Learning program, information processing device, and learning method

Abstract

To provide a learning program, information processing device, and learning method that enable accurate prediction of the likelihood of future major cardiovascular events.SOLUTION: A learning program includes: acquiring first electrocardiogram information and a first BNP value for a first person; generating first teaching data including the first electrocardiogram information and the first BNP value; generating a first learning model by learning the first teaching data; acquiring second electrocardiogram information and a second BNP value for a second person; calculating a corrected BNP value using the second BNP value and a coefficient corresponding to the second person; generating second teaching data including the second electrocardiogram information and the corrected BNP value; and generating a second learning model including at least a portion of layers constituting the first learning model by learning the second teaching data.SELECTED DRAWING: Figure 16

Description

[Technical Field]

[0001] The present invention relates to a learning program, an information processing device, and a learning method. [Background technology]

[0002] For example, the worsening state of cardiac failure in a subject (patient) may be determined by using electrocardiogram waveform data, heart rate, and the like of the subject (see Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 7075611 Summary of the Invention [Problem to be solved by the invention]

[0004] In the medical field, there is a need to accurately predict the likelihood of a subject's future major adverse cardiovascular events (MACE) occurring by using, for example, the subject's electrocardiogram waveform data, heart rate, etc.

[0005] Therefore, an object of the present disclosure is to provide a learning program, an information processing device, and a learning method that enable accurate prediction of the likelihood of future major cardiovascular events. [Means for solving the problem]

[0006] The learning program in the present disclosure causes a computer to execute the following processes: acquire first electrocardiogram information of a first person and a first BNP value of the first person; generate first training data including the first electrocardiogram information and the first BNP value; generate a first learning model by learning the first training data; acquire second electrocardiogram information of a second person and a second BNP value of the second person; calculate a corrected BNP value by using the second BNP value and a coefficient corresponding to the second person; generate second training data including the second electrocardiogram information and the corrected BNP value; and generate a second learning model including at least a portion of the layers constituting the first learning model by learning the second training data. [Effects of the Invention]

[0007] The learning program, information processing device, and learning method of the present invention make it possible to accurately predict the likelihood of a future major cardiovascular event occurring. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a diagram showing an example of the configuration of an information processing system 10 according to the first embodiment. [Figure 2] FIG. 2 is a diagram showing an example of the configuration of the information processing system 10 according to the first embodiment. [Figure 3] FIG. 3 is a diagram for explaining an outline of the first embodiment. [Figure 4] FIG. 4 is a diagram for explaining an outline of the first embodiment. [Figure 5] FIG. 5 is a diagram for explaining an outline of the learning process etc. in the first embodiment. [Figure 6] FIG. 6 is a diagram for explaining an outline of the learning process and the like in the first embodiment. [Figure 7] FIG. 7 is a diagram for explaining an outline of the learning process and the like in the first embodiment. [Figure 8] FIG. 8 is a diagram for explaining an outline of the learning process and the like in the first embodiment. [Figure 9] FIG. 9 is a flowchart illustrating details of the learning process in the first embodiment. [Figure 10] FIG. 10 is a flowchart illustrating details of the learning process in the first embodiment. [Figure 11] FIG. 11 is a flowchart illustrating details of the learning process in the first embodiment. [Figure 12] FIG. 12 is a flowchart illustrating details of the learning process in the first embodiment. [Figure 13] FIG. 13 is a flowchart illustrating details of the learning process in the first embodiment. [Figure 14] FIG. 14 is a flowchart illustrating details of the estimation process in the first embodiment. [Figure 15] FIG. 15 is a diagram illustrating a specific example of the configuration of the learning model MD1 in the first embodiment. [Figure 16] FIG. 16 is a diagram illustrating a specific example of the configuration of the learning model MD2 in the first embodiment. [Figure 17] FIG. 17 is a diagram illustrating the effect of the learning process in the first embodiment. [Figure 18] FIG. 18 is a diagram illustrating the effect of the learning process in the first embodiment. [Figure 19] FIG. 19 is a diagram for explaining an outline of the learning process etc. in the first modified example. [Figure 20] FIG. 20 is a diagram for explaining an outline of the learning process etc. in the first modified example. [Figure 21] FIG. 21 is a flowchart illustrating the details of the learning process in the first modified example. [Figure 22] FIG. 22 is a flowchart illustrating details of the estimation process in the first modified example. [Figure 23] FIG. 23 is a diagram illustrating a specific example of the configuration of the learning model MD1 in the first modified example. [Figure 24]FIG. 24 is a diagram illustrating the effect of the learning process in the first modified example. DETAILED DESCRIPTION OF THE INVENTION

[0009] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. However, such descriptions should not be interpreted in a limiting sense, and do not limit the subject matter described in the claims. Furthermore, various changes, substitutions, and modifications can be made without departing from the spirit and scope of the present disclosure. Furthermore, different embodiments can be combined as appropriate.

[0010] [Configuration Example of Information Processing System 10 in First Embodiment] First, a description will be given of an example of the configuration of an information processing system 10 according to the first embodiment. Figures 1 and 2 are diagrams showing an example of the configuration of an information processing system 10 according to the first embodiment.

[0011] The information processing system 10 includes, for example, an information processing device 1 and a measurement device 2. The information processing device 1 is connected to the measurement device 2, for example, by wire or wirelessly, and is capable of communicating with the measurement device 2.

[0012] The measuring device 2 is, for example, a device that measures electrocardiogram information of the subject OB (hereinafter also simply referred to as a person). Specifically, the measuring device 2 measures, for example, as the electrocardiogram information of the subject OB, at least one of electrocardiogram waveform data of the subject OB, a potential difference in the electrocardiogram waveform data of the subject OB, and the heart rate of the subject OB. The electrocardiogram waveform data of the subject OB may be, for example, electrocardiogram waveform data for a predetermined number of beats in a 12-lead electrocardiogram (e.g., electrocardiogram waveform data for two beats). The electrocardiogram waveform data of the subject OB may also be, for example, electrocardiogram waveform data for a predetermined time in a 12-lead electrocardiogram (e.g., electrocardiogram waveform data for three seconds). The electrocardiogram waveform data of the subject OB may also be, for example, electrocardiogram waveform data for a predetermined time in a unipolar lead electrocardiogram (e.g., electrocardiogram waveform data for three seconds).

[0013] The information processing device 1 is a computer device, for example, a general-purpose PC (Personal Computer). The information processing device 1 performs a process (hereinafter also referred to as a learning process) to generate a learning model that estimates data indicating the likelihood of a future major cardiovascular event occurring in the subject OB (hereinafter also referred to as EHFI (Electrical Heart Failure Indicator) data) by using, for example, electrocardiogram information generated by the measurement device 2. The information processing device 1 also performs a process (hereinafter also referred to as an estimation process) to estimate EHFI data in the subject OB by using, for example, the learning model generated in the learning process. Note that, hereinafter, the learning process and the estimation process are collectively referred to simply as learning process, etc.

