogram analysis device, electrocardiogram analysis method, and program

Abstract

To make it easy to specify whether or not an object person of electrocardiogram analysis has paroxysmal arrhythmia.SOLUTION: An electrocardiogram analysis device 3 includes: a machine learning unit 33 having a machine learning model that has performed machine learning using electrocardiogram data for a teacher in a non-attack period in which an attack of paroxysmal arrhythmia is not occurring, of a patient having paroxysmal arrhythmia; an input processing unit 341 for inputting electrocardiogram data on a person to be analyzed, who is an analysis object, to the machine learning model; and an output control unit 343 for outputting, to an information terminal, abnormality information on whether or not the person to be analyzed has paroxysmal arrhythmia, output from the machine learning model.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to an electrocardiogram analyzer, an electrocardiogram analysis method, and a program for analyzing an electrocardiogram. [Background technology]

[0002] BACKGROUND ART Holter electrocardiographs that are worn by a patient and allow electrocardiograms to be measured over a long period of time are known (see, for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2007-195693 Summary of the Invention [Problem to be solved by the invention]

[0004] Waveform abnormalities that can be evaluated using an electrocardiogram include paroxysmal arrhythmias. Paroxysmal arrhythmias occur at a low frequency, for example, once a day to once every few months. Specific examples of paroxysmal arrhythmias include paroxysmal tachycardias such as paroxysmal atrial fibrillation, ventricular tachycardia, supraventricular tachycardia, and atrial flutter, as well as ventricular fibrillation. When no paroxysmal arrhythmias are occurring, an electrocardiogram exhibits a sinus rhythm waveform. Therefore, even if the electrocardiogram of a subject for electrocardiogram analysis is measured over a long period of time (e.g., 24 hours) using a Holter monitor, the electrocardiogram may not necessarily include a waveform of a paroxysmal arrhythmia, making it difficult to determine whether the subject has a paroxysmal arrhythmia.

[0005] The present invention has been made in consideration of these points, and aims to make it easier to identify whether or not a subject of electrocardiogram analysis has a spasmodic arrhythmia. [Means for solving the problem]

[0006] The electrocardiogram analysis device of the first aspect of the present invention comprises a machine learning unit having a machine learning model trained by machine learning using teacher electrocardiogram data from a patient with a paroxysmal arrhythmia during a non-attack period when the patient is not experiencing an attack of the paroxysmal arrhythmia, an input processing unit that inputs electrocardiogram data of a subject to be analyzed into the machine learning model, and an output control unit that outputs abnormality information regarding whether the subject has the paroxysmal arrhythmia output from the machine learning model to an information terminal.

[0007] The machine learning unit may have a machine learning model that has been machine-learned using the teacher electrocardiogram data during the non-attack period, which is a predetermined period of at least one of before and after the attack period in which the spasmodic arrhythmia is identified as occurring.

[0008] The machine learning unit may have a machine learning model that has been machine-trained using the teacher electrocardiogram data of the patient who has at least one of paroxysmal atrial fibrillation, ventricular tachycardia, supraventricular tachycardia, atrial flutter, and ventricular fibrillation as the paroxysmal arrhythmia.

[0009] The input processing unit may input the electrocardiogram data measured at the same sampling rate as the teacher electrocardiogram data to the machine learning model.

[0010] The output control unit may output the abnormality information indicating the probability that the subject has the paroxysmal arrhythmia to the information terminal.

[0011] The input processing unit may input multiple pieces of electrocardiogram data measured over multiple different periods into the machine learning model, and the output control unit may output the abnormality information regarding changes in the probability over the multiple different periods to the information terminal.

[0012] The input processing unit may input the electrocardiogram data measured by an electrocardiograph worn by the subject during daily life into the machine learning model.

[0013] The output control unit may output to the information terminal the abnormality information representing at least one of the value, degree or score of the probability that the subject has developed the episodic arrhythmia, the value, degree or score of the probability that the subject has not developed the episodic arrhythmia, whether or not the subject has the episodic arrhythmia, and whether or not the subject has signs of the episodic arrhythmia.

