Information processing system, program, and information processing device

Abstract

An information processing system (1) according to an embodiment comprises a determination unit (102) that inputs pulse information relating to the heartbeat of a subject and attribute information pertaining to the subject to a trained model and thereby generates a determination result indicating the presence / absence and the degree of heart disease in the subject. The trained model is trained using, for a plurality of providers who provide training data, pulse information relating to the heartbeat of the provider, attribute information pertaining to the provider, and diagnosis information indicating the state of cardiac function of the provider.

Description

Information processing system, program, and information processing device

[0001] An embodiment of the present invention relates to an information processing system, a program, and an information processing device.

[0002] Heart disease is currently the second leading cause of death in Japan, and known causes include ischemic heart disease, valvular disease, cardiomyopathy, arrhythmia, and heart failure. Because heart disease can be life-threatening once it develops, early detection and treatment are essential. In current medical care, heart disease is classified into several stages depending on the condition, and treatment plans are considered based on the stage.

[0003] For example, heart failure, a type of heart disease, is classified into four stages, A to D, depending on the condition. Depending on the severity of the stage, lifestyle changes, drug treatment, palliative care, etc. are then implemented.

[0004] Patent No. 6893002

[0005] However, while early detection and treatment of heart disease are essential, patients must visit a medical institution to find out whether they have heart disease or not and its severity. Therefore, if patients visit a medical institution after noticing symptoms, their condition may have already progressed.

[0006] An object of the present invention is to provide an information processing system, a program, and an information processing device that can easily determine the risk of heart disease.

[0007] In order to solve the above-mentioned problems and achieve the object, the present invention provides an information processing system that includes a judgment unit that generates a judgment result indicating the presence or absence and severity of cardiac disease in a subject by inputting beat information regarding the subject's heartbeat and attribute information of the subject into a trained model, and the trained model is a trained model that has been trained using beat information regarding the heartbeat of the provider, attribute information of the provider, and diagnostic information indicating the state of the provider's cardiac function for multiple providers who provide training data.

[0008] The present invention is also a program that causes a computer to execute a judgment process that generates a judgment result indicating the presence or absence and severity of cardiac disease in a subject by inputting beat information regarding the subject's heartbeat and attribute information of the subject into a trained model, wherein the trained model is a trained model that has been trained for multiple providers who provide training data using beat information regarding the provider's heartbeat, attribute information of the provider, and diagnostic information indicating the state of the provider's cardiac function.

[0009] The present invention also provides an information processing device that includes a judgment unit that generates a judgment result indicating the presence or absence and severity of cardiac disease in a subject by inputting beat information regarding the subject's heartbeat and attribute information of the subject into a trained model, wherein the trained model is a trained model that is trained for a plurality of providers who provide training data using beat information regarding the provider's heartbeat, attribute information of the provider, and diagnostic information indicating the state of the provider's cardiac function.

[0010] The present invention also provides an information processing device that includes a learning unit that performs machine learning using pulse information regarding the heartbeat of a subject, attribute information of the subject, and diagnostic information indicating the state of cardiac function of the provider for multiple providers who provide training data, in order to generate a trained model that receives input of pulse information regarding the heartbeat of the subject and attribute information of the subject, and outputs a judgment result indicating the presence or absence and severity of cardiac disease in the subject.

[0011] According to the present invention, it is possible to provide an information processing system, a program, and an information processing device that can easily determine the risk of heart disease.

[0012] FIG. 1 is a diagram illustrating an example of a schematic configuration of an information processing system. FIG. 2 is a diagram illustrating an example of a hardware configuration of an information processing device. FIG. 3 is a flowchart illustrating a processing procedure of the information processing system 1. FIG. 4 is a diagram illustrating a feature calculation process. FIG. 5 is a diagram illustrating a method for calculating sample entropy. FIG. 6 is a diagram illustrating a trained model of the information processing system 1. FIG. 7 is a diagram illustrating an example of a schematic configuration of an information processing system 2. FIG. 8A is a diagram illustrating a verification result of a machine learning model according to Example 1. FIG. 8B is a diagram illustrating a verification result of a machine learning model according to Example 1. FIG. 9A is a diagram illustrating a verification result of a machine learning model according to Example 1. FIG. 9B is a diagram illustrating a verification result of a machine learning model according to Example 1. FIG. 10 is a diagram illustrating generation of input data according to Example 2. FIG. 11A is a diagram illustrating a verification result of a machine learning model according to Example 2. FIG. 11B is a diagram illustrating a verification result of a machine learning model according to Example 2. FIG. 12A is a diagram illustrating a verification result of a machine learning model according to Example 2. FIG. 12B is a diagram illustrating a verification result of a machine learning model according to Example 2. FIG. 13A is a diagram illustrating a verification result of a machine learning model according to Example 2. FIG. 13B is a diagram illustrating a verification result of the machine learning model according to the second example.

[0013] An information processing system, a program, and an information processing device according to the present invention will be described below with reference to the drawings. Note that the following embodiments are not limited to the following descriptions. Furthermore, each embodiment can be combined with other embodiments or conventional techniques as long as there is no contradiction in the processing content.

[0014] (Embodiment) First, a schematic configuration of an information processing system 1 according to the present invention will be described with reference to Fig. 1. Fig. 1 is a diagram showing an example of the schematic configuration of the information processing system 1. As shown in Fig. 1, the information processing system 1 according to the embodiment includes an information processing device 10, a wearable sensor 20, and an operation terminal 30.

[0015] The information processing device 10 is a server device that provides a process for determining the risk of heart disease as a service. For example, the information processing device 10 includes an acquisition unit 101, a determination unit 102, and an output control unit 103. Note that the functions of the information processing device 10 are not limited to the acquisition unit 101, the determination unit 102, and the output control unit 103. The acquisition unit 101, the determination unit 102, and the output control unit 103 will be described later.

[0016] The wearable sensor 20 is a sensor device formed in the shape of a flexible plate and attached to the trunk of a subject. For example, the wearable sensor 20 is attached by adhesion or suction to the skin of the trunk. The subject is a person who is to be assessed for the presence or absence and severity of heart disease, and includes, for example, heart disease patients, persons suspected of having heart disease, and healthy individuals.

[0017] For example, the wearable sensor 20 is a multi-sensor device equipped with an electrocardiograph and a motion monitor. The electrocardiograph measures time-series electrocardiographic information (electrocardiographic waveform data) based on electrical signals flowing through the subject's body. The motion monitor (three-axis acceleration sensor) measures time-series acceleration information (acceleration data) based on the subject's body movements. The wearable sensor 20 records the various measured data in its internal memory in association with the measurement time. In this way, the wearable sensor 20 detects the subject's electrocardiographic information and acceleration information. Note that the wearable sensor 20 is an example of a detection unit.

[0018] The operation terminal 30 is an information processing device operated by an operator, such as a personal computer, a workstation, a smartphone, or a tablet. The operation terminal 30 is connected to the information processing device 10 and the wearable sensor 20 via any communication means. Any network, such as a local area network (LAN) or a wide area network (WAN), can be used as the communication means. The operator is a person who operates the operation terminal 30, such as a doctor or a person working in a medical institution. However, the operator is not limited to this. For example, the operator may be the subject or a person assisting the subject.