[0014] Specifically, the information processing device 1 has the hardware configuration of a general-purpose computer device, and includes, for example, a CPU 101 which is a processor, a memory 102, a communication device 103, a storage medium 104 (hereinafter also referred to as the storage unit 104), and an output device 105, as shown in Fig. 2. Each unit is connected to each other via a bus 106.

[0015] The storage medium 104 has, for example, a program storage area (not shown) that stores a program 110 for performing a learning process, etc. The storage medium 104 also has, for example, an information storage area 130 that stores information used when performing the learning process, etc. The storage medium 104 may be, for example, a hard disk drive (HDD) or a solid state drive (SSD).

[0016] As shown in FIG. 2, the CPU 101 performs learning processing and the like by executing a program 110 loaded into the memory 102 from the storage medium 104, for example.

[0017] The communication device 103 communicates with the measurement device 2, for example, by wired communication or wireless communication.

[0018] The output device 105 is, for example, a device such as a display, etc. Specifically, the output device 105 is, for example, a device that outputs EHFI data estimated in the estimation process.

[0019] Note that, although the following description will be given assuming that the learning process and the like are performed in a single information processing device 1, the present invention is not limited to this. Specifically, the learning process and the like may be performed in a distributed manner in, for example, a plurality of information processing devices 1. More specifically, the information processing device 1 that performs the learning process may be a device different from the information processing device 1 that performs the estimation process, for example.

[0020] [Outline of the first embodiment] Next, an outline of the first embodiment will be described. Fig. 3 and Fig. 4 are diagrams for explaining an outline of the learning process etc. in the first embodiment. Specifically, Fig. 3 is a block diagram for explaining the functions of the information processing device 1. Fig. 4 is a diagram for explaining information stored in the information storage area 130. Figs. 5 to 8 are diagrams for explaining an outline of the learning process etc. in the first embodiment.

[0021] 3, the information processing device 1 realizes various functions including, for example, a data acquisition unit 111, a first data generation unit 112, a first model generation unit 113, a second data generation unit 114, a second model generation unit 115, a risk estimation unit 116, and a result output unit 117. Note that the data acquisition unit 111, the first data generation unit 112, the first model generation unit 113, the second data generation unit 114, and the second model generation unit 115 are functions that realize a learning process. Also, the data acquisition unit 111, the risk estimation unit 116, and the result output unit 117 are functions that realize an estimation process.

[0022] 4, the information storage area 130 stores, for example, electrocardiogram information 131, age data 132, BNP (Brain Natriuretic Peptide) data 133, corrected BNP data 134, teacher data DT1 (hereinafter also referred to as first teacher data DT1), teacher data DT2 (hereinafter also referred to as second teacher data DT2), learning model MD1 (hereinafter also referred to as first learning model MD1), and learning model MD2 (hereinafter also referred to as second learning model MD2). Note that the BNP data 133 may be, for example, data on NT-proBNP (human brain natriuretic peptide precursor N-terminal fragment).

[0023] First, the function for realizing the learning process will be described.

[0024] The data acquiring unit 111 acquires, for example, electrocardiogram information 131 measured by the measuring device 2 (electrocardiogram information 131 for each subject OB) for each of the multiple subjects OB from the measuring device 2. The electrocardiogram information 131 is, for example, information including at least one of electrocardiogram waveform data for each subject OB, a potential difference in the electrocardiogram waveform data for each subject OB, and a heart rate for each subject OB. Then, the data acquiring unit 111 stores the acquired electrocardiogram information 131 for each of the multiple subjects OB in the information storage area 130. Note that the electrocardiogram information 131 may be input to the information processing device 1 by an operator, for example.

[0025] Furthermore, the data acquiring unit 111 acquires, for example, age data 132 (data indicating the age of each subject OB) input by the worker for each of the multiple subjects OB. Then, the data acquiring unit 111 stores the acquired age data 132 in the information storage area 130 for each of the multiple subjects OB.

[0026] Furthermore, the data acquiring unit 111 acquires, for example, for each of a plurality of subjects OB, BNP data 133 (data indicating the BNP value of each subject OB) measured by another measuring device (not shown) from the other measuring device. Then, the data acquiring unit 111 stores, for example, for each of a plurality of subjects OB, the acquired BNP data 133 in the information storage area 130. Note that the BNP data 133 may be input to the information processing device 1 by an operator, for example.

[0027] 5(A), the first data generation unit 112 generates, for example, for each of a plurality of subjects OB, a plurality of pieces of teacher data DT1 including electrocardiogram information 131 (electrocardiogram information 131 corresponding to each subject OB) stored in the information storage area 130, age data 132 (age data 132 corresponding to each subject OB) stored in the information storage area 130, and BNP data 133 (BNP data 133 corresponding to each subject OB) stored in the information storage area 130. That is, each of the plurality of pieces of teacher data DT1 is teacher data including, for example, the BNP data 133 as a teacher label (correct label). Then, the first data generation unit 112 stores, for example, the generated plurality of pieces of teacher data DT1 in the information storage area 130.

[0028] The first data generation unit 112 may, for example, convert the BNP data 133 stored in the information storage area 130 before including it in the teacher data DT1. Specifically, the first data generation unit 112 may, for example, convert the BNP data 133 stored in the information storage area 130 using a function such as arctangent before including it in the teacher data DT1. The first data generation unit 112 may also generate multiple pieces of teacher data DT1 so that, for example, electrocardiogram information 131 and BNP data 133 whose measurement timing difference is equal to or less than a predetermined threshold (for example, 30 days) are included in the same teacher data DT1. For example, when multiple combinations of electrocardiogram information 131 and BNP data 133 are measured from the same subject OB, the first data generation unit 112 may also generate multiple pieces of teacher data DT1 corresponding to the subject OB.

[0029] 5(B), the first model generation unit 113 generates a learning model MD1 by learning, for example, a plurality of pieces of training data DT1 stored in the information storage area 130. Then, the first model generation unit 113 stores the generated learning model MD1 in the information storage area 130, for example.

[0030] That is, the first model generation unit 113 generates a learning model MD1 that outputs a predicted value of BNP data 133 (hereinafter also referred to as BNP data 133a) in response to input of, for example, electrocardiogram information 131 and age data 132 acquired by the data acquisition unit 111.

[0031] As shown in Figure 6(A), the second data generation unit 114 acquires BNP data 133a output by inputting, for example, electrocardiogram information 131 (electrocardiogram information 131 corresponding to each subject OB) stored in the information storage area 130 and age data 132 (age data 132 corresponding to each subject OB) stored in the information storage area 130 into a learning model MD1 for each of multiple subjects OB.