[0014] The machine learning unit may receive determination information indicating whether the subject has the spontaneous arrhythmia from the information terminal to which the abnormality information is output, and regenerate the machine learning model by performing machine learning using the determination information and the electrocardiogram data.

[0015] A second aspect of the electrocardiogram analysis method of the present invention includes the steps of: obtaining a machine learning model, executed by a computer, that is trained by machine learning using teacher electrocardiogram data from a patient with episodic arrhythmia during a non-attack period when the patient is not experiencing an attack of the episodic arrhythmia; inputting electrocardiogram data of the subject to be analyzed into the machine learning model; and outputting abnormality information output from the machine learning model regarding whether the subject has the episodic arrhythmia to an information terminal.

[0016] A third aspect of the program of the present invention causes a computer to function as a machine learning unit having a machine learning model trained by machine learning using teacher electrocardiogram data from a patient with episodic arrhythmia during a non-attack period in which the patient is not experiencing an attack of the episodic arrhythmia, an input processing unit that inputs electrocardiogram data of a subject to be analyzed into the machine learning model, and an output control unit that outputs abnormality information regarding whether the subject has the episodic arrhythmia, which is output from the machine learning model, to an information terminal. [Effects of the Invention]

[0017] The present invention provides an advantage in that it is possible to easily identify whether or not a subject of electrocardiogram analysis has a spasmodic arrhythmia. [Brief explanation of the drawings]

[0018] [Figure 1] 1 is a diagram for explaining an overview of an electrocardiogram analysis system according to an embodiment. FIG. [Figure 2] 1 is a block diagram of an electrocardiogram analysis system according to an embodiment. [Figure 3] FIG. 1 is a schematic diagram for explaining a method for performing machine learning in an electrocardiogram analyzer. [Figure 4] FIG. 2 is a schematic diagram for explaining a method in which the electrocardiogram analyzer analyzes electrocardiogram data. [Figure 5] FIG. 10 is a schematic diagram of an analysis result screen displayed on the doctor terminal. [Figure 6] FIG. 2 is a flowchart illustrating an electrocardiogram analysis method executed by the electrocardiogram analysis system according to the embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0019] [Outline of the Electrocardiogram Analysis System S] 1 is a diagram for explaining an overview of an electrocardiogram analysis system S according to this embodiment. The electrocardiogram analysis system S includes an electrocardiograph 1, a doctor terminal 2, and an electrocardiogram analyzer 3. A plurality of electrocardiographs 1 and a plurality of doctor terminals 2 may be provided. The electrocardiogram analysis system S may also include other devices such as a server and a terminal.

[0020] The electrocardiograph 1 is an electrocardiograph worn by a human, and is an electrocardiogram measuring device that generates electrocardiogram data showing the human electrocardiogram by measuring electrical potential while worn on, for example, the human wrist, palm, or chest. That is, the electrocardiograph 1 is a Holter electrocardiograph (also called a wearable electrocardiograph or a continuously worn electrocardiograph). The electrocardiograph 1 transmits the generated electrocardiogram data to the electrocardiogram analyzer 3 via a network N including a wireless communication line. The electrocardiogram data generated by the electrocardiograph 1 may also be delivered to the electrocardiogram analyzer 3 using, for example, a storage medium, without going through the network N.

[0021] The doctor terminal 2 is an information terminal used by a medical professional such as a doctor who examines a human, and includes, for example, a display (display device) and a computer. The doctor terminal 2 is associated in advance with the medical professional who uses the doctor terminal 2 by an ID or the like assigned to the medical professional. The doctor terminal 2 outputs abnormality information output by the electrocardiogram analyzer 3 based on the electrocardiogram data generated by the electrocardiograph 1. The doctor terminal 2 may display the abnormality information on a display, or may output a sound indicating the abnormality information from a speaker.

