Blood pressure estimation device, blood pressure estimation method, and program

Abstract

This blood pressure estimation device (100) is provided with: an acquisition unit (10) for acquiring biological information about a subject; a blood pressure estimation model inference unit (20) for inputting the acquired biological information about the subject to a learned model generated by machine learning to make an inference of a blood pressure value estimation model for the subject; and a blood pressure estimation unit (30) for estimating a blood pressure value of the subject on the basis of the acquired biological information about the subject and the inferred blood pressure value estimation model for the subject.

Description

Blood pressure estimation device, blood pressure estimation method, and program

[0001] The present disclosure relates to a blood pressure estimation device, a blood pressure estimation method, and a program.

[0002] Patent Documents 1 to 3 disclose techniques for estimating a blood pressure value by substituting biological information such as a pulse wave and an electrocardiogram into a prepared equation for estimating a blood pressure value.

[0003] JP-A-7-289526 Patent No. 4971041 Patent No. 5984088

[0004] However, because there are subtle differences in the correspondence between blood pressure values ​​and biological information such as pulse waves and electrocardiograms, the accuracy of estimation formulas for estimating blood pressure values ​​using biological information varies from person to person. In other words, it is difficult to accurately estimate the blood pressure values ​​of a large number of people using a single estimation formula.

[0005] Therefore, the present disclosure provides a blood pressure estimation device and the like that can accurately estimate blood pressure values ​​for each individual.

[0006] The blood pressure estimation device according to the present disclosure includes an acquisition unit that acquires biometric information of a subject, a blood pressure estimation model inference unit that infers an estimation model for the subject's blood pressure value by inputting the acquired biometric information of the subject into a trained model generated by machine learning, and a blood pressure estimation unit that estimates the subject's blood pressure value based on the acquired biometric information of the subject and the inferred estimation model for the subject's blood pressure value.

[0007] The blood pressure estimation method according to the present disclosure is a blood pressure estimation method executed by a computer, and includes an acquisition step of acquiring biometric information of a subject, a blood pressure estimation model inference step of inferring an estimation model of the subject's blood pressure value by inputting the acquired biometric information of the subject into a trained model generated by machine learning, and a blood pressure estimation step of estimating the subject's blood pressure value based on the acquired biometric information of the subject and the inferred estimation model of the subject's blood pressure value.

[0008] A program according to the present disclosure is a program that causes a computer to execute the above-described blood pressure estimation method.

[0009] These comprehensive or specific aspects may be realized as a system, a method, an integrated circuit, a computer program, or a computer-readable recording medium such as a CD-ROM, or may be realized as any combination of a system, a method, an integrated circuit, a computer program, and a recording medium.

[0010] According to the blood pressure estimation device according to one aspect of the present disclosure, it is possible to estimate the blood pressure value for each individual with high accuracy.

[0011] 1 is a block diagram showing an example of a blood pressure estimation device according to embodiment 1. FIG. 2 is a diagram for explaining a specific example of a feature. FIG. 3 is a flowchart showing an example of an operation during blood pressure estimation of the blood pressure estimation device according to embodiment 1. FIG. 4 is a flowchart showing an example of an operation during learning of the blood pressure estimation device according to embodiment 1. FIG. 5 is a diagram for explaining a learning unit according to embodiment 1. FIG. 6 is a block diagram showing an example of a blood pressure estimation device according to a modified example of embodiment 1. FIG. 7 is a block diagram showing an example of a blood pressure estimation device according to embodiment 2. FIG. 8 is a diagram for explaining an example of an operation during learning of the blood pressure estimation device according to embodiment 2. FIG. 9 is a diagram for explaining an example of an operation during additional learning of the blood pressure estimation device according to embodiment 2. FIG. 10 is a diagram for explaining an example of an operation during blood pressure estimation of the blood pressure estimation device according to embodiment 2. FIG. 11 is a diagram for explaining another example of an operation during blood pressure estimation of the blood pressure estimation device according to embodiment 2. FIG. 12 is a diagram for explaining an example of an operation during learning of the blood pressure estimation device according to embodiment 3.

[0012] (Background to the Invention of One Aspect of the Present Disclosure) Conventionally, blood pressure values ​​have been measured using a cuff. However, although the cuff method is highly accurate, it places a heavy burden on the subject because it compresses the subject's arm. Furthermore, the cuff method is not suitable for continuous blood pressure measurement.

[0013] On the other hand, there are also methods for estimating blood pressure values ​​without using a cuff. Specifically, these methods estimate blood pressure values ​​using biological information such as electrocardiograms and pulse waves. More specifically, they estimate blood pressure values ​​from the pulse transmission time (PTT). PTT is the delay time of a pulse wave signal relative to an electrocardiogram signal. It is known that the PTT tends to shorten as blood pressure increases and lengthen as blood pressure decreases. Since blood pressure and PTT are correlated, it is possible to estimate blood pressure values ​​using electrocardiograms and pulse waves. For example, Patent Documents 1 to 3 disclose various estimation formulas for estimating blood pressure values ​​using biological information such as pulse waves and electrocardiograms.

[0014] However, there are subtle differences between individuals in the correspondence between biological information such as pulse waves and electrocardiograms and blood pressure values, making it difficult to accurately estimate the blood pressure values ​​of a large number of people using a single estimation formula. It is also conceivable to prepare multiple estimation formulas in advance and select one from among the multiple estimation formulas depending on the subject, but the selected estimation formula may not be optimal because it is not the estimation formula prepared for the subject, making it difficult to accurately estimate the subject's blood pressure value.

[0015] Therefore, a blood pressure estimation device that can accurately estimate the blood pressure value for each individual will be described below.

[0016] A blood pressure estimation device according to one aspect of the present disclosure includes an acquisition unit that acquires biometric information of a subject, a blood pressure estimation model inference unit that infers an estimation model of the subject's blood pressure value by inputting the acquired biometric information of the subject into a trained model generated by machine learning, and a blood pressure estimation unit that estimates the subject's blood pressure value based on the acquired biometric information of the subject and the inferred estimation model of the subject's blood pressure value.

[0017] According to this method, an estimation model of a subject's blood pressure value is inferred using a trained model generated by machine learning. This estimation model is a model for the subject inferred from the subject's biometric information and is an optimal model for the subject. Therefore, the inferred estimation model can be used to accurately estimate the blood pressure value for each individual. Furthermore, because a cuff is not used, the blood pressure estimation device can be made smaller, and for example, a blood pressure measurement function can be installed in a wearable device. Alternatively, the portability of the blood pressure estimation device can be improved. Furthermore, because a cuff is not used to estimate (measure) blood pressure, the portability of the blood pressure estimation device is improved, making it possible to measure blood pressure continuously or at all times. Therefore, the subject's condition can be understood in detail from instantaneous changes in blood pressure, which was previously difficult to determine.

[0018] For example, the blood pressure estimation device may further include a learning unit that generates the trained model by performing machine learning based on the biological information of the plurality of people.

[0019] In this way, a trained model can be generated using biometric information from multiple people.

[0020] For example, the learning unit may generate the trained model by performing machine learning based on the blood pressure values ​​and biometric information of each of the multiple people.

[0021] Because blood pressure values ​​and biometric information are used when generating the trained model, it is possible to perform additional learning on the generated trained model using the subject's biometric information and blood pressure values.

[0022] For example, the learning unit may generate the trained model by performing machine learning using the biometric information of each of the plurality of people as input data and the blood pressure values ​​of each of the plurality of people as training data.