[0032] 6(B), the second data generation unit 114 calculates corrected BNP data 134 (hereinafter also referred to as corrected BNP data 134) for each of the multiple subjects OB by using the output BNP data 133a (the BNP data 133a corresponding to each subject OB) and a coefficient corresponding to each subject OB. The coefficient corresponding to each subject OB is a coefficient determined according to the occurrence status of major cardiovascular events in each subject OB, for example. Specifically, the second data generation unit 114 calculates the corrected BNP data 134 for each of the multiple subjects OB by multiplying the output BNP data 133a by the coefficient corresponding to each subject OB, for example.

[0033] In addition, the second data generation unit 114 may calculate corrected BNP data 134 for each of a plurality of subjects OB, for example, by using the BNP data 133 (BNP data 133 corresponding to each subject OB) stored in the information storage area 130 and a coefficient corresponding to each subject OB.

[0034] 7(A), for example, for each of the plurality of subjects OB, the second data generation unit 114 generates a plurality of teacher data DT2 including electrocardiogram information 131 (electrocardiogram information 131 corresponding to each subject OB) stored in the information storage area 130, age data 132 (age data 132 corresponding to each subject OB) stored in the information storage area 130, and the calculated corrected BNP data 134. That is, each of the plurality of teacher data DT2 is teacher data including, for example, the corrected BNP data 134 as a teacher label (correct answer label). Then, the second data generation unit 114 stores, for example, the generated plurality of teacher data DT2 in the information storage area 130.

[0035] The second data generation unit 114 may generate the plurality of second teacher data DT2 by, for example, reusing the electrocardiogram information 131 and age data 132 used to generate the first teacher data DT1 from among the electrocardiogram information 131 and age data 132 stored in the information storage area 130, or may generate the plurality of second teacher data DT2 without using the electrocardiogram information 131 and age data 132 used to generate the first teacher data DT1. The second data generation unit 114 may generate the plurality of teacher data DT2 so that the same teacher data DT2 includes electrocardiogram information 131 and BNP data 133 whose measurement timing difference is equal to or less than a predetermined threshold (e.g., 30 days). For example, when multiple combinations of electrocardiogram information 131 and BNP data 133 are measured from the same subject OB, the second data generation unit 114 may generate multiple teacher data DT2 corresponding to the subject OB.

[0036] 7(B), the second model generation unit 115 generates a learning model MD2 by learning, for example, a plurality of pieces of training data DT2 stored in the information storage area 130. Then, the second model generation unit 115 stores the generated learning model MD2 in the information storage area 130, for example.

[0037] That is, the second model generation unit 115 generates a learning model MD2 that outputs a predicted value of corrected BNP data 134 (hereinafter also referred to as EHFI data 134a) in response to input of, for example, electrocardiogram information 131 and age data 132 acquired by the data acquisition unit 111.

[0038] Specifically, the learning model MD2 may be a learning model including, for example, a predetermined number of layers (hereinafter simply referred to as the predetermined number of layers) from the top of the multiple layers constituting the first learning model MD1, and other layers (hereinafter simply referred to as the other layers) that receive the output from the predetermined number of layers as input. In other words, the learning model MD2 may be a learning model generated by transfer learning using the learning results of the learning model MD1. The second model generation unit 115 may, for example, tune parameters in the other layers constituting the learning model MD2.

[0039] The first model generation unit 113 may generate, for example, a plurality of learning models MD1. In this case, other layers in the learning model MD2 may receive, as input, outputs from a predetermined number of layers in each of the plurality of first learning models. In other words, the learning model MD2 may be generated by ensemble learning using, for example, a plurality of learning models MD1.

[0040] Furthermore, the multiple learning models MD1 may each use a different loss function, for example. Specifically, the multiple learning models MD1 may each include, for example, a learning model MD1 using Huber Loss as the loss function and another learning model MD1 using Mean Square Logarithmic Error as the loss function.

[0041] Furthermore, each of the multiple learning models MD1 may include, for example, a learning model MD1 trained using teacher data DT transformed by a function such as arctangent, and another learning model MD1 trained using teacher data DT that has not been transformed by a function such as arctangent.

[0042] Furthermore, the multiple learning models MD1 may include, for example, a learning model MD1 generated by learning multiple training data DT1 adjusted so that the number of training data DT1 whose values ​​corresponding to BNP data 133 are equal to or greater than a predetermined threshold (hereinafter simply referred to as the predetermined threshold) is the same as the number of training data DT1 whose values ​​corresponding to BNP data 133 are less than the predetermined threshold.

[0043] Next, the function for realizing the estimation process will be described.

[0044] The data acquiring unit 111 acquires, for example, electrocardiogram information 131 measured by the measuring device 2 (hereinafter also referred to as new electrocardiogram information 131) from the measuring device 2.

[0045] Furthermore, the data acquiring unit 111 acquires, for example, age data 132 (hereinafter also referred to as new age data 132) input by the worker for each of the plurality of subjects OB. In the following description, it is assumed that the new electrocardiogram information 131 and the new age data 132 are information (data) corresponding to the same subject OB.

[0046] As shown in FIG. 8, the risk estimation unit 116 inputs, for example, new electrocardiogram information 131 and new age data 132 acquired by the data acquisition unit 111 into the learning model MD2 stored in the information storage area 130.

[0047] Then, the risk estimation unit 116 acquires, for example, the EHFI data 134a output from the learning model MD2 in response to the input of the new electrocardiogram information 131 and the new age data 132.

[0048] The result output unit 117 outputs, for example, the EHFI data 134a acquired by the risk estimation unit 116 to the output device 105.

[0049] That is, when generating a learning model MD2 used in the estimation process, the information processing device 1 in this embodiment uses teacher data DT2 including corrected BNP data 134 corrected by a coefficient corresponding to each subject OB. Specifically, in this case, the information processing device 1 determines a coefficient corresponding to each subject OB such that, for example, the coefficient corresponding to the subject OB who has previously experienced a major cardiovascular event is larger than the coefficient corresponding to the subject OB who has not previously experienced a major cardiovascular event. In other words, the information processing device 1 generates a learning model MD2 that determines that a subject OB who has previously experienced a major cardiovascular event is more likely to experience a major cardiovascular event in the future than a subject OB who has not previously experienced a major cardiovascular event.

[0050] This enables the information processing device 1 in this embodiment to generate a learning model MD2 that can accurately predict the likelihood of a future major cardiovascular event, for example.

[0051] Furthermore, the information processing device 1 in this embodiment can generate a learning model MD2 that can estimate the likelihood of a major cardiovascular event occurring according to the age of each subject OB, for example, by including age data 132 indicating the age of each subject OB in each of the teacher data DT1 used to generate the learning model MD1 and the teacher data DT2 used to generate the learning model MD2.

