Sphygmomanometer, blood pressure measurement method, blood pressure measurement program, learning model construction method, and learning model construction program

Abstract

Provided is a sphygmomanometer capable of measuring blood pressure with high accuracy as much as possible without performing data collection for parameter setting of a blood pressure measurement algorithm and calibration of measurement results. This sphygmomanometer is provided with: a cuff wound around the arm of a person to be measured; an inspection unit that inspects the cuff pressure received by the cuff from the arm; a decomposition unit that acquires, on the basis of the cuff pressure, a first waveform indicating a time-series change in a characteristic quantity relating to the blood pressure of the subject, and that decomposes the first waveform into a plurality of second waveforms indicating two or more different characteristic quantities; a learning model constructed by machine learning using the second waveforms for the plurality of subjects as explanatory variables and blood pressure measured by auscultation as target variables; a measurement unit that inputs the plurality of second waveforms into the learning model to measure the blood pressure of the subject; and an output unit that outputs the blood pressure measured by the measurement unit.

Description

Technical Field

[0001] The present invention relates to a sphygmomanometer, a blood pressure measurement method, a blood pressure measurement program, a method for constructing a learning model, and a program for constructing a learning model. Background Art

[0002] People are accustomed to grasping their health status through blood pressure measurement. In the measurement of blood pressure, attempts are being made to improve the measurement accuracy of blood pressure through machine learning including deep learning. As a method for measuring blood pressure using deep learning, for example, a method of inputting the shape of each beat of a pulse and the shape of an envelope of a pulse waveform into a model to measure blood pressure is being studied (see Non-Patent Documents 1 to 3).

[0003] Prior Art Documents

[0004] Non-Patent Documents

[0005] Non-Patent Document 1: Ahmadreza Argha, Branko G. Celler, Nigel H. Lovell, Artificial Intelligence Based Blood Pressure Estimation From Auscultatory and Oscillometric Waveforms: A Methodological Review, 《IEEE REVIEWS IN BIOMEDICAL ENGINEERING》, (USA), 2022, VOL. 15, p. 152 - p. 168 <​​​​Non-Patent Literature 3: Ahmed S. Alghamdi, Kemal Polat, Abdullah Alghoson, Abdulrahman A. Alshdadi, Ahmed A. Abd El-Latif, Gaussian process regression (GPR) based non-invasive continuous blood pressure prediction method from cuff oscillometric signals, [online], February 20, 2020, [retrieved March 28, 2023], Internet <URL: https: / / www.sciencedirect.com / science / article / abs / pii / S0003682X20301079?via%3Dihub>. Summary of the Invention

[0008] The technical problem that the invention aims to solve

[0009] In oscillometric blood pressure measurement, since blood pressure is measured based on the waveform of the pulse wave, the accuracy of the measurement will decrease if the pulse wave contains noise. The noise in the pulse wave can be increased by factors such as the method of wrapping the cuff and the position of the cuff. Therefore, to suppress the influence of this noise, calibration of the measured blood pressure and data collection for setting the parameters of the blood pressure measurement algorithm are implemented. However, even with calibration and parameter setting of the blood pressure measurement algorithm, the cost of data collection is still incurred unless there is a guarantee against measurement errors caused by noise removal.

[0010] One aspect of the disclosed technology aims to provide a blood pressure monitor, blood pressure measurement method, blood pressure measurement program, method for constructing a learning model, and program for constructing a learning model that can perform blood pressure measurement with the highest possible accuracy even without implementing data collection for parameter setting of the blood pressure measurement algorithm and calibration of measurement results.

[0011] Solutions for solving technical problems

[0012] One aspect of the disclosed technology is illustrated by a blood pressure monitor as follows. This blood pressure monitor includes: a cuff, wrapped around the arm of a person being measured; a measurement unit, measuring the cuff pressure received from the arm; a decomposition unit, acquiring a first waveform representing a temporal change in a characteristic quantity related to the blood pressure of the person being measured based on the cuff pressure, and decomposing the first waveform into multiple second waveforms representing two or more different characteristic quantities; a learning model, constructed through machine learning using the second waveforms of the subject as explanatory variables and blood pressure measured by auscultation as the target variable; a measurement unit, inputting the multiple second waveforms into the learning model to measure the blood pressure of the person being measured; and an output unit, outputting the blood pressure measured by the measurement unit.

[0013] According to this blood pressure monitor, the first waveform representing the time-series changes of characteristic quantities related to the blood pressure of the subject is decomposed into multiple second waveforms representing two or more different characteristic quantities. That is, from multiple perspectives, the second waveforms can be considered as waveforms representing the first waveform. Therefore, it can also be considered that when noise is present in the first waveform, even if the noise has a significant impact on the second waveform related to one characteristic quantity, the noise has a relatively small impact on the second waveform related to other characteristic quantities. Therefore, according to this blood pressure monitor, even without data collection for parameter setting of the blood pressure measurement algorithm or calibration of the measurement results, high-precision blood pressure measurement can be performed as much as possible.