[0019] A representative example of processing in the information processing system 1 shown in FIG. 1 will now be described. The wearable sensor 20 is constantly worn by the subject for a fixed period of time, such as several days to several weeks, and sequentially collects electrocardiographic information during that period. When determining the presence or absence and severity of cardiac disease, the subject removes the wearable sensor 20 and hands it over to an operator. The operator operates the operation terminal 30 to access the memory inside the wearable sensor 20 and reads the electrocardiographic information collected over the fixed period. The operator then transmits the read electrocardiographic information to the information processing device 10. The information processing device 10 determines the presence or absence and severity of cardiac disease in the subject based on the electrocardiographic information and transmits the determination result to the operation terminal 30. The operator views the determination result transmitted from the information processing device 10 and takes various actions based on the determination result.

[0020] 1 is merely an example, and the present invention is not limited thereto. For example, wearable sensor 20 may include any type of sensor other than an electrocardiograph or a body motion monitor, such as a sphygmograph that measures pulse waves or a thermistor that measures temperature (body temperature). Furthermore, wearable sensor 20 does not necessarily have to include a body motion monitor. In other words, wearable sensor 20 only needs to include an electrocardiograph that detects electrocardiographic information.

[0021] Furthermore, for example, the operation terminal 30 does not need to be constantly connected to the information processing device 10 and the wearable sensor 20. It is sufficient for the devices to be connected when the various types of information described above are exchanged between them. Furthermore, when information is exchanged via a recording medium or the like, the devices do not need to be connected. Furthermore, the wearable sensor 20 may transmit various types of information directly to the information processing device 10 without going through the operation terminal 30.

[0022] Furthermore, it is preferable that the wearable sensor 20 is always attached to the subject, but it may be removed to the extent that it does not affect the processing of the information processing device 10. Furthermore, various measurement data recorded in the wearable sensor 20 may be automatically transmitted to the information processing device 10 periodically (e.g., once every few hours) via a network.

[0023] Next, the hardware configuration of the information processing device 10 provided in the information processing system 1 will be described with reference to Fig. 2. Fig. 2 is a diagram showing an example of the hardware configuration of the information processing device 10. Note that Fig. 2 describes the hardware configuration of the information processing device 10, but the hardware configuration of the operation terminal 30 is basically the same, so description thereof will be omitted.

[0024] As shown in FIG. 2 , the information processing device 10 includes a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, an auxiliary storage device 14, an input device 15, a display device 16, and an external I / F (Interface) 17.

[0025] The CPU 11 is a processor (processing circuit) that executes a program to comprehensively control the operation of the information processing device 10 and realize various functions of the information processing device 10. For example, the functions of the acquisition unit 101, the determination unit 102, and the output control unit 103 of the information processing device 10 are realized by the CPU 11.

[0026] The ROM 12 is a non-volatile memory that stores various data (information written during the manufacturing stage of the information processing device 10) including a program for starting up the information processing device 10. The RAM 13 is a volatile memory that has a work area for the CPU 11. The auxiliary storage device 14 stores various data such as the program executed by the CPU 11. The auxiliary storage device 14 is configured, for example, by a hard disk drive (HDD), a solid state drive (SSD), or the like.

[0027] The input device 15 is a device for performing various operations by the operator of the information processing device 10. The input device 15 is configured by, for example, a mouse, a keyboard, a touch panel, or hardware keys.

[0028] The display device 16 displays various types of information. For example, the display device 16 displays image data, model data, a GUI (Graphical User Interface) for receiving various operations from an operator, medical images, etc. The display device 16 is configured as, for example, a liquid crystal display, an organic EL (Electro Luminescence) display, or a cathode ray tube display. Note that the input device 15 and the display device 16 may be integrated into one unit, for example, in the form of a touch panel.

[0029] The external I / F 17 is an interface for connecting (communicating) with any external device such as a server device 20 .

[0030] 2 is merely an example, and the present invention is not limited to this. For example, the hardware configuration of the information processing device 10 may be any configuration of a known computer or workstation.

[0031] The processing of the information processing device 10 provided in the information processing system 1 will be described using Fig. 3. Fig. 3 is a flowchart showing the processing procedure of the information processing system 1. The processing procedure shown in Fig. 3 is started, for example, in response to a request from an operator. Note that Fig. 3 will be described with appropriate reference to Figs. 1, 2, and 4 to 6.

[0032] As shown in FIG. 3, the acquisition unit 101 of the information processing device 10 receives electrocardiogram information and attribute information of a subject (step S101).

[0033] Here, the electrocardiographic information is continuous data in which potentials measured by an electrocardiograph provided in the wearable sensor 20 are continuously recorded as numerical data. If the subject wears the wearable sensor 20 for seven days, the electrocardiographic information of the subject corresponds to approximately 168 hours of continuous data. Note that the electrocardiographic information of the subject does not necessarily have to be 168 hours. For example, the electrocardiographic information of the subject preferably includes 48 hours or more of continuous data, and more preferably includes 144 hours or more of continuous data.

[0034] The attribute information includes at least one of the subject's age, sex, height, weight, and BMI (Body Mass Index). BMI can be calculated using a known calculation method. It is preferable to use the subject's age and BMI as the attribute information.

[0035] For example, the operator operates the operation terminal 30 to read time-series electrocardiogram information from the wearable sensor 20 that has been worn by a subject for seven days (approximately 168 hours). The operator also operates the operation terminal 30 to input attribute information of the subject. As a result, the operation terminal 30 acquires the electrocardiogram information and attribute information of the subject. The operation terminal 30 then transmits the acquired electrocardiogram information and attribute information to the information processing device 10. The acquisition unit 101 of the information processing device 10 receives the time-series electrocardiogram information and attribute information transmitted from the operation terminal 30.

[0036] Next, the acquisition unit 101 of the information processing device 10 acquires heartbeat information of the subject based on the electrocardiogram information of the subject (step S102). For example, the acquisition unit 101 calculates heartbeat information about the subject based on the time-series electrocardiogram information detected from the subject.

[0037] Specifically, the acquisition unit 101 divides the electrocardiogram information at predetermined intervals in the time series direction to generate a plurality of divided data items arranged in time series. The acquisition unit 101 then calculates a plurality of types of statistical values ​​for each of the plurality of divided data items. The acquisition unit 101 then calculates pulsation information for each of the plurality of statistical values. The pulsation information is an example of a feature quantity that represents the characteristics of the electrocardiogram information in the time series direction.

[0038] Here, the calculation process of the feature quantities will be described with reference to FIG. 4 . FIG. 4 is a diagram for explaining the calculation process of the feature quantities. As shown in FIG. 4 , for example, the electrocardiogram information is continuous data including n pieces of numerical data from "numerical data 1" to "numerical data n" (n is a natural number). The acquisition unit 101 generates a plurality of divided data by dividing the n pieces of numerical data included in this continuous data at 5-minute intervals, starting from the first piece. For example, if 100 pieces of numerical data are included in a 5-minute period, the acquisition unit 101 defines "numerical data 1" to "numerical data 100" as "divided data 1" and "numerical data 101" to "numerical data 200" as "divided data 2." In this way, the acquisition unit 101 generates m pieces of divided data from "divided data 1" to "divided data m" (m is a natural number smaller than n).

[0039] Next, the acquiring unit 101 calculates multiple types of statistical values ​​for each of the m pieces of divided data. The multiple types of statistical values ​​can be arbitrarily selected from any type of statistical value, such as mean value, variance, standard deviation, maximum value, minimum value, and skewness. For example, the acquiring unit 101 calculates the average value of 100 pieces of numerical data included in divided data 1 as statistical value data 1, and calculates the average value of 100 pieces of numerical data included in divided data 2 as statistical value data 2. In this way, the acquiring unit 101 calculates the average value of each of the m pieces of divided data to calculate m pieces of statistical value data (average value data) from "statistical value data 1" to "statistical value data m." Therefore, the acquiring unit 101 calculates m pieces of statistical value data for each type of statistical value.