[0052] In particular, the normal value of a child's heart rate tends to vary more significantly with age than, for example, the normal value of an adult's heart rate. Furthermore, the normal value of the potential difference (hereinafter simply referred to as potential difference) in a child's electrocardiogram waveform data tends to vary more significantly with age than, for example, the normal value of the potential difference in an adult's electrocardiogram waveform data. Therefore, the information processing device 1 according to the present embodiment can generate a learning model MD2 capable of accurately estimating the likelihood of a major cardiovascular event occurring according to the age of each subject OB by using, for example, training data DT including the subject OB's heart rate, the subject OB's potential difference, and the subject OB's age data 132. Therefore, by using, for example, the learning model MD2, the information processing device 1 according to the present embodiment can accurately estimate the likelihood of a future major cardiovascular event occurring even when the subject OB is a child.

[0053] [Details of Learning Process in First Embodiment] Next, the learning process in the first embodiment will be described in detail below. Figures 9 to 13 are flowcharts illustrating the learning process in the first embodiment in detail.

[0054] [Electrocardiogram information storage processing] First, a process of storing electrocardiogram information 131 (hereinafter also referred to as electrocardiogram information storage process) among the learning processes in the first embodiment will be described. Fig. 9 is a flowchart illustrating the electrocardiogram information storage process in the first embodiment.

[0055] 9, the data acquiring unit 111 waits, for example, until it receives electrocardiogram information 131 measured by the measuring device 2 from the measuring device 2 (NO in S11). Specifically, the data acquiring unit 111 waits, for example, until it receives electrocardiogram information 131 measured by the measuring device 2 from the measuring device 2.

[0056] Then, when the electrocardiogram information 131 is received from the measurement device 2 (YES in S11), the data acquisition unit 111 stores, for example, the acquired electrocardiogram information 131 in the information storage area 130 (S12).

[0057] [Age data memory processing] Next, a process for storing the age data 132 (hereinafter also referred to as age data storage process) of the learning process in the first embodiment will be described. Fig. 10 is a flowchart illustrating the age data storage process in the first embodiment.

[0058] As shown in FIG. 10, the data acquisition unit 111 waits until it receives input of age data 132 by the worker (NO in S21), for example.

[0059] Then, when input of age data 132 is accepted (YES in S11), the data acquiring unit 111 stores, for example, the accepted input of age data 132 in the information storage area 130 (S22).

[0060] [BNP Data Storage Processing] Next, a description will be given of the process of storing the BNP data 133 (hereinafter also referred to as BNP data storage process) out of the learning process in the first embodiment. Fig. 11 is a flowchart illustrating the BNP data storage process in the first embodiment.

[0061] 11, the data acquiring unit 111 waits, for example, until it receives BNP data 133 measured in another measuring device (not shown) from the other measuring device (NO in S31). Specifically, the data acquiring unit 111 waits, for example, until it receives BNP data 133 measured in the other measuring device and transmits it from the other measuring device.

[0062] Then, when BNP data 133 is received from another measurement device (YES in S31), the data acquisition section 111 stores, for example, the acquired BNP data 133 in the information storage area 130 (S32).

[0063] [First model generation process] Next, a process for generating a learning model MD1 (hereinafter also referred to as a first model generation process) will be described among the learning processes in the first embodiment. Fig. 12 is a flowchart illustrating the first model generation process in the first embodiment.

[0064] As shown in FIG. 12, the first data generation unit 112 generates, for example, for each of a plurality of subjects OB, a plurality of teacher data DT1 including electrocardiogram information 131 (electrocardiogram information 131 corresponding to each subject OB) stored in the information storage area 130, age data 132 (age data 132 corresponding to each subject OB) stored in the information storage area 130, and BNP data 133 (BNP data 133 corresponding to each subject OB) stored in the information storage area 130 (S41).

[0065] In the process of S41, the first data generation unit 112 may generate a plurality of pieces of teacher data DT1 so that, for example, electrocardiogram information 131 and BNP data 133 whose difference in measurement timing is equal to or less than a predetermined threshold (for example, 30 days) are included in the same teacher data DT1. In addition, in the process of S41, for example, when a plurality of combinations of electrocardiogram information 131 and BNP data 133 are measured from the same subject OB, the first data generation unit 112 may generate a plurality of pieces of teacher data DT1 corresponding to the subject OB.

[0066] Then, the first data generating unit 112 stores, for example, the plurality of pieces of teacher data DT1 generated in the process of S41 in the information storage area 130 (S42).

[0067] Next, the first model generation unit 113 generates a learning model MD1 by, for example, learning the plurality of pieces of training data DT1 stored in the information storage area 130 (S43).

[0068] That is, in the process of S43, the first model generation unit 113 generates a learning model MD1 that outputs BNP data 133a in response to input of the electrocardiogram information 131 and age data 132 acquired by the data acquisition unit 111, for example.

[0069] Then, the first model generation unit 113 stores, for example, the learning model MD1 generated in the process of S43 in the information storage area 130 (S44).

[0070] [Second model generation process] Next, a process for generating a learning model MD2 (hereinafter also referred to as a second model generation process) will be described in the learning process in the second embodiment. Fig. 13 is a flowchart for explaining the second model generation process in the first embodiment.

[0071] As shown in FIG. 13, the second data generating unit 114, for example, identifies a coefficient corresponding to each of the plurality of subjects OB (S51).

[0072] Specifically, the second data generation circuitry 114 may specify a first coefficient as the coefficient corresponding to the subject OBa if the subject OBa has experienced one major cardiovascular event within a predetermined period (hereinafter simply referred to as the predetermined period) from the time when the electrocardiogram information 131 of a specific subject OB (hereinafter also referred to as the subject OBa) among multiple subjects OB was measured, and may specify a second coefficient smaller than the first coefficient as the coefficient corresponding to the subject OBa if the subject OBa has not experienced a major cardiovascular event within the predetermined period. The predetermined period may be, for example, 180 days. The first coefficient may be, for example, "2.0," and the second coefficient may be, for example, "1.0."

[0073] Furthermore, for example, if the subject OBa has experienced one major cardiovascular event within a period shorter than the predetermined period, the second data generating circuitry 114 may specify a third coefficient greater than the first coefficient as the coefficient corresponding to the subject OBa. Note that the third coefficient may be, for example, 2.5.

[0074] Furthermore, for example, if the subject OBa has experienced multiple major cardiovascular events within a predetermined period, the second data generating circuitry 114 may identify a fourth coefficient greater than the first coefficient as the coefficient corresponding to the subject OBa. The fourth coefficient may be, for example, "3.0."

[0075] Furthermore, for example, when the subject OBa does not experience a major cardiovascular event within a period shorter than a predetermined period, the second data generating circuitry 114 may specify a fifth coefficient, which is larger than the second coefficient and smaller than the first coefficient, as the coefficient corresponding to the subject OBa. Note that the fifth coefficient may be, for example, "1.5."