[0014] In this blood pressure monitor, the first waveform may include a cuff pressure waveform representing the temporal change of the cuff pressure. The decomposition unit decomposes the cuff pressure waveform into multiple third waveforms representing two or more different characteristic quantities. The learning model is constructed using the third waveform with respect to the subject as the explanatory variable. The measurement unit inputs the third waveform into the learning model to measure the blood pressure of the subject.

[0015] Alternatively, in this blood pressure monitor, the first waveform may include a pulse waveform representing the temporal change of the pulse wave of the subject obtained based on the temporal change of the cuff pressure. The decomposition unit decomposes the pulse waveform into multiple fourth waveforms representing two or more different feature quantities. The learning model is constructed using the fourth waveform of the subject as the explanatory variable. The measurement unit inputs the fourth waveform into the learning model to measure the blood pressure of the subject.

[0016] Furthermore, in this blood pressure monitor, the first waveform may include a cuff pressure waveform representing the temporal change of the cuff pressure and a pulse waveform representing the temporal change of the subject's pulse wave obtained based on the temporal change of the cuff pressure. The decomposition unit decomposes the cuff pressure waveform into multiple third waveforms representing two or more different feature quantities, and decomposes the pulse waveform into a fourth waveform obtained from the temporal change of the cuff pressure. The learning model is constructed using the third and fourth waveforms related to the subject as explanatory variables. The measurement unit inputs the third and fourth waveforms into the learning model to measure the subject's blood pressure. Alternatively, the decomposition unit may include a Wave-U-Net.

[0017] The technology disclosed herein can also be mastered from the aspects of blood pressure measurement methods, blood pressure measurement procedures, methods for constructing learning models, and procedures for constructing learning models.

[0018] The effects of the invention

[0019] According to the technology disclosed herein, even without data collection for setting parameters for the blood pressure measurement algorithm or calibration of the measurement results, it is possible to perform blood pressure measurement with the highest possible accuracy. Attached Figure Description

[0020] Figure 1 This is a diagram illustrating an example of the configuration of a blood pressure monitor according to an embodiment.

[0021] Figure 2 This is a schematic diagram showing the state of the blood pressure monitor after it has been placed on the wrist of the person being measured.

[0022] Figure 3 This diagram schematically illustrates the processing of the waveform decomposition section according to the embodiment.

[0023] Figure 4 The first figure schematically illustrates the processing of the blood pressure measuring unit according to the embodiment.

[0024] Figure 5 The second figure schematically illustrates the processing of the blood pressure measuring unit according to the embodiment.

[0025] Figure 6 The first figure schematically illustrates the construction of the blood pressure model involved in the implementation method.

[0026] Figure 7 The second figure schematically illustrates the construction of the blood pressure model involved in the implementation method.

[0027] Figure 8 The third figure schematically illustrates the construction of the blood pressure model involved in the implementation method.

[0028] Figure 9 The fourth figure schematically illustrates the construction of the blood pressure model involved in the implementation method.

[0029] Figure 10 The fifth figure schematically illustrates the construction of the blood pressure model involved in the implementation method.

[0030] Figure 11 This is a diagram illustrating an example of the processing flow of a blood pressure monitor according to an embodiment.

[0031] Figure 12 This is a graph illustrating the verification results of the blood pressure measurement accuracy of the blood pressure monitor according to the illustrative embodiment. Detailed Implementation

[0032] <Application Examples>

[0033] An application example of the present invention will be described. The blood pressure monitor 100 involved in the application example is a device for measuring the blood pressure of a subject by wrapping a cuff 120 around the upper arm and wrist of the subject. In the blood pressure monitor 100, the measuring unit 114 inputs at least one of the temporal variation of the cuff pressure and the temporal variation of the pulse wave into the blood pressure model 117, and obtains the blood pressure of the subject from the blood pressure model 117.

[0034] The measuring unit 114 controls the pressure of the cuff 120 and acquires the temporal changes of characteristic quantities related to the blood pressure of the subject from the pressure sensor 113. For example, the temporal changes of characteristic quantities related to the blood pressure of the subject can be listed as the temporal changes of cuff pressure and the temporal changes of the waveform of the Oscillo Metric Waveform (OMW).