[0040] The acquiring unit 101 then calculates a median and an interquartile range for each of the multiple types of statistical values. For example, the acquiring unit 101 calculates a median or an interquartile range using m pieces of statistical value data corresponding to the average values ​​of the m pieces of divided data. As a result, the acquiring unit 101 calculates a median or an interquartile range as beat information from the electrocardiogram information, which is continuous data in time series. Note that the calculated median and interquartile range are calculated from the divided data divided along the time series, and therefore can be said to represent the characteristics of the electrocardiogram information in the time series direction.

[0041] The statistical values ​​include at least one of various statistical values ​​of heart rate variability obtained from electrocardiogram information, a peak frequency of a predetermined band component in a power spectrum based on frequency analysis of heart rate variability, an integral value of the predetermined band component, and a value based on nonlinear analysis of heart rate variability. The beat information also includes at least one of a median and an interquartile range of the statistical values.

[0042] Specifically, the beat information includes at least one of information regarding the entropy of heart rate fluctuations, the average heartbeat interval in heart rate fluctuations, the heart rate based on electrocardiographic information, information regarding the integral value or peak frequency of the HF (High Frequency) component based on frequency analysis of heart rate fluctuations, information regarding the integral value or peak frequency of the LF (Low Frequency) component based on frequency analysis of heart rate fluctuations, and information regarding the standard deviation of the Lorenz plot of heart rate fluctuations.

[0043] Here, although the number of types of feature quantities (beat information) that can be calculated in the present invention reaches several hundred, useful feature quantities are shown in Table 1. In Table 1, "feature quantity" indicates the type of beat information that is useful for determining the presence or absence and severity of heart disease. "Notes" also indicates the meaning of each feature quantity.

[0044]

[0045] For example, the interquartile range of sample entropy of heart rate variability is an index corresponding to the disorder (difficulty in predicting) of the time series change of the R-wave interval (RR interval: RRI). Sample entropy is a value calculated based on the number of sampling groups that have a similar value change pattern to a group (sampling group) containing any consecutive multiple samples, and indicates the disorder of consecutive values. Note that sample entropy is an example of a statistical value.

[0046] For example, sample entropy is calculated using the following formula 1. In formula 1, SampEn represents sample entropy. m represents the number of samples included in a sampling group. r represents the width (range) of values ​​used to determine whether the values ​​of each sample are close to each other. n represents the number of samples. Ai represents the number of sampling groups in which three consecutive samples have similar change patterns in their values. Bi represents the number of sampling groups in which two consecutive samples have similar change patterns in their values.

[0047]

[0048] Here, a method for calculating sample entropy will be described with reference to FIG. 5. FIG. 5 is a diagram for explaining the method for calculating sample entropy. FIG. 5 illustrates an example of a process for deriving Ai and Bi in Equation 1 using any continuous value included in the RRI time series. In FIG. 5, the vertical direction corresponds to the magnitude of the RRI value, and the horizontal direction corresponds to the elapsed time. Note that the sample entropy is calculated for each divided data in FIG. 4.

[0049] In the upper part of Fig. 5, the acquisition unit 101 derives Bi in Equation 1 by counting the number of sampling groups whose value change pattern is similar to that of sampling group SGB0, which includes two consecutive samples i and i+1. In the upper part of Fig. 5, range R1 is a range of values ​​that can be determined to be close to the value of sample i. Range R2 is a range of values ​​that can be determined to be close to the value of sample i+1. The width (magnitude) of the values ​​in ranges R1 and R2 is both r.

[0050] For example, the acquisition unit 101 identifies sample j as a sample included in range R1. Then, the acquisition unit 101 determines whether the value of sample j+1, which is the next sample after the identified sample j, is included in range R2. In this case, since the value of sample j+1 is included in range R2, the acquisition unit 101 determines that sampling group SGB1, which includes two consecutive samples j and j+1, has a value change pattern similar to that of sampling group SGB0.

[0051] The acquisition unit 101 also identifies sample k as a sample included in range R1. The acquisition unit 101 then determines whether the value of sample k+1, which is the next sample after the identified sample k, is included in range R2. In this case, because the value of sample k+1 is included in range R2, the acquisition unit 101 determines that sampling group SGB1, which includes two consecutive samples k and k+1, has a value change pattern similar to that of sampling group SGB0.

[0052] The acquisition unit 101 then executes similar processing to identify four sampling groups SGB1 to SGB4 as sampling groups whose value change patterns are similar to that of sampling group SGB0. Since the number of identified sampling groups is "4," the acquisition unit 101 derives "4" as Bi.

[0053] In the lower part of Figure 5, one sample to be analyzed is added to the derivation process of Bi, and Ai in Equation 1 is derived by comparing it with the change pattern of sampling group SGA0, which also includes sample i+2. In the lower part of Figure 5, range R1 is a range of values ​​that can be determined to be close to the value of sample i. Range R2 is a range of values ​​that can be determined to be close to the value of sample i+1. Range R3 is a range of values ​​that can be determined to be close to the value of sample i+2. The width (magnitude) of the values ​​in ranges R1, R2, and R3 is all r.

[0054] For example, the acquisition unit 101 determines whether the value of sample j+2 is included in range R3. In this case, since the value of sample j+2 is not included in range R3, the acquisition unit 101 determines that the value change pattern of the sampling group including three consecutive samples j, j+1, and j+2 is not similar to that of sampling group SGA0.

[0055] The acquisition unit 101 also determines whether the value of sample k+2 is within range R3. In this case, since the value of sample k+2 is within range R3, the acquisition unit 101 determines that the value change pattern of sampling group SGA1, which includes three consecutive samples k, k+1, and k+2, is similar to that of sampling group SGA0.

[0056] The acquisition unit 101 then executes similar processing to identify two sampling groups SGA1 to SGA2 as sampling groups with a similar value change pattern to that of the sampling group SGA0. Since the number of identified sampling groups is "2", the acquisition unit 101 derives "2" as Ai.

[0057] In this way, the acquisition unit 101 derives Ai and Bi. Then, the acquisition unit 101 calculates the sample entropy of heart rate variability as a statistical value by inputting the derived Ai and Bi into Equation 1. The acquisition unit 101 calculates the interquartile range of the calculated sample entropy of heart rate variability as beat information.

[0058] Note that the content described in Fig. 5 is merely an example, and the present invention is not limited to this. For example, the number of samples included in a sampling group is not limited to the values ​​shown in Fig. 5, and any value can be set.

[0059] Returning to the explanation of Table 1, for example, the interquartile range of the standard deviation of the heart rate obtained from electrocardiogram information is an index corresponding to the change in the variability of the heart rate.

[0060] The median SD2 / SD1 of heart rate variability is an index of sympathetic nervous activity. SD1 indicates the standard deviation in the y = -x axis direction of the Lorenz plot of RRI. SD2 indicates the standard deviation in the y = x axis direction of the Lorenz plot of RRI.

[0061] Furthermore, the interquartile range of SD2 / SD1 of heart rate variability is an index corresponding to the variability of the sympathetic nervous index.

[0062] The median value of the cardiac vagal index (CVI) of heart rate variability is an index of the parasympathetic nervous system. The CVI corresponds to the logarithm (common logarithm) of the product of SD1 and SD2.