[0076] Next, the second data generation unit 114 acquires BNP data 133a output by inputting, for example, for each of a plurality of subjects OB, electrocardiogram information 131 (electrocardiogram information 131 corresponding to each subject OB) stored in the information storage area 130 and age data 132 (age data 132 corresponding to each subject OB) stored in the information storage area 130 into a learning model MD1 (S52).

[0077] Then, the second data generation unit 114 calculates corrected BNP data 134 for each of a plurality of subjects OB, for example, by using the BNP data 133a obtained in the processing of S52 (BNP data 133a corresponding to each subject OB) and the coefficient identified in the processing of S51 (coefficient corresponding to each subject OB) (S53).

[0078] In the processing of S53, the second data generation unit 114 may calculate corrected BNP data 134 for each of a plurality of subjects OB by using the BNP data 133 (BNP data 133 corresponding to each subject OB) stored in the information storage area 130 and the coefficient (coefficient corresponding to each subject OB) identified in the processing of S51. In this case, the second data generation unit 114 may not perform the processing of S52.

[0079] Next, the second data generation unit 114 generates, for example, for each of a plurality of subjects OB, a plurality of teacher data DT2 including electrocardiogram information 131 (electrocardiogram information 131 corresponding to each subject OB) stored in the information storage area 130, age data 132 (age data 132 corresponding to each subject OB) stored in the information storage area 130, and corrected BNP data 134 (corrected BNP data 134 corresponding to each subject OB) calculated in the processing of S53 (S54).

[0080] Then, the second data generating unit 114 stores, for example, the plurality of pieces of teacher data DT2 generated in the process of S54 in the information storage area 130 (S55).

[0081] Thereafter, the second model generation unit 115 generates a learning model MD2 by, for example, learning the plurality of pieces of training data DT2 stored in the information storage area 130 (S56).

[0082] Then, the second model generation unit 115 stores the learning model MD2 generated in the process of S56 in the information storage area 130 (S57).

[0083] That is, the second model generation unit 115 generates a learning model MD2 that outputs EHFI data 134a in response to input of electrocardiogram information 131 and age data 132 acquired by the data acquisition unit 111, for example.

[0084] [Details of Estimation Processing in the First Embodiment] Next, the details of the estimation process in the first embodiment will be described below with reference to Fig. 14, which is a flowchart illustrating the details of the estimation process in the first embodiment.

[0085] The data acquiring unit 111 acquires, for example, new electrocardiogram information 131 measured by the measuring device 2 from the measuring device 2. In addition, the data acquiring unit 111 acquires, for example, new age data 132 measured by another measuring device from the other measuring device (S61).

[0086] Then, the risk estimation unit 116 inputs, for example, the new electrocardiogram information 131 and the new age data 132 acquired in the process of S61 into the learning model MD2 (S62).

[0087] Next, the risk estimation unit 116 acquires, for example, the EHFI data 134a output from the learning model MD2 in response to the input of the new electrocardiogram information 131 and the new age data 132 in the processing of S62 (S63).

[0088] Thereafter, the result output unit 117 outputs, for example, the EHFI data 134a acquired in the process of S63 to the output device 105 (S64).

[0089] This enables the information processing device 1 in the present embodiment to output, for example, EHFI data 134a corresponding to the subject OB for which the new electrocardiogram information 131 and new age data 132 received in the processing of S61 have been measured. In other words, the information processing device 1 can present, for example, the EHFI data 134a corresponding to each subject OB as one of the indices indicating the likelihood of a future major cardiovascular event occurring in each subject OB.

[0090] [Specific examples of the configuration of learning model MD1 and learning model MD2] Next, a specific example of the configuration of the learning model MD1 and the learning model MD2 in the first embodiment will be described. Specifically, FIG. 15 is a diagram illustrating a specific example of the configuration of the learning model MD1 in the first embodiment. Also, FIG. 16 is a diagram illustrating a specific example of the configuration of the learning model MD2 in the first embodiment. Below, the description will be given assuming that the learning model MD1 and the learning model MD2 are each learning models configured by a convolutional neural network. Also, below, the description will be given assuming that the learning model MD2 includes four learning models MD1.

[0091] In the example shown in FIG. 15, the learning model MD1 is denoted as "CNN Model." In the example shown in FIG. 16, the four learning models MD1 are denoted as "Model A," "Model B," "Model C," and "Model D," respectively. In the examples shown in FIGS. 15 and 16, the convolutional layer is denoted as "Convolution," the batch normalization layer is denoted as "Batch Normalization," the squeeze-and-excitation layer is denoted as "Squeeze-and-Excitation," the dropout layer is denoted as "Dropout," and the fully connected layer is denoted as "Dense." In the example shown in FIG. 15, a block including a convolutional layer, a batch normalization layer, and a squeeze-and-excitation layer (hereinafter also referred to as a convolutional block) is denoted as a "Convolutional Block." 15 and 16, the electrocardiogram waveform data included in the electrocardiogram information 131 is expressed as "12-lead ECG 2 beats", the potential difference included in the electrocardiogram information 131 is expressed as "Voltage", the heart rate included in the electrocardiogram information 131 is expressed as "HR", the age data 132 is expressed as "Age", and the EHFI data 134a is expressed as "EHFI". In the example shown in Fig. 16, the notations for the four learning models MD1 included in the learning model MD2 are abbreviated.

[0092] Specifically, in the learning model MD1, for example, electrocardiogram waveform data included in electrocardiogram information 131 is input to seven layers of convolutional blocks, a drop layer, and a fully connected layer, in that order, as shown in Fig. 15. Also, in the learning model MD1, for example, the potential difference and heart rate included in the electrocardiogram information 131 and age data 132 are input to the squeeze-and-excitation layer, as shown in Fig. 15. Also, in the learning model MD1, for example, the age data 132 included in the electrocardiogram information 131 is input to the fully connected layer, as shown in Fig. 15.

[0093] In addition, in the learning model MD2, for example, the outputs from the seven-layer convolutional blocks in each of the four learning models MD1 are input to the first drop layer, the first fully connected layer, the second drop layer, and the second fully connected layer in that order, as shown in Fig. 16. In addition, in the learning model MD2, for example, age data 132 is input to the first drop layer, as shown in Fig. 16.

[0094] Then, in the learning model MD2, for example, EHFI data 134a is output from the second fully connected layer as shown in Fig. 16. That is, the learning model MD2 is a learning model trained by, for example, transfer learning and ensemble learning, as shown in Fig. 16.

[0095] This makes it possible for the learning model MD2 in this embodiment to improve the estimation accuracy of the EHFI data 134a corresponding to each subject OB, for example.

[0096] [Effects of the learning process in the first embodiment] 17 and 18 are diagrams illustrating the effect of the learning process in the first embodiment. Specifically, FIG. 17 is an ROC curve showing the performance of the learning model MD2 generated in the learning process in the first embodiment. Also, FIG. 18 is a graph showing the prediction accuracy of each of the prediction of major cardiovascular events using BNP data 133 (i.e., prediction by the conventional method) and the prediction of major cardiovascular events using EHFI data 134a output from the learning model MD2 (i.e., prediction by the method in the first embodiment).