[0035] The measurement unit 114 includes a waveform decomposition unit 115 and a blood pressure measurement unit 116. The waveform decomposition unit 115 decomposes the temporal changes of characteristic quantities related to the blood pressure of the measured subject into multiple second waveforms (pattern waveforms). The pattern waveforms may include, for example, waveforms obtained by decomposing the temporal changes of characteristic quantities related to the blood pressure of the measured subject based on waveform characteristics such as frequency and amplitude. The pattern waveforms may also include waveforms that globally show the overall temporal changes of characteristic quantities related to the blood pressure of the measured subject, or waveforms that locally show a portion of the temporal changes of characteristic quantities related to the blood pressure of the measured subject. The waveform decomposition unit 115 may include, for example, a Wave-U-Net.

[0036] The blood pressure measurement unit 116 inputs multiple pattern waveforms decomposed by the waveform decomposition unit 115 into the blood pressure model 117, and acquires the subject's blood pressure from the blood pressure model 117. The blood pressure measurement unit 116 then outputs the acquired blood pressure to the output unit 118.

[0037] Blood pressure model 117 is a learning model that outputs the blood pressure of the subject after being input with multiple pattern waveforms. Blood pressure model 117 is constructed using machine learning, with multiple pattern waveforms obtained by decomposing the temporal variations of characteristic quantities related to the subject's blood pressure as explanatory variables and blood pressure measured by auscultation as the target variable. Machine learning includes deep learning.

[0038] According to this application example, the pattern waveform includes a waveform showing the global changes in the time-series variations of the characteristic quantities related to the blood pressure of the subject, and a waveform showing the local changes in the time-series variations of the characteristic quantities related to the blood pressure of the subject. Therefore, the blood pressure model 117 can comprehensively learn both the global and local characteristics of the time-series variations of the characteristic quantities related to the blood pressure of the subject. According to such a blood pressure model 117, even if the time-series variations of the characteristic quantities related to the blood pressure of the subject contain noise, the blood pressure of the subject can be output with high accuracy. That is, according to this application example, even if the time-series variations of the characteristic quantities related to the blood pressure of the subject acquired by the measurement unit 114 contain noise, high-precision blood pressure measurement can be achieved without data collection for parameter setting of the blood pressure measurement algorithm or calibration of the measurement results.

[0039] <Implementation Method>

[0040] The embodiments will now be described with reference to the accompanying drawings. Figure 1 This is a diagram illustrating an example of the configuration of the blood pressure monitor 100 according to the embodiment. Additionally, Figure 2 This diagram schematically illustrates the state of the blood pressure monitor 100 after it has been worn on the wrist H1 of the person being measured. The blood pressure monitor 100 includes a main body 110 and a cuff 120. The blood pressure monitor 100 measures the blood pressure of the person being measured by wrapping the cuff 120 around the wrist H1 of the person being measured. It should be noted that the position of wrapping the cuff 120 is not limited to the wrist H1, but can also be on the upper arm of the person being measured.

[0041] The cuff 120 is formed in the shape of a strip and has a pocket for supplying air to the inside. When air is supplied to the pocket inside the cuff 120 while the cuff 120 is wrapped around the upper arm or similar part of the subject being measured, the pressure of the cuff on the upper arm or similar part increases. Conversely, when air is drawn out from the pocket inside the cuff 120 while the cuff 120 is wrapped around the upper arm or similar part of the subject being measured, the pressure of the cuff on the upper arm or similar part decreases.

[0042] By configuring a processor and a storage unit in the main body 110, and having the processor execute programs stored in the storage unit, the various processing units, such as the start switch 111, control unit 112, pressure sensor 113, measurement unit 114, blood pressure model 117, and output unit 118, are implemented. The measurement unit 114 includes a waveform decomposition unit 115 and a blood pressure measurement unit 116. In other words, the blood pressure monitor 100 can be regarded as a computer with a cuff 120. It should be noted that the various processing units, such as the start switch 111, control unit 112, pressure sensor 113, measurement unit 114, blood pressure model 117, and output unit 118, can also be dedicated hardware circuits. In the blood pressure monitor 100, by turning on the start switch 111, the blood pressure monitor 100 is started, and blood pressure measurement begins.

[0043] The control unit 112 controls the pressure of the cuff 120 on the upper arm of the person being measured by supplying air to and drawing air out of the cuff 120. When the start switch 111 is turned on, the control unit 112 begins to control the pressure of the cuff 120.

[0044] Pressure sensor 113 is a sensor that detects the pressure (cuff pressure) of cuff 120. Pressure sensor 113 outputs the detected cuff pressure to measuring unit 114. Cuff pressure includes, for example, the pressure of cuff 120 pressing against the upper arm and the pressure vibration generated by the pulsation of blood flow in the arteries of the upper arm.