[0022] The electrocardiogram analyzer 3 is a device, such as a server, that outputs abnormality information regarding whether or not the subject has a paroxysmal arrhythmia based on the electrocardiogram data generated by the electrocardiograph 1. The abnormality information is information indicating the value, degree, or score of the probability that the subject has developed a paroxysmal arrhythmia, or the presence or absence of symptoms of a paroxysmal arrhythmia in the subject. The abnormality information may also be information indicating the value, degree, or score of the probability that the subject has not developed a paroxysmal arrhythmia. The abnormality information may also be information indicating the presence or absence of a paroxysmal arrhythmia in the subject. The abnormality information may also be information regarding the occurrence of other paroxysmal arrhythmias. The paroxysmal arrhythmia is at least one of paroxysmal tachycardia, such as paroxysmal atrial fibrillation, ventricular tachycardia, supraventricular tachycardia, and atrial flutter, and ventricular fibrillation.

[0023] FIG. 2 is a block diagram of an electrocardiogram analysis system S according to this embodiment. In FIG. 2, arrows indicate main data flows, and there may be other data flows not shown in FIG. 2. In FIG. 2, each block indicates a functional configuration rather than a hardware (device) configuration. Therefore, the blocks shown in FIG. 2 may be implemented in a single device, or may be implemented separately in multiple devices. Data may be exchanged between blocks via any means, such as a data bus, a network, or a portable storage medium.

[0024] The electrocardiogram analyzer 3 includes a communication unit 31, a storage unit 32, a machine learning unit 33, and a control unit 34. The control unit 34 includes an input processing unit 341, a result acquisition unit 342, and an output control unit 343.

[0025] The communication unit 31 has a communication controller for transmitting and receiving data between the electrocardiograph 1 and the doctor terminal 2 via the network N. The communication unit 31 notifies the control unit 34 of data received from the electrocardiograph 1 and the doctor terminal 2 via the network N. The communication unit 31 also transmits data output from the control unit 34 to the doctor terminal 2 via the network N.

[0026] The storage unit 32 is a storage medium including a ROM (Read Only Memory), a RAM (Random Access Memory), a hard disk drive, etc. The storage unit 32 stores in advance programs to be executed by the control unit 34. The storage unit 32 may be provided outside the electrocardiogram analyzer 3, in which case data may be exchanged between the storage unit 32 and the control unit 34 via a network N.

[0027] The machine learning unit 33 generates a machine learning model that outputs abnormality information regarding whether or not input electrocardiogram data has a paroxysmal arrhythmia by learning based on teacher electrocardiogram data used as teacher data, and stores the generated machine learning model. The machine learning model is a model generated by machine learning using normal electrocardiogram data (i.e., electrocardiogram data of a person who does not have a paroxysmal arrhythmia) and abnormal electrocardiogram data (i.e., electrocardiogram data of a person who has a paroxysmal arrhythmia) as teacher data. The machine learning unit 33 may also store an externally generated machine learning model.

[0028] The internal configuration of the machine learning model is arbitrary, but may be, for example, a convolutional neural network (CNN) or a recurrent neural network (RNN). The machine learning unit 33 includes, for example, a processor that executes various calculations using the CNN and a memory that stores CNN coefficients. The machine learning model of the machine learning unit 33 outputs abnormality information regarding whether the subject has developed a sporadic arrhythmia based on the input electrocardiogram data. The machine learning unit 33 may include only a memory that stores an externally generated machine learning model. At least some of the functions of the machine learning unit 33 may be built into the control unit 34.

[0029] The control unit 34 is a processor such as a CPU (Central Processing Unit), and functions as an input processing unit 341, a result acquisition unit 342, and an output control unit 343 by executing a program stored in the storage unit 32. At least some of the functions of the control unit 34 may be performed by an electric circuit. Furthermore, at least some of the functions of the control unit 34 may be realized by the control unit 34 executing a program executed via a network.

[0030] The input processing unit 341 inputs electrocardiogram data of the subject, who is the subject of analysis, to the machine learning model possessed by the machine learning unit 33. The result acquisition unit 342 acquires information output by the machine learning model possessed by the machine learning unit 33. The output control unit 343 outputs abnormality information corresponding to whether or not the subject has a spasmodic arrhythmia, output from the machine learning model possessed by the machine learning unit 33, to the doctor terminal 2. Detailed processing performed by the input processing unit 341, the result acquisition unit 342, and the output control unit 343 will be described later.