[0023] When generating a trained model, machine learning is performed using biometric information as input data and blood pressure values ​​as training data. Therefore, it is possible to perform additional learning on the generated trained model using the subject's biometric information as input data and the subject's blood pressure values ​​as training data.

[0024] For example, the learning unit may generate an estimation model of the blood pressure values ​​of each of the plurality of people based on the blood pressure values ​​and biometric information of each of the plurality of people, and generate the trained model by performing machine learning using the biometric information of each of the plurality of people as input data and the generated estimation model of the blood pressure values ​​of each of the plurality of people and the blood pressure values ​​of each of the plurality of people as training data.

[0025] This allows for the generation of an estimation model for each of the blood pressure values ​​of multiple people based on the relationship between each of the people's blood pressure values ​​and their biometric information. Then, machine learning is performed using each of the people's biometric information as input data and each of the people's estimation models and blood pressure values ​​as training data, thereby generating a trained model that uses the biometric information as input and the estimation model as output. In this way, the blood pressure values ​​of each of the multiple people are also used as training data, and machine learning is performed in advance to minimize the error between the estimated blood pressure values ​​of each of the multiple people estimated by the estimation model and the blood pressure values ​​used as training data, making it possible to perform additional learning using the subjects' biometric information and blood pressure values.

[0026] For example, the learning unit may generate the trained model for the subject by performing additional machine learning using the acquired biometric information of the subject as input data and the subject's blood pressure value as training data.

[0027] Because the trained model is a model trained using biometric information of people different from the subject, the estimation model output from the trained model may not be able to accurately estimate blood pressure values ​​for some subjects. In response to this, by performing additional machine learning using the subject's biometric information as input data and the subject's blood pressure values ​​as training data, it is possible to generate a trained model for the subject that outputs an estimation model that can accurately estimate the subject's blood pressure values.

[0028] For example, the blood pressure estimation model inference unit may infer multiple estimation models of the subject's blood pressure value by inputting multiple pieces of biometric information of the subject used when additional machine learning was performed into the trained model for the subject, and generate a single estimation model for the subject from the multiple inferred estimation models of the subject's blood pressure value, and the blood pressure estimation unit may estimate the subject's blood pressure value based on the acquired biometric information of the subject and the estimation model for the subject.

[0029] This makes it possible to generate an estimation model for one subject by averaging multiple estimation models inferred using multiple pieces of biometric information of the subject used when additional machine learning was performed. For example, it is no longer necessary to infer an estimation model every time the subject's blood pressure value is estimated, and after an estimation model for a subject is generated, the subject's blood pressure value can be estimated using the estimation model for the subject.

[0030] For example, the learning unit may generate an estimation model of the blood pressure values ​​of each of the plurality of people based on the blood pressure values ​​and biometric information of each of the plurality of people, and generate the trained model by performing machine learning based on the biometric information of each of the plurality of people and the generated estimation model of the blood pressure values ​​of each of the plurality of people.

[0031] This makes it possible to generate an estimation model for the blood pressure values ​​of each of the multiple people from the relationship between the blood pressure values ​​and biometric information of each of the multiple people. Then, by performing machine learning using the biometric information of each of the multiple people as input data and the estimation models of each of the multiple people as training data, it is possible to generate a trained model that uses the biometric information as input and the estimation model as output.

[0032] For example, the learning unit may generate an estimation model of the blood pressure values ​​of each of the plurality of people by regression analysis.

[0033] This makes it possible to generate an estimation model for the blood pressure values ​​of each of a plurality of people by regression analysis based on the relationship between the blood pressure values ​​and biological information of each of a plurality of people.

[0034] For example, the blood pressure estimation model inference unit may infer a polynomial using the acquired biological information of the subject as a variable as an estimation model for the blood pressure value of the subject.

[0035] In this way, the estimation model for the subject's blood pressure value may be a polynomial.

[0036] For example, the blood pressure estimation model inference unit may infer weighting coefficients of a neural network that receives acquired biological information of the subject as input as an estimation model for the subject's blood pressure value.

[0037] In this way, the estimation model for the subject's blood pressure value may be a neural network.

[0038] For example, the biological information may be biological information related to an electrocardiogram and a pulse wave.

[0039] According to this, since the electrocardiogram and pulse wave are correlated with blood pressure, the blood pressure value can be estimated based on biological information relating to the electrocardiogram and pulse wave.

[0040] For example, the blood pressure estimation model inference unit may infer an estimation model of the subject's blood pressure value by inputting the electrocardiogram waveform and pulse wave waveform, as well as information for synchronizing the time between the electrocardiogram waveform and the pulse wave waveform, as the acquired biometric information of the subject into the trained model, and the blood pressure estimation unit may estimate the subject's blood pressure value based on feature quantities related to the electrocardiogram and pulse wave as the acquired biometric information of the subject and the inferred estimation model of the subject's blood pressure value.

[0041] This allows an optimal estimation model for a subject to be inferred from the subject's electrocardiogram waveform and pulse wave waveform. Furthermore, when estimating the subject's blood pressure using the inferred estimation model, the electrocardiogram waveform and pulse wave waveform do not need to be used, and the subject's blood pressure can be easily estimated by using feature quantities related to the electrocardiogram and pulse wave.

[0042] For example, the information for synchronizing the time between the electrocardiogram waveform and the pulse waveform may include the pulse wave transit time (PTTv) from the R wave of the electrocardiogram waveform to the trough of the pulse waveform, and the time from the peak of one pulse waveform to the peak of the next pulse waveform (PPI: Peak to Peak Interval).

[0043] According to this, by using PTTv and PPI, it is possible to align the time of the subject's electrocardiogram waveform and pulse waveform, thereby making it possible to infer a more optimal estimation model for the subject.

[0044] For example, the blood pressure estimation unit may estimate the subject's blood pressure value by multiplying the acquired biometric information of the subject by a coefficient included in an estimation model of the subject's inferred blood pressure value.

[0045] In this way, blood pressure values ​​can be estimated using a simple method.

[0046] A blood pressure estimation method according to one aspect of the present disclosure is a method executed by a computer, and includes: an acquisition step of acquiring biometric information of a subject; a blood pressure estimation model inference step of inferring an estimation model of the subject's blood pressure value by inputting the acquired biometric information of the subject into a trained model generated by machine learning; and a blood pressure estimation step of estimating the subject's blood pressure value based on the acquired biometric information of the subject and the inferred estimation model of the subject's blood pressure value.

[0047] This provides a blood pressure estimation method that can accurately estimate the blood pressure value for each individual.

[0048] A program according to one aspect of the present disclosure is a program that causes a computer to execute the blood pressure estimation method.

[0049] This makes it possible to provide a program that can accurately estimate the blood pressure value for each individual.

[0050] Hereinafter, the embodiments will be specifically described with reference to the drawings.

[0051] The embodiments described below are all comprehensive or specific examples, and the numerical values, shapes, materials, components, arrangement and connection of the components, steps, and order of steps shown in the following embodiments are merely examples and are not intended to limit the present disclosure.

[0052] (Embodiment 1) Hereinafter, a blood pressure estimation device according to embodiment 1 will be described.

[0053] FIG. 1 is a block diagram illustrating an example of a blood pressure estimation device 100 according to the first embodiment.