[0045] The measuring unit 114 measures blood pressure based on the pressure detected by the pressure sensor 113. It acquires from the pressure sensor 113 the temporal variation of the cuff pressure of the cuff 120 as it is increased to a pressure higher than the highest blood pressure and then gradually reduced to a pressure lower than the lowest blood pressure. Furthermore, the measuring unit 114 extracts from the temporal variation of the cuff pressure acquired from the pressure sensor 113 the temporal variation of pressure oscillations caused by blood flow pulsation, wherein the blood flow is arterial blood flow. The waveform representing the extracted temporal variation of the cuff pressure (hereinafter also referred to as the cuff pressure waveform) and the waveform representing the temporal variation of pressure oscillations (hereinafter also referred to as the pulse waveform) are processed by the waveform decomposition unit 115 and the blood pressure measuring unit 116, respectively.

[0046] The waveform decomposition unit 115 decomposes the cuff pressure waveform and pulse waveform into multiple pattern waveforms. Figure 3 This diagram schematically illustrates the processing of the waveform decomposition unit 115 according to the embodiment. For example, the waveform decomposition unit 115 decomposes the cuff compression waveform W11 into 14-channel pattern waveforms, each with a data length of 16384 bits, and generates a pattern waveform group W12. Additionally, the waveform decomposition unit 115 decomposes the pulse waveform W21 into 14-channel pattern waveforms, each with a data length of 16384 bits, and generates a pattern waveform group W22.

[0047] The waveform decomposition unit 115 may include, for example, Wave-U-Net. Wave-U-Net is, for example, a learning model for waveform decomposition of sound. Sound can be considered to contain non-periodic components in addition to periodic components; similarly, cuff compression waveforms and pulse waveforms also contain non-periodic components in addition to periodic components. Therefore, in this embodiment, the waveform decomposition unit 115 for decomposing cuff compression waveforms and pulse waveforms is assumed to include Wave-U-Net. Alternatively, the waveform decomposition unit 115 may also include, for example, any one of the following instead of Wave-U-Net: Deep Music Source Extractor (DEMUCS), Two-Stage Attention Symmetric Framework (D3NET), Fully Convolutional Temporal Audio Separation Network (Conv-TasNet), WaveNet, Slice-Based Attention Neural Network (Sams-Net), or Meta-Learning Temporal Audio Separation Network (Meta-TasNet).

[0048] Here, the pattern waveform, as also described in the application examples, may include waveforms obtained by decomposing the temporal changes of characteristic quantities related to the blood pressure of the measured subject based on waveform characteristics such as frequency and amplitude. Furthermore, the pattern waveform may also include waveforms that globally show the overall temporal changes of characteristic quantities related to the blood pressure of the measured subject, and waveforms that locally show a portion of the temporal changes of characteristic quantities related to the blood pressure of the measured subject. For example, a waveform showing the temporal changes of characteristic quantities related to the blood pressure of the measured subject can be listed as a globally shown waveform. A waveform showing a portion can be listed as a waveform obtained by extracting (selecting) a portion of the waveform showing the temporal changes of characteristic quantities related to the blood pressure of the measured subject.

[0049] The blood pressure measuring unit 116 measures the blood pressure of the subject based on the pattern waveform obtained by the waveform decomposition unit 115. Figure 4 and Figure 5 This diagram schematically illustrates the processing of the blood pressure measurement unit 116 according to the embodiment. The blood pressure measurement unit 116 applies a one-dimensional convolutional neural network (1D-CNN) to the pattern waveform group W12 input from the waveform decomposition unit 115 to replace it with 30-channel intermediate feature quantities W13, each with a data length of 1025 bits. The blood pressure measurement unit 116 further applies 1D-CNN to the intermediate feature quantities W13 to replace it with 5-channel intermediate feature quantities W14, each with a data length of 35 bits.

[0050] Furthermore, the blood pressure measurement unit 116 applies a one-dimensional convolutional neural network (1D-CNN) to the pattern waveform group W22 input from the waveform decomposition unit 115 to replace it with 30-channel intermediate feature quantities W23, each with a data length of 1025 bits. The blood pressure measurement unit 116 further applies 1D-CNN to the intermediate feature quantities W23 to replace it with 5-channel intermediate feature quantities W24, each with a data length of 35 bits. The blood pressure measurement unit 116 inputs the intermediate feature quantities W14 and W24 into itself and acquires the subject's blood pressure as the output of the blood pressure measurement unit 116.

[0051] Blood pressure model 117 is a learned model that outputs blood pressure after being input with intermediate feature quantity W14 generated from the cuff compression waveform and intermediate feature quantity W24 generated from the pulse waveform. In other words, blood pressure model 117 is a learned model that outputs blood pressure after being input with feature quantities from the cuff compression waveform and feature quantities from the pulse waveform. Blood pressure model 117 is stored, for example, in a storage unit housed in the main body 110.