[0031] The electrocardiogram analysis system S according to this embodiment is not limited to the specific configuration shown in Fig. 2. The electrocardiograph 1, the doctor terminal 2, and the electrocardiogram analyzer 3 may each be configured by two or more physically separate devices connected by wire or wirelessly. The electrocardiogram analyzer 3 may be configured by a single computer, by multiple computers linked together, or by a cloud, which is a collection of computer resources. Two or more of the electrocardiograph 1, the doctor terminal 2, and the electrocardiogram analyzer 3 may be configured as a single device.

[0032] [Explanation of electrocardiogram analysis method] The electrocardiogram analysis method executed by the electrocardiogram analysis system S according to this embodiment will be described in detail below. The electrocardiogram analysis device 3 generates a machine learning model by performing machine learning in advance using teacher electrocardiogram data. FIG. 3 is a schematic diagram for explaining the machine learning method used by the electrocardiogram analysis device 3. The electrocardiograph 1 measures the electrocardiogram of a patient. The patient is a person diagnosed by a doctor as having an arrhythmia that occurs spasmodically (for example, at least one of paroxysmal tachycardia, such as paroxysmal atrial fibrillation, ventricular tachycardia, supraventricular tachycardia, and atrial flutter, and ventricular fibrillation). The electrocardiograph 1 generates electrocardiogram data H0 indicating the measured electrocardiogram and transmits it to the electrocardiogram analysis device 3 via a network N.

[0033] In the electrocardiogram analyzer 3, the input processing unit 341 acquires electrocardiogram data H0 transmitted from the electrocardiograph 1. A patient with paroxysmal arrhythmia has a non-attack period T0 during which no arrhythmia attacks occur and an attack period T1 during which arrhythmia attacks occur. A doctor can identify the attack period T1 by looking at the electrocardiogram. The non-attack period T0 is a predetermined period at least either before or after the attack period T1 during which the doctor has identified an arrhythmia attack as occurring, and is a period during which the doctor has not identified an arrhythmia attack as occurring. The non-attack period T0 is preferably seven days before or after the attack period T1, and more preferably 24 hours before or after the attack period T1.

[0034] The input processing unit 341 inputs the non-attack period T0 portion of the electrocardiogram data H0 as teacher electrocardiogram data of a person with paroxysmal arrhythmia to the machine learning unit 33. The input processing unit 341 also inputs electrocardiogram data of a person who has been diagnosed by a doctor as not having paroxysmal arrhythmia to the machine learning unit 33 as teacher electrocardiogram data of a person without paroxysmal arrhythmia.

[0035] It is desirable that the input processing unit 341 uses electrocardiogram data from a predetermined period when a person is at rest (for example, while sleeping) as the teacher electrocardiogram data. It is also desirable that the input processing unit 341 uses electrocardiogram data from a period in which a doctor has identified no symptoms different from the spontaneous arrhythmia to be analyzed as the teacher electrocardiogram data. This allows the electrocardiogram analysis system S to exclude data that may be noise and improve the accuracy of machine learning.

[0036] The machine learning unit 33 performs known machine learning (for example, CNN or RNN) using the inputted teacher electrocardiogram data to generate a machine learning model that outputs abnormality information regarding whether the inputted electrocardiogram data is electrocardiogram data of a person who is suffering from a spasmodic arrhythmia. The machine learning unit 33 may also acquire a machine learning model generated by an external device (such as a server) by the above-mentioned machine learning method.

[0037] Next, the electrocardiogram analysis device 3 analyzes the electrocardiogram data of the subject, who is the subject of analysis, using the machine learning model possessed by the machine learning unit 33. FIG. 4 is a schematic diagram for explaining the method by which the electrocardiogram analysis device 3 analyzes the electrocardiogram data. The electrocardiograph 1 measures the electrocardiogram of the subject. The electrocardiograph 1 generates electrocardiogram data H1 indicating the measured electrocardiogram and transmits it to the electrocardiogram analysis device 3 via the network N.