[0054] The blood pressure estimation device 100 is a device that estimates a person's blood pressure value. Hereinafter, the person whose blood pressure value is estimated will also be referred to as a subject. The blood pressure estimation device 100 can estimate blood pressure values ​​using a method that does not use a cuff. Therefore, the blood pressure estimation device 100 can be realized as, for example, a wearable device.

[0055] The blood pressure estimation device 100 includes an estimation unit 50 and a measurement unit 300 .

[0056] The measuring unit 300 measures a person's electrocardiogram and pulse wave. The measuring unit 300 includes an electrode 301, an electrocardiogram signal acquiring unit 302, a light emitting unit 303, a light receiving unit 304, a pulse wave signal acquiring unit 305, and a processing unit 306.

[0057] The electrocardiographic signal acquisition unit 302 acquires electrocardiographic signals via electrodes 301 (specifically, two electrodes) that are in contact with the human body. The electrocardiographic signal acquisition unit 302 has an amplifier circuit, a filter circuit, an AD conversion circuit, etc., and amplifies weak electrocardiographic signals, removes noise, and converts them into digital values.

[0058] Pulse wave signal acquiring unit 305 acquires a pulse wave signal based on the light emitted from light emitting unit 303 and reflected by the human body and received by light receiving unit 304. For example, light receiving unit 304 converts the amount of reflected light received into a voltage value and outputs the voltage value as a pulse wave signal to pulse wave signal acquiring unit 305. Pulse wave signal acquiring unit 305 includes an amplifier circuit, a filter circuit, an AD conversion circuit, etc., and amplifies the weak pulse wave signal, removes noise, and converts it into a digital value. Note that transmitted light that has passed through the human body may be used instead of reflected light that has reflected from the human body.

[0059] Processing unit 306 extracts biological information from the electrocardiogram signal and pulse wave signal acquired by electrocardiogram signal acquisition unit 302 and pulse wave signal acquisition unit 305. For example, the biological information is biological information related to the electrocardiogram and pulse wave, and more specifically, the electrocardiogram waveform, pulse wave waveform, and various feature quantities.

[0060] For example, the processing unit 306 extracts an electrocardiogram waveform and a pulse waveform for one beat. Specifically, the processing unit 306 detects R waves (see FIG. 2) from the electrocardiogram waveform and extracts an electrocardiogram waveform for one beat based on the interval between the detected R waves. The processing unit 306 also detects pulse wave troughs (see FIG. 2) from the pulse waveform and extracts a pulse waveform for one beat based on the interval between the detected troughs.

[0061] For example, 25 types of feature amounts are extracted in the processing unit 306. The 25 types of feature amounts will now be described with reference to FIG.

[0062] Fig. 2 is a diagram illustrating specific examples of feature quantities, in which (a) in Fig. 2 is a diagram illustrating an electrocardiogram signal, (b) in Fig. 2 is a diagram illustrating a pulse wave signal, (c) in Fig. 2 is a diagram illustrating a pulse waveform for one beat, (d) in Fig. 2 is a first-order differential pulse waveform of the pulse waveform shown in (c) in Fig. 2, and (e) in Fig. 2 is a second-order differential pulse waveform of the pulse waveform shown in (c) in Fig. 2.

[0063] In FIG. 2, there are 25 types of feature quantities, １ ~x ２５ is shown.

[0064] Feature x １is the pulse wave transit time (PTTp) from the R wave (the time point of cardiac contraction) of the electrocardiogram to the peak of the pulse wave waveform.

[0065] Feature x ２ is the pulse wave transit time (PTTv) from the R wave of the electrocardiogram waveform to the trough of the pulse waveform.

[0066] Feature x ３ is the time from one peak of the pulse waveform to the next peak of the pulse waveform (PPI).

[0067] Feature x ４ is the time when the value becomes 0 between the e wave (the last peak of the second derivative pulse waveform) and the f wave (the last valley of the second derivative pulse waveform).

[0068] Feature x ５ is the time x of the first derivative pulse waveform ４ is the value.

[0069] Feature x ６ is the time x of the pulse waveform ４ is the value.

[0070] Feature x ７ is the time of the a-wave of the second derivative pulse waveform (the first peak of the second derivative pulse waveform).

[0071] Feature x ８ is the time x of the second derivative pulse waveform ７ is the value.

[0072] Feature x ９ is the time of the b wave of the second derivative pulse waveform (the first valley of the second derivative pulse waveform).

[0073] Feature x １０ is the time x of the second derivative pulse waveform ９ is the value.

[0074] Feature x １１ is the time of the c-wave of the second derivative pulse waveform (the second peak of the second derivative pulse waveform).

[0075] Feature x １２ is the time x of the second derivative pulse waveform １１ is the value.

[0076] Feature x １３is the time of the d wave of the second derivative pulse waveform (the second valley of the second derivative pulse waveform).

[0077] Feature x １４ is the time x of the second derivative pulse waveform １３ is the value.

[0078] Feature x １５ is the time of the e wave of the second derivative pulse waveform.

[0079] Feature x １６ is the time x of the second derivative pulse waveform １５ is the value.

[0080] Feature x １７ is the time of the f wave of the second derivative pulse waveform.

[0081] Feature x １８ is the time x of the second derivative pulse waveform １７ is the value.

[0082] Feature x １９ is the time x of the first derivative pulse waveform ７ is the value.

[0083] Feature x ２０ is the time x of the first derivative pulse waveform ９ is the value.

[0084] Feature x ２１ is the time of the peak of the first derivative pulse waveform.

[0085] Feature x ２２ is the time x of the first derivative pulse waveform ２１ is the value.

[0086] Feature x ２３ is the time of the valley of the first derivative pulse waveform.

[0087] Feature x ２４ is the time x of the first derivative pulse waveform ２３ is the value.

[0088] Feature x ２５ is the time x of the line connecting the peak and the end of the pulse waveform. ４ is the value.

[0089] Alternatively, electrocardiogram waveforms and pulse waveforms for several beats may be extracted, and the average of the electrocardiogram waveforms and pulse waveforms for several beats and the average of each feature in the electrocardiogram waveforms and pulse waveforms for several beats may be extracted.

[0090] The estimation unit 50 estimates the blood pressure value of the subject. The estimation unit 50 includes an acquisition unit 10, a blood pressure estimation model inference unit 20, a blood pressure estimation unit 30, and an output unit 40. The blood pressure estimation device 100 (estimation unit 50) is a computer including a processor, a memory, etc. The memory is a read-only memory (ROM) or a random access memory (RAM), etc., and can store programs executed by the processor. The acquisition unit 10, the blood pressure estimation model inference unit 20, the blood pressure estimation unit 30, and the output unit 40 are realized by a processor, etc., that executes programs stored in the memory.

[0091] The acquisition unit 10 acquires biological information of the subject. Specifically, the acquisition unit 10 acquires the subject's electrocardiogram waveform, pulse waveform, and various feature quantities from the measurement unit 300.

[0092] The blood pressure estimation model inference unit 20 infers an estimation model of the subject's blood pressure value by inputting the acquired biometric information of the subject into a trained model generated by machine learning. For example, the blood pressure estimation model inference unit 20 infers an estimation model of the subject's blood pressure value by inputting the acquired biometric information of the subject, such as an electrocardiogram waveform and a pulse waveform, and information for time-aligning the electrocardiogram waveform and the pulse waveform, into the trained model. The information for time-aligning the electrocardiogram waveform and the pulse waveform includes, for example, the pulse wave transit time (PTTv) from the R wave of the electrocardiogram waveform to the trough of the pulse waveform, and the time from the peak of the pulse waveform to the peak of the next pulse waveform (PPI). For example, the blood pressure estimation model inference unit 20 infers an estimation model of the systolic blood pressure value and an estimation model of the diastolic blood pressure value as estimation models of the subject's blood pressure value. Note that the blood pressure estimation model inference unit 20 may include a learning unit 200. The learning unit 200 generates the trained model by performing machine learning based on the biometric information of multiple people.