[0097] Specifically, Figure 17 shows that the area under the ROC curve (AUC) is 0.826, which indicates that the prediction accuracy of the learning model MD2 is sufficiently high.

[0098] 18 also shows that, for example, the prediction accuracy (graph G1) using the EHFI data 134a is higher than the prediction accuracy (graph G2) using the BNP data 133. In particular, Fig. 18 shows that, for example, the prediction accuracy using the EHFI data 134a for a period up to about 100 days after the subject OB measured the electrocardiogram information 131 is higher than the prediction accuracy using the BNP data 133 for a period up to about 100 days after the subject OB measured the electrocardiogram information 131.

[0099] That is, FIG. 18 shows that, for example, prediction of major cardiovascular events using the learning model MD2 is particularly useful in the period up to about 100 days after subject OB has had the electrocardiogram information 131 measured.

[0100] In this way, the information processing device 1 in this embodiment acquires, for example, electrocardiogram information 131 (hereinafter also referred to as first electrocardiogram information 131) of the subject OB (hereinafter also referred to as first subject OB) and BNP data 133 (hereinafter also referred to as first BNP data 133) of the first subject OB. Then, the information processing device 1 generates, for example, teacher data DT1 (hereinafter also referred to as first teacher data DT1) including the first electrocardiogram information 131 and the first BNP data 133. Furthermore, the information processing device 1 generates a first learning model MD1 by, for example, learning the first teacher data DT1.

[0101] Thereafter, the information processing device 1 acquires, for example, electrocardiogram information 131 (hereinafter also referred to as second electrocardiogram information 131) of the subject OB (hereinafter also referred to as second subject OB) and BNP data 133 (hereinafter also referred to as second BNP data 133) of the second subject OB. Then, the information processing device 1 calculates corrected BNP data 134, for example, by using the second BNP data 133 and a coefficient corresponding to the second subject OB. Subsequently, the information processing device 1 generates, for example, second teacher data DT2 including the second electrocardiogram information 131 and the corrected BNP data 134. Furthermore, the information processing device 1 generates a second learning model MD2 including at least a portion of the layers constituting the first learning model MD1, for example, by training the second teacher data DT2.

[0102] That is, when generating a learning model MD2 used in the estimation process, the information processing device 1 in this embodiment uses teacher data DT2 including corrected BNP data 134 corrected by a coefficient corresponding to each subject OB. Specifically, in this case, the information processing device 1 determines a coefficient corresponding to each subject OB such that, for example, the coefficient corresponding to the subject OB who has previously experienced a major cardiovascular event is larger than the coefficient corresponding to the subject OB who has not previously experienced a major cardiovascular event. In other words, the information processing device 1 generates a learning model MD2 that determines that a subject OB who has previously experienced a major cardiovascular event is more likely to experience a major cardiovascular event in the future than a subject OB who has not previously experienced a major cardiovascular event.

[0103] This enables the information processing device 1 in this embodiment to generate a learning model MD2 that can accurately predict the likelihood of a future major cardiovascular event, for example.

[0104] Specifically, the information processing device 1 in this embodiment acquires, for example, age data 132 of the first subject OB (hereinafter also referred to as first age data 132). Then, the information processing device 1 generates, for example, first teacher data DT1 including the first age data 132. Furthermore, the information processing device 1 generates a first learning model MD1 by, for example, learning the first teacher data DT1 including the first age data 132. Thereafter, the information processing device 1 acquires, for example, age data 132 (hereinafter also referred to as second age data 132) of the subject OB (hereinafter also referred to as second subject OB). Then, the information processing device 1 generates, for example, second teacher data DT2 including the second age data 132. Furthermore, the information processing device 1 generates a second learning model MD2 by, for example, learning the second teacher data DT2 including the second age data 132.

[0105] As a result, the information processing device 1 in this embodiment can generate a learning model MD2 that can estimate the likelihood of a major cardiovascular event occurring according to the age of each subject OB, for example, by including age data 132 indicating the age of each subject OB in each of the teacher data DT1 used to generate the learning model MD1 and the teacher data DT2 used to generate the learning model MD2. Therefore, the information processing device 1 can accurately estimate the likelihood of a future major cardiovascular event occurring even if the subject OB is a child, for example.

[0106] [Outline of the first modified example] Next, an outline of a modified example of the first embodiment (hereinafter also referred to as the first modified example) will be described. Figures 19 and 20 are diagrams for explaining an outline of the learning process etc. in the first modified example. Below, differences from the first embodiment will be described.

[0107] First, the function for realizing the learning process will be described.

[0108] 19(A), for example, for each of a plurality of subjects OB, the first data generation unit 112 calculates corrected BNP data 134 by using the BNP data 133 (BNP data 133 corresponding to each subject OB) stored in the information storage area 130 and a coefficient corresponding to each subject OB. Specifically, for example, the first data generation unit 112 calculates corrected BNP data 134 by multiplying the stored BNP data 133 by the coefficient corresponding to each subject OB for each of a plurality of subjects OB.

[0109] 19(B), for each of the plurality of subjects OB, the first data generation unit 112 generates a plurality of pieces of teacher data DT1 (hereinafter also referred to as a plurality of pieces of teacher data DT1a) including electrocardiogram information 131 (electrocardiogram information 131 corresponding to each subject OB) stored in the information storage area 130, age data 132 (age data 132 corresponding to each subject OB) stored in the information storage area 130, and the calculated corrected BNP data 134. Thereafter, the first data generation unit 112 stores the generated plurality of pieces of teacher data DT1a in the information storage area 130, for example.

[0110] 20(A), the first model generation unit 113 generates a learning model MD1 (hereinafter also referred to as a learning model MD1a) by learning, for example, a plurality of pieces of training data DT1a stored in the information storage area 130. Then, the first model generation unit 113 stores the generated learning model MD1a in the information storage area 130, for example.

[0111] Next, the function for realizing the estimation process will be described.

[0112] As shown in FIG. 20(B), the risk estimation unit 116 inputs, for example, new electrocardiogram information 131 and new age data 132 acquired by the data acquisition unit 111 into the learning model MD1a stored in the information storage area 130.

[0113] Then, the risk estimation unit 116 acquires, for example, the EHFI data 134a output from the learning model MD1a in response to the input of new electrocardiogram information 131 and new age data 132.

[0114] The result output unit 117 outputs, for example, the EHFI data 134a acquired by the risk estimation unit 116 to the output device 105.

[0115] That is, unlike the information processing device 1 in the first embodiment, for example, the information processing device 1 in this modified example generates a single learning model MD1a that can directly output EHFI data 134a in response to input of new electrocardiogram information 131 and new age data 132.