[0038] In the electrocardiogram analyzer 3, the input processing unit 341 acquires electrocardiogram data H1 transmitted by the electrocardiograph 1. It is desirable that the input processing unit 341 sequentially acquires electrocardiogram data H1 measured by the electrocardiograph 1 (i.e., a Holter monitor or a patch-type electrocardiograph) worn by the subject during daily life. This allows the electrocardiogram analysis system S to promptly notify medical personnel of information regarding whether the subject has a sporadic arrhythmia. The input processing unit 341 may also acquire electrocardiogram data indicating the subject's electrocardiogram measured in advance at a hospital or the like, electrocardiogram data acquired from an electrocardiograph mounted on the steering wheel of a car, electrocardiogram data acquired from a 12-lead electrocardiograph, electrocardiogram data acquired from an electrocardiograph mounted on a smartwatch, etc.

[0039] The input processing unit 341 inputs the acquired electrocardiogram data H1 as electrocardiogram data to be analyzed into the machine learning model of the machine learning unit 33. Here, it is desirable that the input processing unit 341 inputs electrocardiogram data to be analyzed that has been measured at the same sampling rate (e.g., 1000 Hz) as the teacher electrocardiogram data into the machine learning model. This enables the electrocardiogram analysis device 3 to prevent the analysis results from being affected by differences in sampling rate between electrocardiogram data. When the sampling rate of the teacher electrocardiogram data and the sampling rate of the electrocardiogram data to be analyzed differ, the input processing unit 341 may perform a process to convert the sampling rate of the electrocardiogram data to be analyzed before inputting it into the machine learning model.

[0040] When electrocardiogram data is input, the machine learning model of the machine learning unit 33 outputs abnormality information regarding whether the input electrocardiogram data is electrocardiogram data of a person who is suffering from episodic arrhythmia. For example, the machine learning model outputs, as the abnormality information, abnormality information regarding whether the person who measured the input electrocardiogram data (i.e., the person being analyzed) is suffering from episodic arrhythmia. The machine learning model outputs, as the abnormality information, at least one of the value of the probability that the person being analyzed is suffering from episodic arrhythmia, the degree of probability that the person being analyzed is suffering from episodic arrhythmia, a score of the probability that the person being analyzed is suffering from episodic arrhythmia, whether or not the person being analyzed has episodic arrhythmia, and whether or not the person being analyzed has signs of episodic arrhythmia.

[0041] The result acquisition unit 342 acquires, as an analysis result, abnormality information regarding whether or not the subject has developed a paroxysmal arrhythmia, output by the machine learning model of the machine learning unit 33. The output control unit 343 outputs the abnormality information acquired by the result acquisition unit 342 to the doctor terminal 2. The abnormality information is represented by a value, degree, or score of the probability that the subject has developed a paroxysmal arrhythmia. The abnormality information may also be information representing a value, degree, or score of the probability that the subject has not developed a paroxysmal arrhythmia. The degree of probability is represented, for example, by letters or symbols such as high, low, O, or X associated with each range of probability values. The probability score is represented, for example, by a score obtained by converting the probability using a predetermined formula.

[0042] Furthermore, the abnormality information may be information indicating the presence or absence of a spontaneous arrhythmia in the subject or the presence or absence of signs of arrhythmia. The presence or absence of arrhythmia is represented, for example, by a character or symbol indicating the determination result of the machine learning model as to the presence or absence of arrhythmia or the presence or absence of signs of arrhythmia. Furthermore, the abnormality information may be output for the entire electrocardiogram data, or may be output in association with each of multiple periods constituting the electrocardiogram data. Furthermore, the abnormality information may be information regarding the occurrence of other spontaneous arrhythmias.

[0043] The doctor terminal 2 may display the abnormality information on a display or may output a sound indicating the abnormality information from a speaker. The doctor terminal 2 may also directly display the abnormality information or may convert the abnormality information into other information (such as "reexamination is required" or "medication is required" if the probability is greater than or equal to a predetermined value) and then output it.

[0044] 5 is a schematic diagram of an analysis result screen displayed by the doctor terminal 2. The analysis result screen includes an electrocardiogram H and abnormality information P. The electrocardiogram H represents at least a portion of the electrocardiogram data used in the analysis. The abnormality information P represents, for example, the value or degree of the probability that the subject has a spasmodic arrhythmia, output from a machine learning model.