[0052] The output unit 118 outputs the blood pressure measured by the blood pressure measuring unit 116. The output unit 118 can be, for example, a liquid crystal display (LCD), a plasma display panel (PDP), an inorganic electroluminescence (EL) panel, or an organic EL panel. The output unit 118 can also be a printer or a speaker.

[0053] <Construction of Blood Pressure Model 117>

[0054] Figures 6 to 10 This is a schematic diagram illustrating the construction of the blood pressure model 117 involved in the embodiment. (Refer to...) Figures 6 to 10 The construction of the blood pressure model 117 will be explained. In constructing the blood pressure model 117, multiple subjects (E1, E2, ..., EN) are prepared. It should be noted that although multiple subjects are prepared in this embodiment, only one subject may be prepared. In step T1, the measurement unit 114 acquires the cuff pressure waveform P11 and pulse waveform P21 of subject E1.

[0055] In step T2, the waveform decomposition unit 115 decomposes the cuff compression waveform P11 into 14-channel pattern waveforms with a data length of 16384 bits for each channel, and generates pattern waveform group P12. In step T3, the waveform decomposition unit 115 decomposes the pulse waveform P21 into 14-channel pattern waveforms with a data length of 16384 bits for each channel, and generates pattern waveform group P22.

[0056] In step T4, the blood pressure measurement unit 116 applies 1D-CNN to the pattern waveform group P12 input from the waveform decomposition unit 115 to replace it with 30-channel intermediate feature quantities P13, each with a data length of 1025 bits. In step T5, the blood pressure measurement unit 116 applies 1D-CNN to the intermediate feature quantity P13 to replace it with 5-channel intermediate feature quantities P14, each with a data length of 35 bits.

[0057] In step T6, the blood pressure measurement unit 116 applies 1D-CNN to the pattern waveform group P22 input from the waveform decomposition unit 115 to replace it with 30-channel intermediate feature quantities P23, each with a data length of 1025 bits. In step T7, the blood pressure measurement unit 116 applies 1D-CNN to the intermediate feature quantity P23 to replace it with 5-channel intermediate feature quantities P24, each with a data length of 35 bits.

[0058] In step T8, a blood pressure model 117 is generated using machine learning with intermediate feature quantity P14 generated from the cuff compression waveform and intermediate feature quantity P24 generated from the pulse waveform as explanatory variables and blood pressure as the target variable. It should be noted that the blood pressure as the target variable is, for example, blood pressure measured by auscultation. The processing in steps T1 to T8 is also performed on subjects E2 to EN respectively.

[0059] <Processing Flow>

[0060] Figure 11 This is a diagram illustrating an example of the processing flow of the blood pressure monitor 100 according to an embodiment. Referring below... Figure 4 An example of the processing procedure for a blood pressure monitor 100 will be explained.

[0061] In step S1, the blood pressure monitor 100 is activated by turning on the start switch 111. In step S2, the control unit 112 delivers air to the cuff 120 to apply pressure to the wrist of the person being measured. After applying pressure to the cuff 120 until it reaches a pressure higher than the highest blood pressure, the control unit 112 gradually reduces the pressure of the cuff 120.

[0062] In step S3, the measuring unit 114 acquires the cuff pressure waveform and pulse waveform based on the cuff pressure detected by the pressure sensor 113. Then, the waveform decomposition unit 115 decomposes the waveform acquired in step S3 into multiple pattern waveforms. In step S4, the blood pressure measuring unit 116 measures the subject's blood pressure using the blood pressure model 117 based on the pattern waveforms obtained in step S3.

[0063] In step S5, the measurement unit 114 determines whether the blood pressure measurement of the subject is complete. For example, if the measurement unit 114 measures both the subject's highest and lowest blood pressure, it determines that the blood pressure measurement is complete. If it is complete (yes in step S5), the process proceeds to step S6. If it is not complete (no in step S5), the process proceeds to step S2.

[0064] In step S6, the control unit 112 removes air from the cuff 120. In step S7, the output unit 118 outputs the blood pressure measured in step S4.

[0065] <Verification>

[0066] Since the accuracy of blood pressure measurement performed by the blood pressure monitor 100 according to the above-described embodiment has been verified, the explanation will be elaborated here. Figure 12 This is a graph illustrating the verification results of blood pressure measurement accuracy performed using the blood pressure monitor 100 according to the embodiment. Figure 12 In this study, blood pressure measured by a sphygmomanometer 100 and blood pressure measured by an oscilloscope were compared.