[0063] The interquartile range of the LF (Low Frequency) corrected value of heart rate variability is an index corresponding to the variability of the sympathetic nervous index. The LF corrected value is equivalent to the value obtained by dividing LF by the sum of LF and HF (High Frequency). LF and HF are calculated based on frequency analysis of heart rate variability. For example, LF is obtained by integrating the power of the LF component (0.04 Hz to 0.15 Hz) in the power spectrum calculated from the RRI time series. HF is obtained by integrating the power of the HF component (0.15 Hz to 0.40 Hz) in the power spectrum calculated from the RRI time series.

[0064] The median value of the peak frequency of the LF component of heart rate fluctuation is an index corresponding to the period of blood pressure fluctuation.

[0065] The interquartile range of the mean value of heart rate variability is an index corresponding to the dispersion of time-series changes in RRI.

[0066] Furthermore, the interquartile range of SD2 of heart rate variability is an index corresponding to the variability in the total power of autonomic nervous activity.

[0067] Furthermore, the interquartile range of the peak frequency of the HF component of heart rate fluctuation is an index corresponding to the variation in the respiratory cycle.

[0068] Furthermore, the median of the integral value of the LF component of heart rate variability is an index that includes both the parasympathetic and sympathetic components.

[0069] The interquartile range of the approximate entropy of heart rate variability is an index corresponding to the randomness of the RRI time series change. Note that the calculation method for the approximate entropy is different from that for the sample entropy, and it can be calculated by any known calculation method.

[0070] The median value of the logarithm of the HF component of the heart rate variability is an index obtained by logarithmically converting the index of the parasympathetic nervous system. The logarithm of the HF component is obtained, for example, by common logarithm.

[0071] The median of the LF correction values ​​of heart rate variability is an index of the sympathetic nervous system. Note that the method for calculating the LF correction value is the same as the calculation method described above, and therefore a description thereof will be omitted.

[0072] Note that the contents described in Table 1 are merely examples, and the present invention is not limited thereto. For example, the acquisition unit 101 may acquire at least one of the feature amounts shown in Table 1. However, it is preferable to use, as the feature amount, the interquartile range of the sample entropy of heart rate variability, the interquartile range of the standard deviation of the heart rate, the median of SD2 / SD1 of heart rate variability, the interquartile range of SD2 / SD1 of heart rate variability, the median of the cardiac vagal index (CVI) of heart rate variability, the interquartile range of the low frequency (LF) corrected value of heart rate variability, and the median of the peak frequency of the LF component of heart rate variability.

[0073] In addition, any feature may be defined by any calculation method other than those shown in Table 1. The frequency analysis includes at least one of fast Fourier transform, Lomb-Scargle periodogram, and autoregressive model. However, the beat information is different from the electrocardiogram waveform included in the electrocardiogram information.

[0074] Returning to the explanation of FIG. 3 , the determination unit 102 of the information processing device 10 generates a determination result by inputting the pulsation information and attribute information into the trained model (step S103). Note that this trained model is trained using pulsation information related to the heartbeats of multiple providers who provide training data, attribute information about the providers, and diagnostic information indicating the state of the provider's cardiac function. Note that the providers include healthy individuals and individuals diagnosed with at least heart failure by a doctor. In other words, the determination result is information indicating the presence or absence and severity of heart failure in the subject. The diagnostic information is information indicating the state of the provider's cardiac function related to heart failure.

[0075] Here, the trained model of the information processing system 1 will be described with reference to FIG. 6. FIG. 6 is a diagram for explaining the trained model of the information processing system 1. The upper part of FIG. 6 shows the processing during training of the trained model, and the lower part of FIG. 6 shows the processing during operation of the trained model. Here, the trained model is constructed in advance and stored in a predetermined storage area (e.g., ROM 12) available to the determination unit 102. The processing of the "learning unit" for constructing the trained model will be described later.

[0076] As shown in the upper part of FIG. 6 , during learning, machine learning is performed using, for example, the pulse information, attribute information, and diagnostic information of multiple providers 1 to N. Here, the provider's pulse information is calculated from the provider's electrocardiogram information. The method for calculating the provider's pulse information is the same as the method for calculating the subject's pulse information described in the acquisition unit 101, and therefore, description thereof will be omitted. The provider's attribute information is information including at least one of the provider's age, sex, height, weight, and BMI (Body Mass Index). The BMI can be calculated using a known calculation method. The provider's diagnostic information is information indicating the state of the provider's cardiac function and is the result of a diagnosis by a doctor.

[0077] For example, a trained model is constructed by performing machine learning using XGBoost. This trained model is a model that receives pulse information and attribute information of a subject and outputs a determination result indicating the presence or absence and severity of heart disease in the subject.

[0078] 6, during operation, the determination unit 102 inputs the pulse information and attribute information of the subject to the trained model constructed by machine learning, thereby causing the trained model to output a determination result. In this way, the determination unit 102 generates a determination result.

[0079] 6 can be executed at any time before the operation process is executed. The learning process can also be executed to update (additionally learn) an already generated trained model.

[0080] In addition, in the present invention, the information input to the trained model is only beat information or attribute information, and is not the value of the electrocardiogram information itself. In other words, the beat information is different from the electrocardiogram waveform included in the electrocardiogram information.

[0081] Then, the output control unit 103 outputs the determination result (step S104). For example, the output control unit 103 transmits the determination result to the operation terminal 30. The operation terminal 30 displays the determination result transmitted from the output control unit 103 on a display device or stores it in a predetermined storage device.

[0082] As described above, in the information processing system 1 according to the present invention, the determination unit 102 of the information processing device 10 inputs pulse information related to the subject's heartbeat and attribute information of the subject into a trained model to generate a determination result indicating the presence or absence and severity of heart disease in the subject. This enables the information processing system 1 to easily determine the risk of heart disease.

[0083] For example, by using the information processing system 1, a user can obtain the results of a heart disease risk assessment using a trained model without visiting a medical institution. As a result, the information processing system 1 is expected to contribute to the early detection and treatment of heart disease.

[0084] In the above embodiment, the continuous data acquired from the subject for a certain period is used as the processing target for calculating the feature amount, but the present invention is not limited to this. For example, the acquisition unit 101 may select numerical data corresponding to sleep from the acquired continuous data for a certain period and use the selected numerical data for calculating the feature amount.

[0085] That is, the acquisition unit 101 identifies the sleeping time based on at least one of a preset time, time-series electrocardiogram information detected from the subject, time-series acceleration information detected from the subject, and time-series posture information detected from the subject. Then, the acquisition unit 101 calculates pulse information based on the electrocardiogram information during sleeping among the time-series electrocardiogram information.

[0086] For example, a period from midnight to 6 a.m. may be preset as the time for identifying sleeping time. In this case, the acquisition unit 101 identifies the period from midnight to 6 a.m. every day as sleeping time. Then, the acquisition unit 101 extracts electrocardiographic information included in the period from midnight to 6 a.m. every day from the time-series electrocardiographic information, and calculates beat information based on the extracted electrocardiographic information.

[0087] Furthermore, for example, the acquisition unit 101 identifies a time when the subject is asleep based on time-series acceleration information detected from the subject. Here, the acceleration information is collected by a three-axis acceleration sensor provided in the wearable sensor 20. As an example, the acquisition unit 101 estimates the subject's movement (body movement) based on the acceleration information. Then, if the magnitude of the estimated movement is less than a threshold for three hours or more, the acquisition unit 101 identifies the time as a time when the subject is asleep. Then, the acquisition unit 101 extracts electrocardiographic information included in the identified time from the time-series electrocardiographic information and calculates pulse information based on the extracted electrocardiographic information.