[0045] In this way, by analyzing electrocardiograms using machine learning, the electrocardiogram analysis system S can detect signs of episodic arrhythmia during non-attack periods that are difficult for a doctor to detect with the naked eye, and display this as information on whether or not the subject has episodic arrhythmia. By referring to the electrocardiogram H and abnormality information P displayed on the analysis result screen, a doctor can easily determine whether or not the subject has episodic arrhythmia, and can determine whether or not to perform additional tests, etc., as necessary.

[0046] The electrocardiogram analyzer 3 may display the change in probability over a plurality of different time periods on the doctor terminal 2. In this case, in the electrocardiogram analyzer 3, the input processing unit 341 inputs each of a plurality of electrocardiogram data indicating the electrocardiogram of the subject measured by the electrocardiograph 1 over a plurality of different time periods (for example, specific days in each of a plurality of months) to the machine learning model of the machine learning unit 33 as electrocardiogram data to be analyzed. The result acquisition unit 342 acquires abnormality information output by the machine learning model of the machine learning unit 33, relating the abnormality information to whether or not the subject has developed a spasmodic arrhythmia, in association with each of the plurality of different time periods.

[0047] The output control unit 343 then displays the abnormality information regarding the change in probability over multiple different time periods acquired by the result acquisition unit 342 on the doctor terminal 2. The abnormality information represents the change in probability, for example, as a graph of the change over time or the degree of change in probability (difference, rate of change, etc.). By referring to the change in probability displayed on the analysis result screen, the doctor can determine the risk of the patient developing a sudden arrhythmia.

[0048] The analysis result screen may include an area (for example, a button or a selection box) for inputting a doctor's diagnosis of whether or not the subject has a paroxysmal arrhythmia. In this case, the doctor terminal 2 transmits the inputted content to the electrocardiogram analyzer 3 as determination information of whether or not the subject has a paroxysmal arrhythmia.

[0049] The electrocardiogram analyzer 3 receives the determination information sent by the doctor terminal 2. The machine learning unit 33 performs the above-mentioned machine learning using the received determination information and the electrocardiogram data of the subject to regenerate a machine learning model that outputs abnormality information regarding whether the input electrocardiogram data is electrocardiogram data of a person suffering from a spasmodic arrhythmia. This allows the electrocardiogram analysis system S to receive feedback on the doctor's diagnosis results and improve the accuracy of the abnormality information output by the machine learning model.

[0050] [Flowchart of electrocardiogram analysis method] FIG. 6 is a flowchart illustrating an electrocardiogram analysis method executed by the electrocardiogram analysis system S according to this embodiment. This embodiment includes the electrocardiogram analysis method illustrated in FIG. 6, a program for executing the electrocardiogram analysis method, and a computer-readable recording medium storing the program. In the electrocardiogram analyzer 3, the input processing unit 341 acquires electrocardiogram data transmitted from the electrocardiograph 1 of a patient suffering from paroxysmal arrhythmia (S11). The input processing unit 341 inputs a non-episode period portion of the patient's electrocardiogram data to the machine learning unit 33 as teacher electrocardiogram data for a person with paroxysmal arrhythmia. The input processing unit 341 also inputs electrocardiogram data of a person diagnosed by a doctor as not suffering from paroxysmal arrhythmia to the machine learning unit 33 as teacher electrocardiogram data for a person without paroxysmal arrhythmia.

[0051] The machine learning unit 33 performs machine learning using the input teacher electrocardiogram data to generate a machine learning model that outputs abnormality information regarding whether the input electrocardiogram data is electrocardiogram data of a person who is suffering from a spastic arrhythmia (S12). When the machine learning unit 33 uses an externally generated machine learning model, steps S11 to S12 may be omitted.