[0093] The blood pressure estimation unit 30 estimates the blood pressure value of the subject based on the acquired biometric information of the subject and an estimation model of the subject's inferred blood pressure value. For example, the blood pressure estimation unit 30 estimates the blood pressure value of the subject based on feature quantities related to the electrocardiogram and pulse wave as the acquired biometric information of the subject and an estimation model of the subject's inferred blood pressure value. The feature quantities related to the electrocardiogram and pulse wave are feature quantities extracted from the electrocardiogram waveform and pulse wave waveform, and more specifically, the above-mentioned feature quantity x １ ~x ２５ However, the feature quantity used to estimate the blood pressure value of the subject is not limited to this, and the feature quantity x １ ~x ２５ It may be a part of the above or a different feature amount.

[0094] The output unit 40 outputs the estimated blood pressure values. For example, the output unit 40 outputs the estimated systolic blood pressure value and diastolic blood pressure value. For example, the blood pressure estimation device 100 may be provided with a display unit (such as a display) or an audio output unit (speaker), and the estimated blood pressure values ​​may be displayed or output as audio.

[0095] The measurement unit 300 does not have to be a component of the blood pressure estimation device 100. For example, the blood pressure estimation device 100 (estimation unit 50) may acquire an electrocardiogram waveform, a pulse waveform, and various feature quantities from a measurement unit 300 provided separately from the blood pressure estimation device 100.

[0096] Next, the operation of the blood pressure estimation device 100 will be described in detail with reference to FIGS. 3 to 5. FIG.

[0097] FIG. 3 is a flowchart showing an example of the operation of the blood pressure estimation device 100 according to the first embodiment during blood pressure estimation.

[0098] First, the acquisition unit 10 acquires biological information of the subject (step S11). As described above, for example, the acquisition unit 10 acquires the subject's electrocardiogram waveform, pulse waveform, and various feature quantities (for example, feature quantity x １ ~x ２５ ) to obtain the

[0099] Next, the blood pressure estimation model inference unit 20 infers an estimation model of the subject's blood pressure value by inputting the acquired biological information of the subject into a trained model generated by machine learning (step S12). For example, the blood pressure estimation model inference unit 20 infers an estimation model of the subject's blood pressure value by inputting the acquired biological information of the subject, the electrocardiogram waveform, the pulse waveform, the PTTv (feature quantity x ２ ) and PPI (feature x ３ ) into the trained model to infer an estimated model for the subject's systolic blood pressure value and diastolic blood pressure value.

[0100] For example, the blood pressure estimation model inference unit 20 infers a polynomial with the acquired biological information of the subject as a variable as an estimation model of the subject's blood pressure value (an estimation model of the systolic blood pressure value and an estimation model of the diastolic blood pressure value). For example, the variable is a feature quantity related to the electrocardiogram and pulse wave, specifically, a feature quantity extracted from the electrocardiogram waveform and the pulse wave waveform, more specifically, the above-mentioned feature quantity x １ ~x ２５ However, this variable is １ ~x ２５ Not limited to, feature x １ ~x ２５ It may be a part of the above or a different feature amount.

[0101] For example, the estimation model for the systolic blood pressure value and the estimation model for the diastolic blood pressure value can be expressed by polynomials such as the following Equation 1 and Equation 2. ｉ and b ｉ are coefficients, c and d are constants, and x ｉ is a feature.

[0102]

[0103]

[0104] Next, the blood pressure estimation unit 30 estimates the subject's blood pressure value based on the acquired biometric information of the subject and the estimated model of the subject's blood pressure value (step S13). For example, the blood pressure estimation unit 30 estimates the subject's blood pressure value by multiplying the acquired biometric information of the subject by a coefficient included in the estimated model of the subject's blood pressure value. For example, the blood pressure estimation unit 30 estimates the subject's blood pressure value by multiplying the acquired biometric information of the subject by a coefficient included in the estimated model of the subject's blood pressure value.１ ~x ２５ and the coefficient a in the above formulas 1 and 2. ｉ and b ｉ By multiplying by and, the subject's systolic and diastolic blood pressure values ​​are estimated.

[0105] The output unit 40 then outputs the estimated blood pressure values ​​of the subject (step S14). For example, the output unit 40 outputs the estimated systolic and diastolic blood pressure values ​​of the subject to a display or the like for display.

[0106] Next, the operation of the blood pressure estimation device 100 during learning of the learned model will be described with reference to FIGS. 4 and 5. FIG.

[0107] FIG. 4 is a flowchart showing an example of the operation during learning of the blood pressure estimation device 100 according to the first embodiment.

[0108] Fig. 5 is a diagram for explaining the learning unit 200 according to embodiment 1. The upper part of Fig. 5 shows a blood pressure estimation model generation unit 201 and a blood pressure estimation model inferor learning unit 202 as functional components used for learning by the learning unit 200. The lower part of Fig. 5 also shows a schematic diagram of the inference flow of an estimation model using the generated trained model.

[0109] First, the learning unit 200 acquires blood pressure values ​​and biometric information of each of the plurality of people (step S21). For example, the learning unit 200 acquires systolic and diastolic blood pressure values ​​as blood pressure values ​​of each of the plurality of people. The systolic and diastolic blood pressure values ​​are, for example, blood pressure values ​​measured by a method using a cuff. The learning unit 200 also acquires electrocardiogram waveforms, pulse waveforms, and feature quantities related to the electrocardiogram and pulse wave as biometric information of each of the plurality of people. For example, the feature quantities related to the electrocardiogram and pulse wave are feature quantities extracted from the electrocardiogram waveform and pulse waveform, and specifically, the above-mentioned feature quantity x １ ~x ２５ For the sake of simplicity, persons A, B, and C are shown in FIG. 5 as an example of a plurality of persons, but for example, blood pressure values ​​and biological information of many persons may be acquired and used for learning.

[0110] Next, the learning unit 200 (blood pressure estimation model generating unit 201) generates an estimation model of the blood pressure values ​​of each of the plurality of people based on the blood pressure values ​​and biometric information of each of the plurality of people (step S22). For example, the learning unit 200 (blood pressure estimation model generating unit 201) generates an estimation model of the blood pressure values ​​of each of the plurality of people by regression analysis. For example, the blood pressure values ​​measured by a cuff method are used as training data, and the biometric information (feature values) are used as input data, and a blood pressure estimation model for each individual is generated by regression analysis. For example, for person A, the systolic blood pressure value of person A is used as training data, and the feature value x １ ~x ２５ Using the above as input data, a model for estimating the systolic blood pressure of person A is generated by multiple regression analysis. Also, using the diastolic blood pressure of person A as training data, the feature quantity x １ ~x ２５ Using the above as input data, an estimation model of the diastolic blood pressure value of person A is generated by multiple regression analysis. In FIG. 5, the estimation model of the systolic blood pressure value and the estimation model of the diastolic blood pressure value of person A are shown as estimation model A. The estimation model A is generated by using, for example, biometric information of person A (for example, feature quantity x １ ~x ２５ ) as a variable. Similarly, for persons B and C, an estimation model B for person B (specifically, an estimation model for the systolic blood pressure value and the diastolic blood pressure value of person B) and an estimation model C for person C (specifically, an estimation model for the systolic blood pressure value and the diastolic blood pressure value of person C) are generated.