[0088] Furthermore, for example, the acquisition unit 101 identifies the sleeping time based on time-series posture information detected from the subject. Here, the posture information is collected by a gyro sensor provided in the wearable sensor 20. As an example, the acquisition unit 101 estimates the posture of the subject based on the posture information. Then, the acquisition unit 101 identifies the time when the subject is lying down as the sleeping time. Then, the acquisition unit 101 extracts electrocardiographic information included in the identified time from the time-series electrocardiographic information, and calculates pulse information based on the extracted electrocardiographic information.

[0089] In this way, the information processing device 10 can select electrocardiogram information corresponding to sleep from continuous data for a certain period of time and use it for feature calculation processing. As a result, the information processing device 10 can make a judgment after meeting conditions related to the body movements of the subject and the provider, thereby improving the accuracy of the judgment.

[0090] Other Embodiments The present invention may be implemented in various different forms other than the above-described embodiment.

[0091] (System Configuration) In the above embodiment, the schematic configuration of the information processing system 1 is described using Fig. 1, but the present invention is not limited to this. For example, the processing functions of the acquisition unit 101, the determination unit 102, and the output control unit 103 provided in the information processing device 10 may be provided in any device provided in the information processing system 1. As an example, the acquisition unit 101 may be provided in the operation terminal 30.

[0092] That is, the information processing system 1 includes a determination unit that generates a determination result of the risk of heart disease by inputting pulse information related to the subject's heartbeat and attribute information of the subject to a trained model. The trained model is trained using pulse information related to the heartbeat of the provider, attribute information of the provider, and diagnostic information indicating the state of the provider's cardiac function for multiple providers who provide training data.

[0093] (Program) Furthermore, each process of the information processing system 1 described in the above embodiment and modified examples may be provided as a program that causes a computer to execute the process.

[0094] That is, the program causes the computer to execute a determination process for generating a determination result indicating the presence or absence and severity of cardiac disease in the subject by inputting pulse information related to the subject's heartbeat and attribute information of the subject into the trained model. The trained model is a trained model trained using pulse information related to the heartbeats of the providers, attribute information of the providers, and diagnostic information indicating the state of cardiac function of the providers for multiple providers who provide training data.

[0095] (Information Processing Device) Of the information processing systems 1 described in the above embodiments and modifications, the information processing device 10 can easily determine the risk of heart disease.

[0096] That is, the information processing device 10 includes a determination unit that generates a determination result indicating the presence or absence and severity of cardiac disease in the subject by inputting pulse information related to the subject's heartbeat and attribute information of the subject to a trained model. The trained model is trained using pulse information related to the heartbeats of the providers, attribute information of the providers, and diagnostic information indicating the state of the provider's cardiac function for multiple providers who provide training data.

[0097] The processing function of the wearable sensor 20 as a detector may be provided in the information processing device 10 .

[0098] (Learning Unit) The system according to the present invention may also include a processing function for constructing the trained model described in the above embodiment.

[0099] The schematic configuration of an information processing system 2 according to the present invention will be described using Fig. 7. Fig. 7 is a diagram showing an example of the schematic configuration of the information processing system 2. As shown in Fig. 7, the information processing system 2 according to the embodiment includes a wearable sensor 20, an operation terminal 30, and an information processing device 40. The configurations of the wearable sensor 20 and the operation terminal 30 shown in Fig. 7 are basically the same as the configurations of the wearable sensor 20 and the operation terminal 30 shown in Fig. 1, and therefore description thereof will be omitted.

[0100] The information processing device 40 is a server device that constructs a trained model for determining the risk of heart disease. For example, the information processing device 40 includes an acquisition unit 401, a learning unit 402, and an output control unit 403. The configurations of the acquisition unit 401 and the output control unit 403 shown in Fig. 7 are basically the same as the configurations of the acquisition unit 101 and the output control unit 103 shown in Fig. 1, and therefore description thereof will be omitted.

[0101] Here, the learning unit 402 performs machine learning using the heartbeat information, attribute information, and diagnosis information of multiple providers 1 to N to construct a trained model for determining the presence or absence and severity of cardiac disease in a subject. For example, the learning unit 402 performs machine learning using XGBoost, as shown in the upper part of FIG. 6 . Through this machine learning, the learning unit 402 receives the heartbeat information and attribute information of the subject and generates a trained model that outputs a determination result indicating the presence or absence and severity of cardiac disease in the subject. The learning unit 402 stores the generated trained model in any storage area available to the determination unit 102.

[0102] In other words, the learning method of the present invention performs machine learning using pulse information about the heartbeat of a subject, attribute information about the subject, and diagnostic information indicating the state of the subject's cardiac function for multiple providers who provide training data, in order to generate a trained model that receives input of pulse information about the subject's heartbeat and attribute information about the subject and outputs a judgment result indicating the presence or absence and severity of cardiac disease in the subject.

[0103] Although the case where machine learning is performed using XGBoost has been described here, the present invention is not limited to this. The learning unit 402 is not limited to XGBoost and can apply any known machine learning algorithm.

[0104] Although the case where the information processing device 40, which is different from the information processing device 10, includes the learning unit 402 has been described here, the present invention is not limited to this. For example, the learning unit 402 may be included in the information processing device 10.

[0105] According to the embodiment and modified examples described above, the risk of heart disease can be easily determined.

[0106] The programs executed by the information processing systems 1 and 2 according to the above-described embodiments and modifications may be provided by being recorded in an installable or executable file format on a computer-readable recording medium such as a CD-ROM, a flexible disk (FD), a CD-R, a DVD, or a USB (Universal Serial Bus), or may be provided or distributed via a network such as the Internet. Also, various programs may be provided by being pre-installed in a non-volatile storage medium such as a ROM.

[0107] EXAMPLES The present invention will be described in more detail below based on examples, but the present invention is not limited to these examples.

[0108] Example 1 Using the method described in the above embodiment, two machine learning models (trained models) for classifying heart failure into four stages A to D were created, and their accuracy was verified.

[0109] Here, the stage classification of heart failure will be described. For example, heart failure is classified into four stages, A to D, according to the ACCF / AHA (American College of Cardiology Foundation / American Heart Association) stage classification. This classification is based mainly on the presence or absence of "organic heart disease," such as cardiac hypertrophy and reduced cardiac output, and the presence or absence of "heart failure symptoms," such as chest pain, shortness of breath, and leg swelling. Stage A is a stage in which risk factors, such as high blood pressure and diabetes, are present but neither organic heart disease nor heart failure symptoms are present. Stage B is a stage in which organic heart disease is present but no heart failure symptoms are present. Stage C is a stage in which both organic heart disease and heart failure symptoms are present. Stage D is a stage in which both organic heart disease and heart failure symptoms are present and the condition is intractable (treatment-resistant, end-stage). Note that the stages of heart failure that can be classified in the present application are not limited to the above examples, and the definition of the stage classification can be set arbitrarily.

[0110] Table 2 shows the samples used in the two machine learning models (machine learning model A and machine learning model B) and their validation results. Both machine learning models were created using the Nystroem Kernel SVM Classifier.