[0052] The input processing unit 341 acquires electrocardiogram data transmitted by the electrocardiograph 1 of the subject to be analyzed (S13). The input processing unit 341 inputs the acquired electrocardiogram data to the machine learning model of the machine learning unit 33 as electrocardiogram data to be analyzed (S14). When the electrocardiogram data is input, the machine learning model of the machine learning unit 33 outputs abnormality information regarding whether the input electrocardiogram data is electrocardiogram data of a person who is suffering from a spasmodic arrhythmia. The result acquisition unit 342 acquires the abnormality information output by the machine learning model of the machine learning unit 33 regarding whether the subject is suffering from a spasmodic arrhythmia as an analysis result (S15). The output control unit 343 causes the doctor terminal 2 to display the abnormality information regarding whether the result acquisition unit 342 has acquired it (S16).

[0053] [Effects of the embodiment] Because paroxysmal arrhythmias occur at low frequency, a patient's electrocardiogram measured within a limited period of time may not necessarily include a waveform of the arrhythmia. However, the electrocardiogram of a patient with paroxysmal arrhythmia may contain signs of arrhythmia that are difficult for a doctor to detect, even during periods when no attacks are occurring. Therefore, the electrocardiogram analysis system S inputs the electrocardiogram data of the subject into a machine learning model that has machine-learned the electrocardiogram data of patients with paroxysmal arrhythmias during periods when no attacks are occurring, and outputs abnormality information regarding whether the subject has paroxysmal arrhythmia. This makes it easier for the electrocardiogram analysis system S to determine whether the subject has paroxysmal arrhythmia.

[0054] [Variations] In the above-described embodiment, the electrocardiogram analyzer 3 is used to assist a doctor in making a diagnosis, but it may also be used for other purposes. For example, the electrocardiogram analyzer 3 may be used in a health check result display system that displays, based on abnormality information, whether or not the subject is likely to have a sudden arrhythmia and whether or not a re-examination or detailed examination is necessary, an insurance examination support system that includes abnormality information in the criteria for the subject's insurance application examination, or an insurance contract document device that includes abnormality information in the display items on the subject's insurance contract document. In this way, the electrocardiogram analyzer 3 can assist in insurance application examinations, etc., based on the subject's health condition.

[0055] The electrocardiogram analyzer 3 may also be used in a clinical trial participant selection system that includes abnormality information in the display items of information about clinical trial candidates, thereby enabling the electrocardiogram analyzer 3 to assist in decisions such as excluding candidates who may have spasmodic arrhythmia from the clinical trial participants for electrocardiogram evaluation.

[0056] The electrocardiogram analyzer 3 may also be used in a drug administration decision support system or drug administration contraindication decision support device that uses abnormality information, which enables the electrocardiogram analyzer 3 to, for example, support the decision on whether to administer a drug that cannot be administered in the presence of arrhythmia to a patient who may have a paroxysmal arrhythmia, or to use the abnormality information as a digital biomarker for classifying patients with paroxysmal diseases.

[0057] The electrocardiogram analyzer 3 may also be used in an automobile driving control device that safely stops a vehicle when it detects the occurrence of a sudden arrhythmia while driving the vehicle based on abnormality information.The electrocardiogram analyzer 3 may also be used in an alert system that recommends visiting a medical institution by displaying the occurrence of a sudden arrhythmia on a smart device such as a smartphone based on abnormality information.

[0058] The present invention has been described above using embodiments, but the technical scope of the present invention is not limited to the scope described in the above embodiments, and various modifications and changes are possible within the scope of the gist of the present invention. For example, all or part of the device can be configured by functionally or physically distributing or integrating any unit. Furthermore, new embodiments resulting from any combination of multiple embodiments are also included in the embodiments of the present invention. The effects of the new embodiments resulting from the combination also have the effects of the original embodiments.