[0111] Thus, the estimation model generated by the blood pressure estimation model generating unit 201 through multiple regression analysis is an individual-specific model, and feature quantities (input data) and blood pressure values ​​(teaching data) are prepared for each individual for learning. The estimation model is a polynomial as shown in Equation 1 and Equation 2 above. However, in some cases, certain coefficients may be set to 0, and some terms may not be used during estimation. As will be described later, the method for generating the estimation model by the blood pressure estimation model generating unit 201 is not limited to a method using multiple regression analysis, but may also be a method using other regression analyses such as neural networks. For example, when multiple regression analysis is used in the blood pressure estimation model generating unit 201, processing speed can be increased. On the other hand, when a neural network is used in the blood pressure estimation model generating unit 201, the accuracy of machine learning can be improved.

[0112] Next, the learning unit 200 (blood pressure estimation model inferencing unit 202) generates a trained model by performing machine learning based on the acquired biometric information of each of the plurality of people and the generated estimation model of blood pressure values ​​of each of the plurality of people (step S23). For example, the blood pressure estimation model inferencing unit 202 generates a trained neural network model (trained model) by training a neural network that outputs an estimation model of blood pressure values ​​based on the biometric information of each of the plurality of people.

[0113] The blood pressure estimation model inference unit 202 uses the electrocardiogram waveforms, pulse waveforms, PTTv, and PPI of each of a plurality of people (e.g., people A, B, and C) as input data, and the generated polynomials of each of the plurality of people (e.g., people A, B, and C) as training data to train a neural network that outputs an estimation model (polynomial) of blood pressure values. In other words, the blood pressure estimation model inference unit 202 performs machine learning based on the input data and training data of the plurality of people, and calculates the coefficients of the polynomials (e.g., a ｉ and b ｉ) and constants (e.g., c and d). By using such a trained neural network, the blood pressure estimation model inference unit 20 can output the coefficients and constants of an estimation model suitable for an unknown person (i.e., the subject) not used in the training, based on the electrocardiogram waveform, pulse waveform, and features such as PTTv and PPI of the subject, when estimating blood pressure. Then, the coefficients and constants are calculated based on the feature values ​​of the subject (e.g., feature value x １ ~x ２５ ) and adding a constant to estimate the subject's blood pressure value.

[0114] As described above, an estimation model of a subject's blood pressure value is inferred using a trained model generated by machine learning. This estimation model is inferred from the subject's biometric information and is an optimal model for the subject. Therefore, the inferred estimation model can be used to accurately estimate the blood pressure value for each individual. Furthermore, because a cuff is not used, the blood pressure estimation device 100 can be made smaller, and, for example, a wearable device can be equipped with a blood pressure measurement function. Alternatively, the portability of the blood pressure estimation device 100 can be improved. Furthermore, because a cuff is not used to estimate (measure) blood pressure, the portability of the blood pressure estimation device is improved, making it possible to measure blood pressure continuously or at all times. Therefore, the subject's condition can be understood in detail from instantaneous changes in blood pressure, which was previously difficult to determine.

[0115] (Variation of Embodiment 1) In the first embodiment, an example has been described in which the blood pressure estimation model inference unit 20 includes the learning unit 200. However, the blood pressure estimation model inference unit 20 does not necessarily have to include the learning unit 200. This will be described as a variation of the first embodiment with reference to FIG. 6 .

[0116] FIG. 6 is a block diagram showing an example of a blood pressure estimation device 100a according to a modification of the first embodiment.

[0117] The blood pressure estimation device 100a according to the variation of the first embodiment differs from the blood pressure estimation device 100 according to the first embodiment in that it includes an estimation unit 50a instead of the estimation unit 50, and the blood pressure estimation model inference unit 20a of the estimation unit 50a does not include the learning unit 200. The other points are basically the same as those in the first embodiment, and therefore detailed description thereof will be omitted.

[0118] For example, the learning unit 200 may be provided in a computer external to the blood pressure estimation device 100a. The external computer may be a server or the like. The learning unit 200 provided in the external computer acquires biometric information (electrocardiogram waveforms, pulse waveforms, and feature values) of multiple people and blood pressure values ​​(systolic blood pressure values ​​and diastolic blood pressure values) of multiple people used as training data, and generates a trained model. The blood pressure estimation device 100a then receives the trained model generated by the learning unit 200 via a communication unit (not shown) for communicating with the external computer, and the blood pressure estimation model inference unit 20a may use the received trained model to infer an estimation model for the subject's blood pressure value.

[0119] (Embodiment 2) Next, a blood pressure estimation device according to embodiment 2 will be described.

[0120] FIG. 7 is a block diagram showing an example of a blood pressure estimation device 100b according to the second embodiment.

[0121] The blood pressure estimation device 100b according to the second embodiment differs from the blood pressure estimation device 100 according to the first embodiment in that it includes an estimation unit 50b instead of the estimation unit 50. The estimation unit 50b according to the second embodiment differs from the estimation unit 50 according to the first embodiment in that it includes a blood pressure estimation model inference unit 20b instead of the blood pressure estimation model inference unit 20. The blood pressure estimation model inference unit 20b differs from the blood pressure estimation model inference unit 20 according to the first embodiment in that it includes a learning unit 200b instead of the learning unit 200. Other points are basically the same as those in the first embodiment, and therefore detailed description thereof will be omitted.

[0122] The learning unit 200b generates a trained model by performing machine learning based on the blood pressure values ​​and biometric information of each of multiple people. The learning unit 200b also has a function of performing additional training on the trained model using the subject's blood pressure values ​​(e.g., systolic blood pressure values ​​and diastolic blood pressure values ​​measured using a cuff) and blood pressure values ​​estimated by the blood pressure estimation unit 30 (e.g., estimated systolic blood pressure values ​​and diastolic blood pressure values). Details of the learning unit 200b will be described using Figures 8 and 9.

[0123] FIG. 8 is a diagram for explaining an example of the operation during learning of the blood pressure estimation device 100b according to the second embodiment.

[0124] First, the learning unit 200b acquires blood pressure values ​​and biometric information of each of the plurality of people. For example, the learning unit 200b acquires systolic and diastolic blood pressure values ​​as blood pressure values ​​of each of the plurality of people. The systolic and diastolic blood pressure values ​​are, for example, blood pressure values ​​measured using a cuff. The learning unit 200b also acquires electrocardiogram waveforms, pulse waveforms, and feature quantities related to the electrocardiogram and pulse wave as biometric information of each of the plurality of people. For example, the feature quantities related to the electrocardiogram and pulse wave are feature quantities extracted from the electrocardiogram waveform and pulse waveform, and specifically, the above-mentioned feature quantity x １ ~x ２５ For the sake of simplicity, persons A, B, and C are shown in FIG. 8 as an example of a plurality of persons, but for example, blood pressure values ​​and biological information of many persons may be acquired and used for learning.