[0111]

[0112] In Table 2, the patient group indicates the number of data samples obtained from patients diagnosed with heart failure by a doctor, and the healthy group indicates the number of data samples obtained from healthy individuals. That is, the samples in stage A described above are included in the healthy group, and the samples in stages B to D are included in the patient group. Note that the samples used in machine learning model A and the samples used in machine learning model B overlap with each other.

[0113] The results in Table 2 confirm that for both machine learning model A and machine learning model B, both the Recall value and the Specificity value were high, and that machine learning models with sufficient performance were created.

[0114] Next, the sample data used in the machine learning was input to each machine learning model, and the determination results were re-output. The verification results are shown in Figures 8A, 8B, 9A, and 9B. Figures 8A, 8B, 9A, and 9B are diagrams showing the verification results of the machine learning models according to Example 1.

[0115] The graph in Figure 8A shows the re-output scores from machine learning model A for each patient group / healthy group. The graph in Figure 8B shows the re-output scores from machine learning model B for each patient group / healthy group. In Figures 8A and 8B, the vertical axis shows the re-output score from each machine learning model. This score is expressed as a number between 0 and 1, with scores closer to "1" indicating a higher likelihood of being a patient and scores closer to "0" indicating a higher likelihood of being a healthy individual. In other words, the prediction threshold is 0.5. The horizontal axis shows the attribute of whether the subject is a patient or a healthy individual, and the numbers in parentheses correspond to the number of samples. In other words, Figures 8A and 8B show boxplots in which the data for each sample listed in Table 2 is re-input into machine learning model A, and the scores (re-output scores) obtained as the judgment results are plotted by attribute (patient group and healthy group).

[0116] As shown in Figures 8A and 8B, although some samples exceeded the prediction threshold, good results were obtained with both machine learning models.

[0117] The graph in Figure 9A shows the re-output scores by machine learning model A for each stage. The graph in Figure 9B shows the re-output scores by machine learning model B for each stage. In Figures 9A and 9B, the vertical axis indicates the re-output scores from each machine learning model. The explanation of the scores is the same as in Figures 8A and 8B. The horizontal axis indicates the stage classification of heart failure diagnosed by a doctor, and the numbers in parentheses correspond to the number of samples. Note that "Unknown" on the horizontal axis in Figure 9B indicates samples that were diagnosed with heart failure but whose heart failure stage was not yet classified. In other words, Figures 9A and 9B show boxplots in which the data of each sample listed in Table 2 was re-input into machine learning model B, and the scores (re-output scores) obtained as the judgment results were plotted by heart failure stage.

[0118] As shown in Figures 9A and 9B, although some samples exceeded the prediction threshold in Stage A and Stage C, good results were obtained with both machine learning models. In particular, in Stage B, no samples exceeded the prediction threshold, allowing for highly accurate classification.

[0119] Heart failure is generally considered a difficult-to-treat disease, and early detection and treatment are essential. However, by the time a patient becomes aware of the disease due to heart failure symptoms, it has usually progressed to stage C, making early detection at stage B difficult. The results in Figures 9A and 9B demonstrate that stage B can be classified with high accuracy, demonstrating the effectiveness of the information processing device 10 described in this embodiment. Therefore, for example, by wearing a wearable sensor 20 for purposes such as a health check, measuring data, and determining the stage of heart failure using the information processing device 10 according to this embodiment, early detection at stage B, when no heart failure symptoms are present, becomes possible.

[0120] Example 2 Next, verification was performed regarding the number of days of measurement (number of measurement days) using the wearable sensor 20. In Example 2, the same sample as that used in the machine learning model A in Example 1 was used, and a machine learning model was created by the method described in the above embodiment.

[0121] 10 is a diagram illustrating generation of input data according to Example 2. As shown in FIG. 10 , when there are seven days' worth of measurement data in total, seven sets of input data are generated for one day, two days, three days, four days, five days, six days, and seven days. Each set of input data corresponds to the number of days counted from the start of measurement, and is continuous data extracted during sleep (i.e., from midnight to 6:00 AM). Therefore, for example, three days' worth of input data includes continuous data from midnight to 6:00 AM on the first day, continuous data from midnight to 6:00 AM on the second day, and continuous data from midnight to 6:00 AM on the third day.

[0122] Furthermore, depending on the sample (subject), seven days' worth of measurement data may not be obtained due to reasons such as the wearable sensor 20 being removed within seven days. Therefore, for example, if a total of three days' worth of measurement data was obtained, three sets of input data for one day, two days, and three days were generated. Furthermore, for example, if a total of five days' worth of measurement data was obtained, five sets of input data for one day, two days, three days, four days, and five days were generated. The input data samples were the same samples as those used in the machine learning model A in Example 1. The generated input data for each day was input into the previously created machine learning model, and a score was output for each divided number of days.

[0123] 11A, 11B, 12A, 12B, 13A, and 13B are diagrams illustrating the verification results of the machine learning model according to the second example.

[0124] The graph in Figure 11A shows the scores obtained from samples from the patient group for each number of days of measurement. The graph in Figure 11B shows the scores obtained from samples from the healthy group for each number of days of measurement. In Figures 11A and 11B, the vertical axis indicates the score output from the machine learning model. The explanation of the scores is the same as in Figures 8A and 8B. The horizontal axis corresponds to the number of days of measurement. In other words, Figures 11A and 11B are line graphs in which the scores obtained as the judgment results are plotted for each number of days of measurement, and plots from the same sample are connected by a line.

[0125] As shown in Figures 11A and 11B, in both the patient group and the healthy group, the deviation in scores increased as the number of measurement days decreased, and there were occasional cases where the scores deviated from the prediction threshold (0.5).

[0126] The graph in Figure 12A shows the deviation in scores for each number of measurement days shown in Figure 11A. The graph in Figure 12B shows the deviation in scores for each number of measurement days shown in Figure 11B. The graph in Figure 13A is a box plot converted from the graph in Figure 12A. The graph in Figure 13B is a box plot converted from the graph in Figure 12B. In Figures 12A to 13B, the vertical axis corresponds to the difference (deviation) between the score for the maximum number of measurement days for each subject (patient and healthy subject) and the score for each measurement day. In other words, this difference is calculated assuming that the score for the maximum number of measurement days for each subject is correct. The horizontal axis corresponds to the number of measurement days.

[0127] 12A to 13B, in both the patient group and the healthy group, the blurring was significant for the 1-day and 2-day data, but the blurring was generally reduced for the 3-day and 4-day data, and became smaller for the 5-day data or more. As a result, it can be said that it is preferable to perform measurements for 3 days or more, and it is even more preferable to perform measurements for 5 days or more.

[0128] The present invention also provides the technical solutions described in the following appendix.

[0129] (Supplementary Note 1) An information processing system comprising: a judgment unit that generates a judgment result indicating the presence or absence and severity of cardiac disease in the subject by inputting beat information regarding the subject's heartbeat and attribute information of the subject into a trained model, wherein the trained model is a trained model that has been trained for a plurality of providers who provide training data using beat information regarding the provider's heartbeat, attribute information of the provider, and diagnostic information indicating the state of the provider's cardiac function.

[0130] (Supplementary Note 2) The information processing system according to Supplementary Note 1, wherein the providers include healthy individuals and patients diagnosed with heart failure by a doctor.

[0131] (Supplementary Note 3) The information processing system according to Supplementary Note 1, further comprising an acquisition unit that acquires the heartbeat information of the subject.

[0132] (Supplementary Note 4) The information processing system according to Supplementary Note 1, wherein the attribute information of the subject includes at least one of the subject's age, sex, height, weight, and BMI (Body Mass Index).