[0116] This makes it possible for the information processing device 1 in this modification to reduce the time and workload required to generate a learning model (learning model MD1a) that predicts the likelihood of a future major cardiovascular event occurring, for example.

[0117] [Details of the learning process in the first modified example] Next, the details of the learning process in the first modified example will be explained. Fig. 21 is a flowchart illustrating the details of the learning process in the first modified example. Note that the electrocardiogram information storage process, age data storage process, and BNP data storage process in the first modified example are the same as those in the first embodiment, so their explanations will be omitted.

[0118] [First model generation process] FIG. 21 is a flowchart illustrating the first model generation process in the first modified example.

[0119] As shown in FIG. 21, the first data generating unit 112, for example, identifies a coefficient corresponding to each of the plurality of subjects OB (S71).

[0120] Then, the first data generation unit 112 calculates corrected BNP data 134 for each of a plurality of subjects OB, for example, by using the BNP data 133 (BNP data 133 corresponding to each subject OB) stored in the information storage area 130 and the coefficient (coefficient corresponding to each subject OB) identified in the processing of S71 (S72).

[0121] Next, the first data generation unit 112 generates a plurality of teacher data DT1a including, for example, for each of a plurality of subjects OB, electrocardiogram information 131 (electrocardiogram information 131 corresponding to each subject OB) stored in the information storage area 130, age data 132 (age data 132 corresponding to each subject OB) stored in the information storage area 130, and corrected BNP data 134 (corrected BNP data 134 corresponding to each subject OB) calculated in the processing of S72 (S73).

[0122] Then, the first data generating unit 112 stores, for example, the plurality of pieces of teacher data DT1a generated in the process of S73 in the information storage area 130 (S74).

[0123] Thereafter, the first model generation unit 113 generates a learning model MD1a by, for example, learning the plurality of training data DT1a stored in the information storage area 130 (S75).

[0124] Then, the first model generation unit 113 stores, for example, the learning model MD1a generated in the process of S75 in the information storage area 130 (S76).

[0125] [Details of Estimation Processing in the First Modification] Next, the details of the estimation process in the first modified example will be described below. Fig. 22 is a flowchart illustrating the details of the estimation process in the first modified example.

[0126] The data acquiring unit 111 acquires, for example, new electrocardiogram information 131 measured by the measuring device 2 from the measuring device 2. The data acquiring unit 111 also acquires, for example, new age data 132 measured by another measuring device from the other measuring device (S81).

[0127] Then, the risk estimation unit 116 inputs, for example, the new electrocardiogram information 131 and the new age data 132 acquired in the process of S81 into the learning model MD1a (S82).

[0128] Next, the risk estimation unit 116 acquires, for example, the EHFI data 134a output from the learning model MD1a in response to the input of the new electrocardiogram information 131 and the new age data 132 in the processing of S82 (S83).

[0129] Thereafter, the result output unit 117 outputs, for example, the EHFI data 134a acquired in the process of S83 to the output device 105 (S84).

[0130] [Specific example of the configuration of learning model MD1a] Next, a specific example of the configuration of the learning model MD1a in the first modified example will be described. Specifically, Fig. 23 is a diagram illustrating a specific example of the configuration of the learning model MD1a in the first modified example. In the example shown in Fig. 23, three seconds of electrocardiogram waveform data in a 12-lead electrocardiogram is input to the learning model MD1a as electrocardiogram waveform data included in the electrocardiogram information 131. In addition, in the example shown in Fig. 23, three seconds of electrocardiogram waveform data in a 12-lead electrocardiogram is expressed as "12-lead ECG 3 Second."

[0131] The following description will be given of the case where 3 seconds of electrocardiogram waveform data from a 12-lead electrocardiogram is input, but the learning model MD1a may also be configured to input, for example, electrocardiogram waveform data of a duration shorter than 3 seconds (e.g., electrocardiogram waveform data of approximately 1.5 seconds), or electrocardiogram waveform data of a duration longer than 3 seconds.

[0132] Specifically, in the learning model MD1a of this modification, for example, electrocardiogram waveform data included in electrocardiogram information 131 is input to five layers of convolutional blocks, a drop layer, and a fully connected layer in this order, as shown in Fig. 23. Also, in the learning model MD1a of this modification, for example, the potential difference and age data 132 included in the electrocardiogram information 131 are input to the fully connected layer, as shown in Fig. 23. Then, in the learning model MD1a of this modification, for example, re-learning (fine-tuning) is performed for each of the five layers of convolutional blocks and the fully connected layer.

[0133] Note that the learning model MD1a in this modification may not require input of the heart rate included in the electrocardiogram information 131, as shown in Fig. 23. Also, the learning model MD1a in this modification may not have a Squeeze-and-Excitation layer, as shown in Fig. 23.

[0134] As a result, the learning model MD1a in this modification can shorten the estimation time by realizing simplification (weight reduction) of the learning model MD1a itself while maintaining at least the estimation accuracy of the EHFI data 134a corresponding to each subject OB, compared to, for example, the case where the learning models MD1 and MD2 in the first embodiment are used. Specifically, the learning model MD1a in this modification can achieve simplification (weight reduction) of the learning model MD1a itself by reducing the number of times that the potential difference and age data 132 included in the electrocardiogram information 131 are input, compared to, for example, the case where the learning models MD1 and MD2 in the first embodiment are used.

[0135] Therefore, the learning model MD1a in this modified example not only makes it possible to reduce time and workload by not generating the learning model MD2 in the first embodiment, but also makes it possible to shorten the estimation time by simplifying (reducing the weight of) the learning model MD1a itself while at least maintaining the estimation accuracy of the EHFI data 134a corresponding to each subject OB.

[0136] Furthermore, in this modified example, when heart rate is not input, the learning model MD1a can reduce the impact of fluctuations in the heart rate of each subject OB on the estimation accuracy of the EHFI data 134a, thereby making it possible to further improve the estimation accuracy of the EHFI data 134a corresponding to each subject OB.

[0137] [Effect of the learning process in the first modified example] Fig. 24 is a diagram illustrating the effect of the learning process in the first modified example. Specifically, Fig. 24 is a graph showing the prediction accuracy of each of the prediction of major cardiovascular events using BNP data 133 (i.e., prediction by the conventional method) and the prediction of major cardiovascular events using EHFI data 134a output from the learning model MD1a in this modified example (i.e., prediction by the method in the first modified example).

[0138] Specifically, Fig. 24 shows that the area under the ROC curve (AUC) is 0.871, for example. That is, Fig. 24 can be judged to indicate that the prediction accuracy of the learning model MD1a in this modification is sufficiently high.

[0139] In this way, the information processing device 1 in this embodiment acquires, for example, first electrocardiogram information 131 of the first subject OB and first BNP data 133 of the first subject OB. Then, the information processing device 1 calculates corrected BNP data 134 by using, for example, the first BNP data 133 and a coefficient corresponding to the first subject OB. Furthermore, the information processing device 1 generates, for example, teacher data DT1a (hereinafter also referred to as first teacher data DT1a) including the first electrocardiogram information 131 and the corrected BNP data 134. Thereafter, the information processing device 1 generates a first learning model MD1a by, for example, learning the first teacher data DT1a.