[0125] Next, the learning unit 200b (blood pressure estimation model generating unit 201) generates an estimation model of the blood pressure values ​​of each of the plurality of people based on the blood pressure values ​​and biological information of each of the plurality of people. For example, the blood pressure estimation model generating unit 201 generates an estimation model of the blood pressure values ​​of each of the plurality of people by regression analysis. For example, a blood pressure estimation model for each individual is generated by regression analysis using blood pressure values ​​measured by a cuff-based method as training data and biological information (feature values) as input data. For example, for person A, the systolic blood pressure value of person A is used as training data, and the feature value x １ ~x ２５Using the above as input data, a model for estimating the systolic blood pressure of person A is generated by multiple regression analysis. Also, using the diastolic blood pressure of person A as training data, the feature quantity x １ ~x ２５ Using the above as input data, an estimation model of the diastolic blood pressure value of person A is generated by multiple regression analysis. In FIG. 8, the estimation model of the systolic blood pressure value and the estimation model of the diastolic blood pressure value of person A are shown as estimation model A. The estimation model A is generated by using, for example, biometric information of person A (for example, feature quantity x １ ~x ２５ ) as a variable. Similarly, for persons B and C, an estimation model B for person B (specifically, an estimation model for the systolic blood pressure value and the diastolic blood pressure value of person B) and an estimation model C for person C (specifically, an estimation model for the systolic blood pressure value and the diastolic blood pressure value of person C) are generated.

[0126] Thus, the estimation model generated by the blood pressure estimation model generating unit 201 through multiple regression analysis is an individual-specific model, and feature quantities (input data) and blood pressure values ​​(teaching data) are prepared for each individual for learning. The estimation model is a polynomial as shown in Equation 1 and Equation 2 above. However, in some cases, certain coefficients may be set to 0, and some terms may not be used during estimation. As will be described later, the method for generating the estimation model by the blood pressure estimation model generating unit 201 is not limited to a method using multiple regression analysis, but may also be a method using other regression analyses such as neural networks. For example, when multiple regression analysis is used in the blood pressure estimation model generating unit 201, processing speed can be increased. On the other hand, when a neural network is used in the blood pressure estimation model generating unit 201, the accuracy of machine learning can be improved.

[0127] Next, the learning unit 200b (blood pressure estimation model inference unit 202b) uses the biometric information of each of the multiple people as input data, and performs machine learning using the generated estimation models of the blood pressure values ​​of each of the multiple people and the blood pressure values ​​of each of the multiple people as training data to generate a trained model. For example, the blood pressure estimation model inference unit 202b trains a neural network that outputs estimation models of blood pressure values ​​based on the biometric information of each of the multiple people, thereby generating a trained neural network model (trained model). In the first embodiment, the blood pressure estimation model inference unit 202b does not use the blood pressure values ​​of each of the multiple people as training data, but in the second embodiment, the blood pressure estimation model inference unit 202b uses the blood pressure values ​​of each of the multiple people as training data.

[0128] Specifically, the blood pressure estimation model inference unit learning unit 202b uses the electrocardiogram waveforms, pulse waveforms, PTTv, and PPI of each of a plurality of people (e.g., people A, B, and C) as input data, and uses the generated polynomials of each of the plurality of people (e.g., people A, B, and C) and the blood pressure values ​​of each of the plurality of people (e.g., people A, B, and C) as teacher data to learn a neural network that outputs an estimation model (polynomial) of blood pressure values. Each time the blood pressure estimation model inference unit 202b learns, in other words, each time the weight coefficients of the neural network are updated, the blood pressure estimation model inference unit 202b infers an estimation model based on the input data, and applies a feature (e.g., feature x １ ~x ２５ ) to estimate the blood pressure values ​​of multiple people (e.g., people A, B, and C). The blood pressure estimation model inference unit 202b compares the estimated blood pressure values ​​with blood pressure values ​​that are training data to determine the error, and performs learning to reduce this error. In this way, the blood pressure estimation model inference unit 202b generates a trained model.

[0129] By using such a trained neural network, the blood pressure estimation model inference unit 20b can output the coefficients and constants of an estimation model suitable for an unknown person (i.e., a subject) not used in the training, from the electrocardiogram waveform, pulse waveform, and features such as PTTv and PPI of the subject, when estimating blood pressure. Then, the coefficients and constants of the estimation model suitable for the subject are calculated based on the feature values ​​of the subject (e.g., feature value x １ ~x ２５ ) and adding a constant, the subject's blood pressure value can be estimated. However, because the trained model is a model trained using biometric information from people different from the subject, the estimation model output from the trained model may not be able to accurately estimate the blood pressure value for some subjects.

[0130] Therefore, in the second embodiment, additional learning of the trained model is performed using biometric information of the subject, etc.

[0131] FIG. 9 is a diagram for explaining an example of the operation during additional learning of the blood pressure estimation device 100b according to the second embodiment.

[0132] The learning unit 200b (blood pressure estimation model inference unit 202b) generates a trained model for the subject by performing additional machine learning using the acquired subject's biometric information as input data and the subject's blood pressure values ​​as training data. Since it is difficult to prepare a correct estimation model for the subject when performing additional machine learning, the subject's blood pressure values ​​are used as training data. For example, when the subject uses the blood pressure estimation device 100b for the first time, the blood pressure values ​​and biometric information (such as electrocardiogram waveform and pulse waveform) are measured multiple times (e.g., three times) to allow the blood pressure estimation model inference unit 202b to acquire data for additional learning. For example, the biometric information is measured by the measurement unit 300, and the blood pressure values ​​are measured using a blood pressure monitor that uses a cuff or the like.

[0133] The blood pressure estimation model inference unit learning unit 202b performs additional learning using the subject's electrocardiogram waveform, pulse waveform, PTTv, and PPI as input data and the subject's blood pressure value as training data. The blood pressure estimation model inference unit 202b infers an estimation model in the same way as in the pre-learning, and applies a feature (e.g., feature x１ ~x ２５ ) to estimate the subject's blood pressure value. The blood pressure estimation model inference unit 202b compares the estimated blood pressure value with the blood pressure value of the training data to determine the error, and performs learning to reduce this error. This makes it possible to generate a trained model for the subject that outputs an estimation model that can accurately estimate the subject's blood pressure value.

[0134] In embodiment 2, as described in Figure 8, pre-training of the trained model is performed so that the error between the estimated blood pressure value and the correct blood pressure value is reduced, and therefore, additional training of the trained model can be performed so that the error between the subject's estimated blood pressure value and the correct blood pressure value is reduced.

[0135] FIG. 10 is a diagram for explaining an example of the operation of the blood pressure estimation device 100b according to the second embodiment during blood pressure estimation.

[0136] By using such a neural network that has been additionally trained for a subject, the blood pressure estimation model inference unit 20b can output coefficients and constants (blood pressure estimation unit 30) of an estimation model that is more suitable for the subject from the subject's electrocardiogram waveform, pulse waveform, and features such as PTTv and PPI. Then, the coefficients and constants of the estimation model that are more suitable for the subject (for example, feature x １ ~x ２５ ) and adding a constant, the subject's blood pressure value can be estimated more accurately.

[0137] For example, while Fig. 10 shows an example in which an estimation model is inferred each time the subject's blood pressure value is estimated, an estimation model for the subject may be generated when the subject uses the blood pressure estimation device 100b for the first time, and the estimation model for the subject may be used thereafter when estimating the subject's blood pressure value. This will be described with reference to Fig. 11.

[0138] 11 is a diagram illustrating another example of the operation of the blood pressure estimation device 100b according to embodiment 2 during blood pressure estimation. In this example, the blood pressure estimation device 100b includes a blood pressure estimation unit 30b that estimates the blood pressure value of the subject using an estimation model for the subject, instead of the blood pressure estimation unit 30.