[0133] (Supplementary Note 5) The information processing system according to Supplementary Note 3, wherein the acquisition unit calculates the pulsation information based on time-series electrocardiogram information detected from the subject.

[0134] (Supplementary Note 6) The information processing system described in Supplementary Note 5, wherein the acquisition unit generates a plurality of divided data arranged in a time series by dividing the electrocardiogram information at predetermined intervals in a time series direction, calculates a plurality of types of statistical values ​​for each of the plurality of divided data, and calculates the beat information representing the characteristics of the electrocardiogram information in the time series direction for each of the plurality of types of statistical values.

[0135] (Supplementary Note 7) The information processing system according to Supplementary Note 5, wherein the electrocardiogram information includes continuous data acquired from the subject for 72 hours or more.

[0136] (Supplementary Note 8) The information processing system according to Supplementary Note 5, wherein the electrocardiogram information includes 120 hours or more of continuous data obtained from the subject.

[0137] (Appendix 9) The information processing system described in Appendix 6, wherein the statistical values ​​include at least one of various statistical values ​​of heart rate fluctuations obtained from the electrocardiogram information, a peak frequency of a specified band component in a power spectrum based on frequency analysis of the heart rate fluctuations, an integral value of the specified band component, and a value based on nonlinear analysis of the heart rate fluctuations, and the beat information includes at least one of a median and an interquartile range of the statistical values.

[0138] (Supplementary Note 10) The information processing system according to Supplementary Note 9, wherein the attribute information of the subject is the age and BMI (Body Mass Index) of the subject, and the heartbeat information is an interquartile range of sample entropy of the heart rate variability.

[0139] (Appendix 11) The information processing system described in Appendix 10, wherein the beat information further includes a median of a low frequency (LF) correction value based on a frequency analysis of the heart rate fluctuation, an interquartile range of a standard deviation of the heart rate obtained from the electrocardiogram information, a median of a peak frequency of the LF component based on a frequency analysis of the heart rate fluctuation, a median of a logarithm of a high frequency (HF) component based on a frequency analysis of the heart rate fluctuation, and an interquartile range of the LF correction value based on a frequency analysis of the heart rate fluctuation.

[0140] (Appendix 12) The information processing system described in Appendix 9, wherein the beat information includes at least one of information regarding the entropy of the heart rate fluctuations, the average heartbeat interval in the heart rate fluctuations, the heart rate based on the electrocardiogram information, information regarding the integral value or peak frequency of the HF (High Frequency) component based on frequency analysis of the heart rate fluctuations, information regarding the integral value or peak frequency of the LF (Low Frequency) component based on frequency analysis of the heart rate fluctuations, and information regarding the standard deviation of a Lorenz plot of the heart rate fluctuations.

[0141] (Supplementary Note 13) The beat information may include an interquartile range of sample entropy of the heart rate variability, an interquartile range of standard deviation of the heart rate obtained from the electrocardiogram information, a median of a value obtained by dividing the standard deviation of the Lorenz plot of the heart rate variability in the y=x-axis direction by the standard deviation of the Lorenz plot in the y=-x-axis direction, an interquartile range of a value obtained by dividing the standard deviation of the Lorenz plot of the heart rate variability in the y=x-axis direction by the standard deviation of the Lorenz plot in the y=-x-axis direction, a median of a logarithm of the product of the standard deviation of the Lorenz plot of the heart rate variability in the y=x-axis direction and the standard deviation of the Lorenz plot in the y=-x-axis direction, an LF (Low Frequency Error) based on frequency analysis of the heart rate variability, the interquartile range of a high frequency (HF) corrected value based on the frequency analysis of the heart rate variability, the median of the peak frequency of the LF component based on the frequency analysis of the heart rate variability, the interquartile range of the mean value of the heart rate variability, the interquartile range of the standard deviation in the x-axis direction of the Lorenz plot of the heart rate variability, the interquartile range of a high frequency (HF) corrected value based on the frequency analysis of the heart rate variability, the median of the integral value of the LF component based on the frequency analysis of the heart rate variability, the interquartile range of approximate entropy of the heart rate variability, the median of the logarithm of the HF component based on the frequency analysis of the heart rate variability, and the median of the LF corrected value based on the frequency analysis of the heart rate variability.

[0142] (Supplementary Note 14) The information processing system according to Supplementary Note 13, wherein the frequency analysis includes at least one of a fast Fourier transform, a Lomb-Scargle periodogram, and an autoregressive model.

[0143] (Supplementary Note 15) The information processing system according to Supplementary Note 5, wherein the beat information is different from an electrocardiogram waveform included in the electrocardiogram information.

[0144] (Supplementary Note 16) The information processing system according to Supplementary Note 5, wherein the acquisition unit calculates the pulsation information based on electrocardiogram information during sleep among the time-series electrocardiogram information.

[0145] (Supplementary Note 17) The information processing system according to Supplementary Note 16, wherein the acquisition unit identifies the sleeping time based on at least one of a preset time, time-series electrocardiogram information detected from the subject, time-series acceleration information detected from the subject, and time-series posture information detected from the subject.

[0146] (Supplementary Note 18) The information processing system according to Supplementary Note 16, wherein the electrocardiogram information during sleep corresponds to electrocardiogram information from midnight to 6:00 among the time-series electrocardiogram information.

[0147] (Supplementary Note 19) The information processing system according to Supplementary Note 5, further comprising a detection unit attached to the subject and configured to detect the electrocardiogram information.

[0148] (Supplementary Note 20) The information processing system according to Supplementary Note 19, wherein the detection unit is attached to a trunk of the subject.

[0149] (Supplementary Note 21) The information processing system according to Supplementary Note 19, wherein the detection unit is a flexible plate and is attached to the skin of the trunk of the subject by adhesion or suction.

[0150] (Supplementary Note 22) The information processing system according to Supplementary Note 1, wherein the determination result is information indicating the presence or absence and degree of heart failure of the subject, and the diagnostic information is information indicating the state of cardiac function of the provider regarding heart failure.

[0151] (Supplementary Note 23) A program that causes a computer to execute a judgment process that generates a judgment result indicating the presence or absence and severity of cardiac disease in the subject by inputting beat information regarding the subject's heartbeat and attribute information of the subject into a trained model, wherein the trained model is a trained model that has been trained for multiple providers who provide training data using beat information regarding the provider's heartbeat, attribute information of the provider, and diagnostic information indicating the state of the provider's cardiac function.

[0152] (Supplementary Note 24) An information processing device comprising: a judgment unit that generates a judgment result indicating the presence or absence and severity of cardiac disease in the subject by inputting pulse information regarding the subject's heartbeat and attribute information of the subject into a trained model, wherein the trained model is a trained model that has been trained for a plurality of providers who provide training data using pulse information regarding the provider's heartbeat, attribute information of the provider, and diagnostic information indicating the state of the provider's cardiac function.

[0153] (Supplementary Note 25) An information processing device comprising: a learning unit that performs machine learning for a plurality of providers who provide training data using pulse information about the heartbeats of the providers, attribute information about the providers, and diagnostic information indicating the state of the cardiac function of the providers, in order to generate a trained model that receives pulse information about the heartbeats of the subject and attribute information about the subject, and outputs a determination result indicating the presence or absence and severity of cardiac disease in the subject.