[0067] In this validation, multiple blood pressure measurements were performed on multiple subjects, and the results were obtained. The obtained measurement results were evaluated based on the standard deviation of error (SDE). (Refer to...) Figure 12 In the “Implementation Method”, higher accuracy than the “oscillometric measurement method” is shown for either the lowest blood pressure (DBP) or the highest blood pressure (SBP).

[0068] <Effects of the Implementation Method>

[0069] In this embodiment, the cuff pressure waveform and pulse waveform are each decomposed into multiple pattern waveforms. The pattern waveforms include waveforms showing global changes in the temporal variations of characteristic quantities related to the subject's blood pressure and waveforms showing local changes in the temporal variations of characteristic quantities related to the subject's blood pressure. Therefore, the blood pressure model 117 can comprehensively learn both the global and local characteristics of the temporal variations of characteristic quantities related to the subject's blood pressure. Based on such a blood pressure model 117, even if the cuff pressure waveform and pulse waveform contain noise, the subject's blood pressure can be output with high accuracy. That is, according to this embodiment, even if the cuff pressure waveform and pulse waveform contain noise due to the position of the cuff 120, the subject's posture, etc., high-precision blood pressure measurement can be achieved without calibration or parameter setting of the blood pressure measurement algorithm.

[0070] <Variation Example>

[0071] In the embodiment described above, both the cuff pressure waveform and the pulse waveform are input to the waveform decomposition unit 115, but either the cuff pressure waveform or the pulse waveform can be input to the waveform decomposition unit 115. In this case, the blood pressure model 117 can be constructed using intermediate feature quantities generated from the cuff pressure waveform and intermediate feature quantities generated from the pulse waveform, which are intermediate feature quantities generated from the waveform used for blood pressure measurement, as explanatory variables.

[0072] The implementation methods and variations disclosed above can be combined separately.

[0073] <Computer-readable recording media>

[0074] Information processing programs that enable computers, other devices, or apparatuses (hereinafter referred to as "computers, etc.") to perform any of the aforementioned functions can be recorded on a computer-readable recording medium. Furthermore, the functions can be provided by causing a computer or the like to read and execute the program on the recording medium.

[0075] Here, a computer-readable recording medium refers to a recording medium capable of storing data, programs, and other information through electrical, magnetic, optical, mechanical, or chemical processes and capable of reading such information from a computer or similar device. Examples of such recording media that can be removed from a computer or similar device include floppy disks, optical disks, Compact Disc Read Only Memory (CD-ROM), Compact Disc Recordable (CD-R), Compact Disc Rewriteable (CD-RW), Digital Versatile Disc (DVD), Blu-ray Disc (BD), Digital Audio Tape (DAT), 8mm magnetic tape, flash memory, external hard disk drives, and solid-state drives (SSDs). Additionally, recording media fixed to a computer or similar device include internal hard disk drives, SSDs, and ROMs.

[0076] <Postscript 1>

[0077] A blood pressure monitor (100) has the following features: A cuff (120) is wrapped around the arm (H1) of the person being measured; Inspection section (113) inspects the cuff (120) for the pressure exerted on the arm (H1) by the cuff; The decomposition unit (115) acquires a first waveform (W11, W21) representing the temporal change of characteristic quantities related to the blood pressure of the subject based on the cuff pressure, and decomposes the first waveform (W11, W21) into a plurality of second waveforms (W12, W22) representing two or more different characteristic quantities. The learning model (117) was constructed by machine learning with the second waveform (W12, W22) of the subject as explanatory variables and blood pressure measured by auscultation as the target variable; The measurement unit (116) inputs multiple second waveforms (W12, W22) into the learning model to measure the blood pressure of the subject; and The output unit (118) outputs the blood pressure measured by the measuring unit.

[0078] <Appendix 2>

[0079] According to the sphygmomanometer (100) described in Appendix 1, wherein, The first waveform (W11, W21) includes a cuff pressure waveform (W11) representing the temporal variation of the cuff pressure. The decomposition unit (115) decomposes the cuff pressure waveform (W11) into multiple third waveforms (W12) representing two or more different characteristic quantities. The learning model (117) is constructed using the third waveform (P14) of the subject (E1...EN) as the explanatory variable. The measuring unit (116) inputs the third waveform (W12) into the learning model (117) to measure the blood pressure of the subject.

[0080] <Appendix 3>

[0081] According to Appendix 1 or 2, the sphygmomanometer (100) wherein, The first waveform (W11, W21) includes a pulse waveform (W21) representing the temporal variation of the subject's pulse wave obtained based on the temporal variation of the cuff pressure. The decomposition unit (115) decomposes the pulse waveform (W21) into multiple fourth waveforms (W22) representing two or more different characteristic quantities. The learning model (117) is constructed using the fourth waveform (W24) with respect to the subject (E1...EN) as the explanatory variable. The measuring unit (116) inputs the fourth waveform (W22) into the learning model (117) to measure the blood pressure of the subject.