[0139] The blood pressure estimation model inference unit 20b infers multiple (e.g., three) estimation models of the subject's blood pressure value by inputting multiple (e.g., three) pieces of biological information (e.g., electrocardiogram waveform, pulse waveform, PTTv, and PPI) of the subject used in the additional machine learning into a trained model for the subject (additionally trained neural network). The blood pressure estimation model inference unit 20b then generates an estimation model for one subject from the multiple inferred estimation models of the subject's blood pressure value. For example, the blood pressure estimation model inference unit 20b generates an estimation model for one subject by calculating the average value of each coefficient and each constant of the multiple inferred estimation models (polynomials) of the subject's blood pressure value. Note that when calculating the average value, outliers may be excluded. Alternatively, an estimation model for a subject may be generated by calculating a median or the like instead of the average value.

[0140] Then, the blood pressure estimation unit 30b estimates the acquired biological information of the subject (for example, the feature amount x １ ~x ２５ ) and an estimation model for the subject. In this way, it is no longer necessary to infer an estimation model each time the subject's blood pressure value is estimated, and after the estimation model for the subject is generated, the subject's blood pressure value can be estimated using the estimation model for the subject.

[0141] As described above, because blood pressure values ​​and biometric information are used when generating the trained model, additional learning can be performed on the generated trained model using the subject's biometric information and blood pressure values. Specifically, the blood pressure values ​​of each of the multiple people are also used as training data, and machine learning is performed in advance to reduce the error between the estimated blood pressure values ​​of each of the multiple people estimated by the estimation model and the blood pressure values ​​used as training data. This makes it possible to perform additional learning using the subject's biometric information and blood pressure values. Then, by performing additional machine learning using the subject's biometric information as input data and the subject's blood pressure values ​​as training data, it is possible to generate a trained model for the subject that outputs an estimation model that can accurately estimate the subject's blood pressure value.

[0142] In the second embodiment, as in the modified example of the first embodiment, the learning unit 200b may be provided in a computer external to the blood pressure estimation device 100b.

[0143] (Embodiment 3) Next, a blood pressure estimation device according to embodiment 3 will be described.

[0144] FIG. 12 is a diagram for explaining an example of the operation during learning of the blood pressure estimation device according to the third embodiment.

[0145] The blood pressure estimation device according to the third embodiment differs from the blood pressure estimation device 100b according to the second embodiment in that it includes a learning unit 200c instead of the learning unit 200b. The other points are the same as those in the second embodiment, and therefore, a description thereof will be omitted.

[0146] As shown in Fig. 12, the learning unit 200c (blood pressure estimation model inference unit 202c) generates a trained model by performing machine learning using the biometric information of each of the multiple people as input data and the blood pressure values ​​of each of the multiple people as training data. Unlike the learning unit 200b according to the second embodiment, the learning unit 200c does not include the blood pressure estimation model generation unit 201. In other words, the blood pressure estimation model inference unit 202c does not use the estimation models of the blood pressure values ​​of each of the multiple people as training data. Even in this case, the blood pressure estimation model inference unit 202c trains a neural network that outputs an estimation model of the blood pressure values ​​based on the biometric information of each of the multiple people, thereby generating a trained neural network model (trained model).

[0147] Specifically, the blood pressure estimation model inference unit 202c uses the electrocardiogram waveforms, pulse waveforms, PTTv, and PPI of each of a plurality of people (e.g., people A, B, and C) as input data, and the blood pressure values ​​of each of the plurality of people (e.g., people A, B, and C) as training data to train a neural network that outputs an estimation model (polynomial) of blood pressure values. Each time the blood pressure estimation model inference unit 202c learns, in other words, each time the weight coefficients of the neural network are updated, the blood pressure estimation model inference unit 202c infers an estimation model based on the input data, and applies a feature (e.g., feature x１ ~x ２５ ) to estimate the blood pressure values ​​of multiple people (e.g., people A, B, and C). The blood pressure estimation model inference unit 202c compares the estimated blood pressure values ​​with blood pressure values ​​that are training data to determine the error, and performs learning to reduce this error. In this way, the blood pressure estimation model inference unit 202c generates a trained model.

[0148] In embodiment 3 as well, when generating a trained model, machine learning is performed using biometric information as input data and blood pressure values ​​as training data, so it is possible to perform additional learning on the generated trained model using the subject's biometric information as input data and the subject's blood pressure values ​​as training data.

[0149] In the third embodiment, as in the modified example of the first embodiment, the learning unit 200c may be provided in a computer external to the blood pressure estimation device.

[0150] While the blood pressure estimation device according to one or more aspects of the present disclosure has been described above based on the embodiments, the present disclosure is not limited to these embodiments. As long as they do not deviate from the spirit of the present disclosure, various modifications conceivable by a person skilled in the art to each embodiment and configurations constructed by combining components of different embodiments may also be included within the scope of one or more aspects of the present disclosure.

[0151] For example, in the above embodiment, an example in which multiple regression analysis is used in the blood pressure estimation model generating unit 201 has been described, but a neural network may also be used. For example, the blood pressure estimation model generating unit 201 may generate weighting coefficients of a neural network as an estimation model for each of a plurality of people. The blood pressure estimation model inferrer learning unit 202 may use the electrocardiogram waveforms, pulse waveforms, PTTv, and PPI of each of a plurality of people as input data, and the weighting coefficients of the generated neural networks for the plurality of people as training data to learn a neural network that outputs an estimation model (neural network) of blood pressure values. In this case, the blood pressure estimation model inferrer 20 may use the acquired biological information of the subject (e.g., feature quantity x ) as an estimation model for the subject's blood pressure value. １ ~x ２５ ) to infer the weight coefficients of a neural network that uses the input.

[0152] For example, in the above embodiment, an electrocardiogram waveform, a pulse waveform, and information for synchronizing the time between the electrocardiogram waveform and the pulse waveform (e.g., PTTv and PPI) are described as examples of biological information used by the blood pressure estimation model inference unit for inferring the estimation model, and a feature amount related to the electrocardiogram and pulse wave (e.g., feature amount x １ ~x ２５ ) has been described. That is, an example has been described in which the biological information used by the blood pressure estimation model inference unit to infer the estimation model is different from the biological information used by the blood pressure estimation unit to estimate the blood pressure value. However, the biological information used by the blood pressure estimation model inference unit to infer the estimation model may be the same as the biological information used by the blood pressure estimation unit to estimate the blood pressure value. For example, the biological information used by the blood pressure estimation model inference unit to infer the estimation model and the biological information used by the blood pressure estimation unit to estimate the blood pressure value both contain feature quantities related to the electrocardiogram and pulse wave (for example, feature quantity x １ ~x ２５ ) may also be used.

[0153] For example, in the above embodiment, the information for synchronizing the time between the electrocardiogram waveform and the pulse waveform is the PTTv and PPI, but this is not limiting. For example, the information for synchronizing the time between the electrocardiogram waveform and the pulse waveform may be the pulse wave transit time (PTTv) from the R wave of the electrocardiogram waveform to the peak of the pulse waveform and the R-R interval (RRI) from the R wave of the electrocardiogram waveform to the next R wave of the electrocardiogram waveform.