[0154] REFERENCE SIGNS LIST 1, 2 Information processing system 10, 40 Information processing device 11 CPU 12 ROM 13 RAM 14 Auxiliary storage device 15 Input device 16 Display device 17 External I / F 20 Wearable sensor 30 Operation terminal 101 Acquisition unit 102 Determination unit 103 Output control unit 104 Learning unit

Claims

1. An information processing system comprising: a judgment unit that generates a judgment result indicating the presence or absence and the degree of cardiac disease in the subject by inputting beat information regarding the subject's heartbeat and attribute information of the subject into a trained model, the trained model being a trained model trained for a plurality of providers providing training data using beat information regarding the heartbeat of the provider, attribute information of the provider, and diagnostic information indicating the state of the provider's cardiac function.

2. The information processing system according to claim 1, wherein the providers include healthy individuals and patients diagnosed with heart failure by a doctor.

3. The information processing system according to claim 1, further comprising an acquisition unit that acquires the heartbeat information of the subject.

4. The information processing system according to claim 1, wherein the subject's attribute information includes at least one of the subject's age, sex, height, weight, and BMI (Body Mass Index).

5. The information processing system according to claim 3, wherein the acquisition unit calculates the pulsation information based on time-series electrocardiogram information detected from the subject.

6. The information processing system of claim 5, wherein the acquisition unit generates a plurality of divided data arranged in a time series by dividing the electrocardiogram information at predetermined intervals in the time series direction, calculates a plurality of types of statistical values ​​for each of the plurality of divided data, and calculates the beat information representing the characteristics of the electrocardiogram information in the time series direction for each of the plurality of statistical values.

7. The information processing system according to claim 5, wherein the electrocardiogram information includes continuous data obtained from the subject for 72 hours or more.

8. The information processing system according to claim 5, wherein the electrocardiogram information includes 120 hours or more of continuous data obtained from the subject.

9. The information processing system of claim 6, wherein the statistical values ​​include at least one of various statistical values ​​of heart rate variability obtained from the electrocardiogram information, a peak frequency of a specified band component in a power spectrum based on a frequency analysis of the heart rate variability, an integral value of the specified band component, and a value based on a nonlinear analysis of the heart rate variability, and the beat information includes at least one of a median and an interquartile range of the statistical values.

10. The information processing system of claim 9, wherein the subject's attribute information is the subject's age and BMI (Body Mass Index), and the heartbeat information is the interquartile range of sample entropy of the heart rate variability.

11. The information processing system of claim 10, wherein the beat information further includes a median of LF (Low Frequency) correction value based on frequency analysis of the heart rate variability, an interquartile range of the standard deviation of the heart rate obtained from the electrocardiogram information, a median of the peak frequency of the LF component based on frequency analysis of the heart rate variability, a median of the logarithm of the HF (High Frequency) component based on frequency analysis of the heart rate variability, and an interquartile range of the LF correction value based on frequency analysis of the heart rate variability.

12. The information processing system of claim 9, wherein the beat information includes at least one of information regarding the entropy of the heart rate fluctuation, an average heartbeat interval in the heart rate fluctuation, a heart rate based on the electrocardiogram information, information regarding the integral value or peak frequency of the HF (High Frequency) component based on frequency analysis of the heart rate fluctuation, information regarding the integral value or peak frequency of the LF (Low Frequency) component based on frequency analysis of the heart rate fluctuation, and information regarding the standard deviation of a Lorentz plot of the heart rate fluctuation.

13. The beat information includes an interquartile range of sample entropy of the heart rate fluctuation, an interquartile range of standard deviation of the heart rate obtained from the electrocardiogram information, a median of a value obtained by dividing the standard deviation of the Lorenz plot of the heart rate fluctuation in the y = x axis direction by the standard deviation of the Lorenz plot in the y = -x axis direction, an interquartile range of a value obtained by dividing the standard deviation of the Lorenz plot of the heart rate fluctuation in the y = x axis direction by the standard deviation of the Lorenz plot in the y = -x axis direction, a median of a logarithm of the product of the standard deviation of the Lorenz plot of the heart rate fluctuation in the y = x axis direction and the standard deviation of the Lorenz plot in the y = -x axis direction, and an LF (Low Frequency Shift Register) based on a frequency analysis of the heart rate fluctuation.

10. The information processing system of claim 9, further comprising at least one of an interquartile range of a high frequency (HF) correction value, a median of a peak frequency of an LF component based on a frequency analysis of the heart rate variability, an interquartile range of an average value of the heart rate variability, an interquartile range of a standard deviation in the y=x-axis direction of a Lorentz plot of the heart rate variability, an interquartile range of a peak frequency of an HF (High Frequency) component based on a frequency analysis of the heart rate variability, a median of an integral value of an LF component based on a frequency analysis of the heart rate variability, an interquartile range of an approximate entropy of the heart rate variability, a median of a logarithm of an HF component based on a frequency analysis of the heart rate variability, and a median of an LF correction value based on a frequency analysis of the heart rate variability.

14. The information processing system according to claim 13, wherein the frequency analysis includes at least one of a fast Fourier transform, a Lomb-Scargle Periodogram, and an autoregressive model.

15. The information processing system according to claim 5, wherein the beat information is different from an electrocardiogram waveform included in the electrocardiogram information.

16. The information processing system according to claim 5, wherein the acquisition unit calculates the pulsation information based on electrocardiogram information during sleep from among the time-series electrocardiogram information.

17. The information processing system of claim 16, wherein the acquisition unit identifies the time of sleep based on at least one of a preset time, time-series electrocardiogram information detected from the subject, time-series acceleration information detected from the subject, and time-series posture information detected from the subject.

18. The information processing system according to claim 16, wherein the electrocardiogram information during sleep corresponds to electrocardiogram information from midnight to 6:00 among the time-series electrocardiogram information.

19. The information processing system according to claim 5, further comprising a detection unit attached to the subject and configured to detect the electrocardiogram information.

20. The information processing system according to claim 19, wherein the detection unit is attached to the trunk of the subject.

21. The information processing system according to claim 19, wherein the detection unit is a flexible plate and is attached to the skin of the subject's trunk by adhesion or suction.

22. The information processing system of claim 1, wherein the judgment result is information indicating the presence or absence and degree of heart failure in the subject, and the diagnostic information is information indicating the state of the donor's cardiac function with respect to heart failure.

23. A program that causes a computer to execute a judgment process that generates a judgment result indicating the presence or absence and the degree of cardiac disease in the subject by inputting beat information regarding the subject's heartbeat and attribute information of the subject into a trained model, the trained model being a trained model that has been trained for a plurality of providers providing training data using beat information regarding the heartbeat of the provider, attribute information of the provider, and diagnostic information indicating the state of the provider's cardiac function.

24. An information processing device comprising: a judgment unit that generates a judgment result indicating the presence or absence and the degree of cardiac disease in the subject by inputting beat information regarding the subject's heartbeat and attribute information of the subject into a trained model, wherein the trained model is a trained model trained for a plurality of providers providing training data using beat information regarding the provider's heartbeat, attribute information of the provider, and diagnostic information indicating the state of the provider's cardiac function.

25. An information processing device comprising: a learning unit that performs machine learning for a plurality of providers providing training data using pulse information relating to the heartbeat of the provider, attribute information of the provider, and diagnostic information indicating the state of the provider's cardiac function, in order to generate a trained model that receives pulse information relating to the heartbeat of the subject and attribute information of the subject, and outputs a judgment result indicating the presence or absence and the severity of the subject's cardiac disease.

Images