[0059] The processor of the electrocardiogram analyzer 3 is responsible for each step (process) included in the electrocardiogram analysis method shown in Fig. 6. That is, the processor of the electrocardiogram analyzer 3 reads a program for executing the electrocardiogram analysis method shown in Fig. 6 from a storage unit, and executes the program to control each unit of the electrocardiogram analysis system S, thereby executing the electrocardiogram analysis method shown in Fig. 6. Some of the steps included in the electrocardiogram analysis method shown in Fig. 6 may be omitted, the order of the steps may be changed, or multiple steps may be performed in parallel. [Explanation of symbols]

[0060] S Electrocardiogram Analysis System 1 electrocardiograph 2. Doctor's terminal 3. Electrocardiogram analyzer 31 Communications Department 32 Storage section 33 Machine Learning Department 34 Control Unit 341 Input Processing Unit 342 Result acquisition part 343 Output control section

Claims

1. a machine learning unit having a machine learning model that is machine-learned using first teacher electrocardiogram data of a patient determined to have a paroxysmal arrhythmia during a non-attack period in which the patient is identified as not experiencing an abnormal condition due to the paroxysmal arrhythmia, and second teacher electrocardiogram data of a person determined not to have the paroxysmal arrhythmia and identified as not experiencing a predetermined symptom different from the paroxysmal arrhythmia; an input processing unit that inputs electrocardiogram data of a subject to be analyzed into the machine learning model; an output control unit that outputs abnormality information regarding whether the subject has the paroxysmal arrhythmia output from the machine learning model; and An electrocardiogram analyzer having:

2. the machine learning unit has the machine learning model that has been machine-learned using the first teacher electrocardiogram data in the non-attack period, which is a predetermined period that is at least one of before and after an attack period identified as an occurrence of an attack of the paroxysmal arrhythmia; 2. The electrocardiogram analyzer according to claim 1.

3. the machine learning unit has the machine learning model that has been machine-trained using the first teacher electrocardiogram data of the patient who has been determined to have at least one of paroxysmal atrial fibrillation, ventricular tachycardia, supraventricular tachycardia, atrial flutter, and ventricular fibrillation as the paroxysmal arrhythmia; 3. The electrocardiogram analyzer according to claim 1 or 2.

4. the input processing unit inputs the electrocardiogram data measured at the same sampling rate as the first teacher electrocardiogram data and the second teacher electrocardiogram data into the machine learning model; 4. The electrocardiogram analyzer according to claim 1.

5. The input processing unit inputs the electrocardiogram data measured by an electrocardiograph worn by the subject during daily life into the machine learning model.

5. The electrocardiogram analyzer according to claim 1.

6. the output control unit outputs the abnormality information indicating at least one of a value, degree, or score of the probability that the subject has developed the episodic arrhythmia, a value, degree, or score of the probability that the subject has not developed the episodic arrhythmia, whether or not the subject has the episodic arrhythmia, and whether or not the subject has a symptom of the episodic arrhythmia.

6. The electrocardiogram analyzer according to claim 1.

7. the machine learning unit receives, from the information terminal to which the abnormality information is output, determination information indicating whether or not the subject has the irregular arrhythmia, and performs machine learning using the determination information and the electrocardiogram data to regenerate the machine learning model.

7. The electrocardiogram analyzer according to claim 1.

8. The computer executes A step of acquiring a machine learning model by machine learning using first teacher electrocardiogram data of a patient determined to have a paroxysmal arrhythmia during a non-attack period in which the patient is identified as not experiencing an abnormal condition due to the paroxysmal arrhythmia, and second teacher electrocardiogram data of a person determined not to have the paroxysmal arrhythmia and identified as not experiencing a predetermined symptom different from the paroxysmal arrhythmia; inputting electrocardiogram data of a subject to be analyzed into the machine learning model; outputting abnormality information output from the machine learning model regarding whether the subject has the paroxysmal arrhythmia; The electrocardiogram analysis method includes:

9. Computer, a machine learning unit having a machine learning model that has been machine-learned using first teacher electrocardiogram data of a patient determined to have a paroxysmal arrhythmia during a non-attack period in which the patient is identified as not experiencing an abnormal condition due to the paroxysmal arrhythmia, and second teacher electrocardiogram data of a human being determined not to have the paroxysmal arrhythmia and who is identified as not experiencing a predetermined symptom different from the paroxysmal arrhythmia; an input processing unit that inputs electrocardiogram data of a subject to be analyzed into the machine learning model; and an output control unit that outputs abnormality information regarding whether the subject has the spasmodic arrhythmia output from the machine learning model; A program to function as a

Images