[0082] <Appendix 4>

[0083] According to any one of Annexes 1 to 3, the sphygmomanometer (100) wherein, The first waveform (W11, W21) includes a cuff pressure waveform (W11) representing the temporal variation of the cuff pressure and a pulse waveform (W21) representing the temporal variation of the pulse wave of the subject obtained based on the temporal variation of the cuff pressure. The decomposition unit (115) decomposes the cuff pressure waveform (W11) into multiple third waveforms (W12) representing two or more different characteristic quantities, and decomposes the pulse waveform (W21) into a fourth waveform (W22) obtained from the temporal variation of the cuff pressure. The learning model (117) is constructed using the third waveform (P14) and the fourth waveform (P24) with respect to the subject (E1...EN) as the explanatory variables. The measuring unit (116) inputs the third waveform (W12) and the fourth waveform (W22) into the learning model (117) to measure the blood pressure of the subject.

[0084] <Appendix 5>

[0085] According to any one of Annexes 1 to 4, the sphygmomanometer (100) wherein, The aforementioned decomposition section (115) includes Wave-U-Net.

[0086] <Appendix 6>

[0087] A blood pressure measurement method in which a computer performs the following steps: The inspection procedure involves checking the cuff (120) wrapped around the arm (H1) of the person being measured for the pressure exerted by the cuff on the arm (H1); The decomposition step involves acquiring a first waveform (W11, W21) representing the temporal changes of characteristic quantities related to the blood pressure of the subject based on the cuff pressure, and decomposing the first waveform (W11, W21) into multiple second waveforms (W12, W22) representing two or more different characteristic quantities. The measurement step involves inputting multiple second waveforms (W12, W22) into a learning model (117) constructed through machine learning, using the second waveforms (W12, W22) about the subject (E1…EN) as explanatory variables and blood pressure measured by auscultation as the target variable, to measure the blood pressure of the subject; and The output step outputs the blood pressure measured in the measurement step.

[0088] <Appendix 7>

[0089] A blood pressure measurement program that causes a computer to perform the following steps: The inspection procedure involves checking the cuff (120) wrapped around the arm (H1) of the subject (E1...EN) for the pressure exerted by the cuff on the arm (H1); The decomposition step involves acquiring a first waveform (W11, W21) representing the temporal changes of characteristic quantities related to the blood pressure of the subject based on the cuff pressure, and decomposing the first waveform (W11, W21) into multiple second waveforms (W12, W22) representing two or more different characteristic quantities. The measurement step involves inputting multiple second waveforms (W12, W22) into a learning model (117) constructed through machine learning, using the second waveforms (W12, W22) as explanatory variables and blood pressure measured by auscultation as the target variable, to measure the blood pressure of the subject; and The output step outputs the blood pressure measured in the measurement step.

[0090] <Postscript 8>

[0091] A method for constructing a learning model, in which the computer performs the following steps: The inspection procedure involves checking the cuff (120) wrapped around the arm (H1) of the subject (E1...EN) for the pressure exerted by the cuff on the arm (H1); The decomposition step involves acquiring a first waveform (W11, W21) representing the temporal variation of characteristic quantities related to the subject's blood pressure based on the cuff pressure, and decomposing the first waveform (W11, W21) into multiple second waveforms (W12, W22) representing two or more different characteristic quantities; and The construction step involves building a learning model (117) using machine learning with the second waveform (W12, W22) as explanatory variables and blood pressure measured by auscultation as the target variable.

[0092] <Postscript 9>

[0093] A procedure for building a learning model causes a computer to perform the following steps: The inspection procedure involves checking the cuff (120) wrapped around the subject's arm (H1) for the pressure exerted by the cuff on the arm (H1); The decomposition step involves acquiring a first waveform (W11, W21) representing the temporal variation of characteristic quantities related to the subject's blood pressure based on the cuff pressure, and decomposing the first waveform (W11, W21) into multiple second waveforms (W12, W22) representing two or more different characteristic quantities; and The construction step involves building a learning model (117) using machine learning with the second waveform (W12, W22) as explanatory variables and blood pressure measured by auscultation as the target variable.

[0094] Explanation of reference numerals in the attached figures

[0095] 100. Blood pressure monitor; 110. Main body; 111. Start switch; 112. Control unit; 113. Pressure sensor; 114. Measurement unit; 115. Waveform decomposition unit; 116. Blood pressure measurement unit; 117. Blood pressure model; 118. Output unit; 120. Cuff; H1. Wrist.