[0154] For example, in the above embodiment, the biological information is described as information related to an electrocardiogram and a pulse wave, but this is not limiting. For example, the biological information may be a ballistocardiogram, a phonocardiogram, or bioimpedance. Furthermore, biological information such as height, weight, blood oxygen saturation (SpO2), or body temperature may be used to estimate a blood pressure value.

[0155] For example, the present disclosure can be realized not only as a blood pressure estimation device, but also as a blood pressure estimation method including steps (processing) performed by components that make up the blood pressure estimation device.

[0156] The blood pressure estimation method is a method executed by a computer, and as shown in Figure 3, includes an acquisition step (step S11) of acquiring biometric information of a subject, a blood pressure estimation model inference step (step S12) of inferring an estimation model of the subject's blood pressure value by inputting the acquired biometric information of the subject into a trained model generated by machine learning, and a blood pressure estimation step (step S13) of estimating the subject's blood pressure value based on the acquired biometric information of the subject and the inferred estimation model of the subject's blood pressure value.

[0157] For example, the present disclosure may be realized as a program for causing a processor to execute steps included in a blood pressure estimation method. Furthermore, the present disclosure may be realized as a non-transitory computer-readable recording medium, such as a CD-ROM, on which the program is recorded.

[0158] For example, when the present disclosure is realized as a program (software), each step is performed by running the program using hardware resources such as a computer's CPU, memory, input / output circuits, etc. In other words, each step is performed by the CPU acquiring data from memory or input / output circuits, etc., performing calculations, and outputting the calculation results to memory or input / output circuits, etc.

[0159] In the above-described embodiments, each component included in the blood pressure estimation device may be configured with dedicated hardware or may be realized by executing a software program suitable for each component. Each component may also be realized by a program execution unit such as a CPU or processor reading and executing a software program recorded on a recording medium such as a hard disk or semiconductor memory.

[0160] Some or all of the functions of the blood pressure estimation device according to the above-described embodiments are typically realized as an LSI, which is an integrated circuit. These functions may be individually integrated into a single chip, or some or all of them may be integrated into a single chip. Furthermore, the integrated circuit is not limited to an LSI, and may be realized using a dedicated circuit or a general-purpose processor. An FPGA (Field Programmable Gate Array), which can be programmed after LSI fabrication, or a reconfigurable processor, which can reconfigure the connections and settings of circuit cells within an LSI, may also be used.

[0161] The present disclosure is applicable to devices such as wearable devices that measure blood pressure without using a cuff.

[0162] 10 Acquisition unit 20, 20a, 20b Blood pressure estimation model inference unit 30, 30b Blood pressure estimation unit 40 Output unit 50, 50a, 50b Estimation unit 100, 100a, 100b Blood pressure estimation device 200, 200b, 200c Learning unit 201 Blood pressure estimation model generation unit 202, 202b, 202c Blood pressure estimation model inference device learning unit 300 Measurement unit 301 Electrode 302 Electrocardiogram signal acquisition unit 303 Light emission unit 304 Light reception unit 305 Pulse wave signal acquisition unit 306 Processing unit

Claims

1. An acquisition unit that acquires biometric information of a subject; a blood pressure estimation model inference unit that infers an estimation model of the subject's blood pressure value by inputting the acquired biological information of the subject into a trained model generated by machine learning; and a blood pressure estimation unit that estimates a blood pressure value of the subject based on the acquired biological information of the subject and an estimation model of the inferred blood pressure value of the subject. Blood pressure estimator.

2. The blood pressure estimation device further includes a learning unit that generates the trained model by performing machine learning based on biological information of a plurality of people. The blood pressure estimation device according to claim 1 .

3. The learning unit generates the trained model by performing machine learning based on the blood pressure values ​​and biological information of each of the plurality of people. The blood pressure estimation device according to claim 2 .

4. The learning unit generates the trained model by performing machine learning using biometric information of each of the plurality of people as input data and blood pressure values ​​of each of the plurality of people as teacher data. The blood pressure estimation device according to claim 3.

5. The learning unit is generating an estimation model for the blood pressure value of each of the plurality of persons based on the blood pressure value and biological information of each of the plurality of persons; Using the biometric information of each of the plurality of people as input data, a generated estimation model of the blood pressure values ​​of each of the plurality of people and the blood pressure values ​​of each of the plurality of people as teacher data are used to perform machine learning to generate the trained model. The blood pressure estimation device according to claim 3.

6. The learning unit generates the trained model for the subject by performing additional machine learning using the acquired biological information of the subject as input data and the blood pressure value of the subject as teacher data. The blood pressure estimation device according to claim 3.

7. the blood pressure estimation model inference unit infers a plurality of estimation models of the subject's blood pressure value by inputting a plurality of pieces of biological information of the subject used when additional machine learning was performed into the trained model for the subject, and generates one estimation model for the subject from the plurality of inferred estimation models of the subject's blood pressure value; The blood pressure estimation unit estimates a blood pressure value of the subject based on the acquired biological information of the subject and an estimation model for the subject. The blood pressure estimation device according to claim 6.

8. The learning unit is generating an estimation model for the blood pressure value of each of the plurality of persons based on the blood pressure value and biological information of each of the plurality of persons; generating the trained model by performing machine learning based on the biological information of each of the plurality of people and the generated estimation model of the blood pressure value of each of the plurality of people; The blood pressure estimation device according to claim 2 .

9. The learning unit generates an estimation model of the blood pressure value of each of the plurality of people by regression analysis. The blood pressure estimation device according to claim 5.

10. The blood pressure estimation model inference unit infers a polynomial having the acquired biological information of the subject as a variable as an estimation model of the subject's blood pressure value. The blood pressure estimation device according to claim 1 .

11. The blood pressure estimation model inference unit infers weighting coefficients of a neural network using the acquired biological information of the subject as an input as an estimation model of the subject's blood pressure value. The blood pressure estimation device according to claim 1 .

12. The biological information is biological information related to an electrocardiogram and a pulse wave. The blood pressure estimation device according to claim 1 .

13. the blood pressure estimation model inference unit infers an estimation model of the subject's blood pressure value by inputting, into the trained model, an electrocardiogram waveform, a pulse waveform, and information for synchronizing the time between the electrocardiogram waveform and the pulse waveform as the acquired biological information of the subject; The blood pressure estimation unit estimates a blood pressure value of the subject based on features related to an electrocardiogram and a pulse wave as the acquired biological information of the subject and an estimation model of the inferred blood pressure value of the subject. The blood pressure estimation device according to claim 12.

14. The information for synchronizing the time between the electrocardiogram waveform and the pulse waveform includes a pulse wave propagation time from the R wave of the electrocardiogram waveform to the valley of the pulse waveform, and a time from the peak of the pulse waveform to the peak of the next pulse waveform. The blood pressure estimation device according to claim 13.

15. The blood pressure estimation unit estimates the blood pressure value of the subject by multiplying the acquired biological information of the subject by a coefficient included in an estimation model of the inferred blood pressure value of the subject. The blood pressure estimation device according to any one of claims 1 to 14.

16. 1. A computer-implemented method for blood pressure estimation, comprising: An acquisition step of acquiring biometric information of a subject; a blood pressure estimation model inference step of inferring an estimation model of the subject's blood pressure value by inputting the acquired biological information of the subject into a trained model generated by machine learning; and a blood pressure estimation step of estimating a blood pressure value of the subject based on the acquired biological information of the subject and an estimation model of the inferred blood pressure value of the subject. Blood pressure estimation methods.

17. A program for causing a computer to execute the blood pressure estimation method according to claim 16.