[0140] As a result, the information processing device 1 in this modified example not only makes it possible to reduce time and workload by not generating the learning model MD2 in the first embodiment, but also makes it possible to shorten the estimation time by simplifying (reducing the weight of) the learning model MD1a itself while at least maintaining the estimation accuracy of the EHFI data 134a corresponding to each subject OB. [Explanation of symbols]

[0141] 1: Information processing device 2: Measuring equipment 101:CPU 102: Memory 103: Communication equipment 104:Storage medium 105: Output device 106: Bus 111: Data acquisition section 112: First data generation unit 113: First model generation unit 114: First data generation unit 115: Second model generation unit 116: Risk Estimation Department 117: Result output section 131:Electrocardiogram information 132: Age data 133:BNP data 133a:BNP data 134: Corrected BNP data 134a: Corrected BNP data DT1: Training data DT1a: Training data DT2: Training data MD1: Learning model MD1a: Learning model MD2: Learning Model

Claims

1. obtaining first electrocardiogram information of a first person and a first BNP value of the first person; generating first teacher data including the first electrocardiogram information and the first BNP value; generating a first learning model by learning the first teacher data; acquiring second electrocardiogram information of a second person and a second BNP value of the second person; calculating a corrected BNP value by using the second BNP value and a coefficient corresponding to the second person; generating second teacher data including the second electrocardiogram information and the corrected BNP value; generating a second learning model including at least a part of the layers constituting the first learning model by learning the second training data; A learning program that causes a computer to execute a process.

2. the first electrocardiogram information includes at least one of first electrocardiogram waveform data of the first person, a potential difference in the first electrocardiogram waveform data, and a heart rate of the first person; The learning program of claim 1, characterized in that the second electrocardiogram information includes at least one of second electrocardiogram waveform data of the second person, a potential difference in the second electrocardiogram waveform data, and a heart rate of the second person.

3. The learning program described in claim 1, characterized in that the second learning model is a learning model that includes a predetermined number of layers from the top of the multiple layers that constitute the first learning model and other layers that receive output from the predetermined number of layers as input.

4. 4. The learning program according to claim 3, wherein the process of generating the second learning model involves learning parameters in the other layers.

5. In the process of generating the first learning model, a plurality of first learning models are generated; 4. The learning program according to claim 3, wherein the other layers receive outputs from the predetermined number of layers in each of the plurality of first learning models as inputs.

6. In the process of acquiring the first electrocardiogram information and the first BNP value, a first age of the first person is acquired; In the process of generating the first teacher data, the first teacher data including the first age is generated; In the process of generating the first learning model, the first learning model is generated by learning the first teacher data including the first age; In the process of acquiring the second electrocardiogram information and the second BNP value, a second age of the second person is acquired; In the process of generating the second teacher data, the second teacher data including the second age is generated; In the process of generating the second learning model, the second learning model is generated by learning the second training data including the second age.

2. The learning program according to claim 1, wherein the learning program causes a computer to execute processing.

7. In the process of generating the first teacher data, transforming said first BNP value by using an arctangent; generating the first teacher data including the first electrocardiogram information and the converted first BNP value; 2. The learning program according to claim 1.

8. further specifying a coefficient corresponding to the second person according to an occurrence status of a major cardiovascular event in the second person; 2. The learning program according to claim 1, wherein the learning program causes a computer to execute processing.

9. 9. The learning program according to claim 8, wherein the process of identifying the coefficient identifies a first coefficient as the coefficient if the second person develops the major cardiovascular event within a predetermined period from the time the second electrocardiogram information is measured, and identifies a second coefficient smaller than the first coefficient as the coefficient if the second person does not develop the major cardiovascular event within the predetermined period.

10. 9. The learning program of claim 8, wherein the process of identifying the coefficient identifies a first coefficient as the coefficient if the second person develops the major cardiovascular event within a predetermined period from the time the second electrocardiogram information is measured, and identifies a third coefficient greater than the first coefficient as the coefficient if the second person develops the major cardiovascular event within a period shorter than the predetermined period.

11. 9. The learning program of claim 8, wherein the process of identifying the coefficient identifies a first coefficient as the coefficient if the second person experiences the major cardiovascular event a predetermined number of times within a predetermined period from the time the second electrocardiogram information is measured, and identifies a fourth coefficient greater than the first coefficient as the coefficient if the first person experiences the major cardiovascular event more than a predetermined number of times within the predetermined period.

12. obtaining first electrocardiogram information of a first person and a first BNP value of the first person; calculating a corrected BNP value by using the first BNP value and a coefficient corresponding to the first person; generating first teacher data including the first electrocardiogram information and the corrected BNP value; generating a first learning model by learning the first training data; A learning program that causes a computer to execute a process.

13. a data acquisition unit that acquires first electrocardiogram information of a first person and a first BNP value of the first person; a first data generating unit that generates first teacher data including the first electrocardiogram information and the first BNP value; a first model generation unit that generates a first learning model by learning the first teacher data, The data acquisition unit acquires second electrocardiogram information of a second person and a second BNP value of the second person, and further a second data generating unit that calculates a corrected BNP value by using the second BNP value and a coefficient corresponding to the second person, and generates second teacher data including the second electrocardiogram information and the corrected BNP value; a second model generation unit that generates a second learning model including at least a part of the layers that constitute the first learning model by learning the second teacher data; 1. An information processing device comprising:

14. a data acquisition unit that acquires first electrocardiogram information of a first person and a first BNP value of the first person; a first data generating unit that calculates a corrected BNP value by using the first BNP value and a coefficient corresponding to the first person, and generates first teacher data including the first electrocardiogram information and the corrected BNP value; a first model generation unit that generates a first learning model by learning the first teacher data; 1. An information processing device comprising:

15. obtaining first electrocardiogram information of a first person and a first BNP value of the first person; generating first teacher data including the first electrocardiogram information and the first BNP value; generating a first learning model by learning the first teacher data; acquiring second electrocardiogram information of a second person and a second BNP value of the second person; calculating a corrected BNP value by using the second BNP value and a coefficient corresponding to the second person; generating second teacher data including the second electrocardiogram information and the corrected BNP value; generating a second learning model including at least a part of the layers constituting the first learning model by learning the second training data; A learning method characterized in that the processing is executed by a computer.

16. obtaining first electrocardiogram information of a first person and a first BNP value of the first person; calculating a corrected BNP value by using the first BNP value and a coefficient corresponding to the first person; generating first teacher data including the first electrocardiogram information and the corrected BNP value; generating a first learning model by learning the first training data; A learning method characterized in that the processing is executed by a computer.

Images