Claims

1. A blood pressure monitor, comprising: A cuff is wrapped around the arm of the person being measured. The inspection section checks the cuff pressure exerted on the arm. The decomposition unit acquires a first waveform representing the temporal changes of characteristic quantities related to the blood pressure of the subject based on the cuff pressure, and decomposes the first waveform into multiple second waveforms representing two or more different characteristic quantities. The learning model was constructed through machine learning, using the second waveform of the subject as the explanatory variable and blood pressure measured by auscultation as the target variable. The measurement unit inputs multiple second waveforms into the learning model to measure the blood pressure of the subject; as well as The output unit outputs the blood pressure measured by the measuring unit.

2. The blood pressure monitor according to claim 1, wherein, The first waveform includes a cuff pressure waveform representing the temporal variation of the cuff pressure. The decomposition unit decomposes the cuff compression waveform into multiple third waveforms representing two or more different characteristic quantities. The learning model is constructed using the third waveform of the subject as the explanatory variable. The measurement unit inputs the third waveform into the learning model to measure the blood pressure of the subject.

3. The blood pressure monitor according to claim 1, wherein, The first waveform includes a pulse waveform representing the temporal variation of the subject's pulse wave obtained based on the temporal variation of the cuff pressure. The decomposition unit decomposes the pulse waveform into multiple fourth waveforms representing two or more different characteristic quantities. The learning model is constructed using the fourth waveform of the subject as the explanatory variable. The measurement unit inputs the fourth waveform into the learning model to measure the blood pressure of the subject.

4. The blood pressure monitor according to claim 1, wherein, The first waveform includes a cuff pressure waveform representing the temporal variation of the cuff pressure and a pulse waveform representing the temporal variation of the subject's pulse wave obtained based on the temporal variation of the cuff pressure. The decomposition unit decomposes the cuff pressure waveform into multiple third waveforms representing two or more different characteristic quantities, and decomposes the pulse waveform into a fourth waveform obtained from the temporal variation of the cuff pressure. The learning model is constructed using the third and fourth waveforms of the subject as explanatory variables. The measurement unit inputs the third and fourth waveforms into the learning model to measure the blood pressure of the subject.

5. The sphygmomanometer according to any one of claims 1 to 4, wherein, The decomposition unit includes Wave-U-Net.

6. A blood pressure measurement method, wherein a computer performs the following steps: The inspection procedure involves checking the cuff wrapped around the arm of the person being measured for the pressure exerted by the cuff on the arm. The decomposition step involves acquiring a first waveform representing the temporal changes of characteristic quantities related to the blood pressure of the subject based on the cuff pressure, and decomposing the first waveform into multiple second waveforms representing two or more different characteristic quantities. The measurement step involves inputting multiple second waveforms into a learning model constructed through machine learning, with the second waveform about the subject as an explanatory variable and blood pressure measured by auscultation as a target variable, to measure the blood pressure of the subject. as well as The output step outputs the blood pressure measured in the measurement step.

7. A blood pressure measurement program that causes a computer to perform the following steps: The inspection procedure involves checking the pressure exerted by the cuff wrapped around the subject's arm on the arm. The decomposition step involves acquiring a first waveform representing the temporal changes of characteristic quantities related to the blood pressure of the subject based on the cuff pressure, and decomposing the first waveform into multiple second waveforms representing two or more different characteristic quantities. The measurement step involves inputting multiple second waveforms into a learning model constructed through machine learning, using the second waveform as an explanatory variable and blood pressure measured by auscultation as the target variable, to measure the blood pressure of the subject. as well as The output step outputs the blood pressure measured in the measurement step.

8. A method for constructing a learning model, wherein a computer performs the following steps: The inspection procedure involves checking the pressure exerted by the cuff wrapped around the subject's arm on the arm. The decomposition step involves acquiring a first waveform representing the temporal changes of characteristic quantities related to the blood pressure of the subject based on the cuff pressure, and decomposing the first waveform into multiple second waveforms representing two or more different characteristic quantities. as well as The construction steps involve building a learning model using machine learning with the second waveform as the explanatory variable and blood pressure measured by auscultation as the target variable.

9. A procedure for building a learning model, which causes a computer to perform the following steps: The inspection procedure involves checking the pressure exerted by the cuff wrapped around the subject's arm on the arm. The decomposition step involves acquiring a first waveform representing the temporal changes of characteristic quantities related to the blood pressure of the subject based on the cuff pressure, and decomposing the first waveform into multiple second waveforms representing two or more different characteristic quantities. as well as The construction steps involve building a learning model using machine learning with the second waveform as the explanatory variable and blood pressure measured by auscultation as the target variable.

Images