Blood pressure estimation device, computer program, and recording medium

Abstract

To allow blood pressure to be estimated from a plurality of kinds of biological indexes, and to enhance versatility.SOLUTION: A blood pressure estimation device 100 applies, to a regression equation, a gradient value (GT value or GI value) obtained from each physiological index, or applies an average value of the gradient value of each physiological index and estimates blood pressure using integrated correlation data that integrates correlation data of each kind obtained from a plurality of kinds of biological indexes related to a response of a cardiovascular system, preferably three indexes of heart rate variation (HRV), acoustic pulse wave amplitude variation (AFV) obtained through a body surface, and pulse wave propagation delay time variation (PPDTV).SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present invention relates to a blood pressure estimation device, a computer program, and a recording medium that estimate brachial blood pressure values ​​without wrapping a cuff around the upper arm. [Background technology]

[0002] The applicant has disclosed in Patent Document 1 a blood pressure estimation device that analyzes biosignal data obtained from a biosignal detection sensor using a three-dimensional knitted fabric, captures in vivo vibration information caused by blood flow fluctuations from the ventricular filling phase to the isovolumetric contraction phase, obtains an index showing the fluctuations by applying the Lorentz plot method, and estimates blood pressure values ​​from that index (Patent Document 1).

[0003] Specifically, this technique involves capturing biological signals from the body surface on the back using a biosignal detection sensor, filtering the time waveform within a predetermined frequency band to obtain a filtered waveform that reveals the cardiac cycle, comparing this filtered waveform with electrocardiogram waveform data obtained from an electrocardiograph simultaneously, identifying waveform components from the ventricular filling phase to the isovolumetric contraction phase, then applying the Lorentz plot method to construct a new time waveform based on the angle difference between the slope of a large number of points plotted over a predetermined measurement time and the slope of a point cloud with a shorter measurement time, performing frequency analysis on this time waveform, displaying the frequency analysis results on a log-log axis, finding a regression line for the range from VLF to LF in the analyzed waveform, defining the slope of this regression line as the fractal slope, and estimating blood pressure from the fractal slope. [Prior art documents] [Patent Documents]

[0004] [Patent Document 1] Japanese Patent Publication No. 2019-122502 [Overview of the project] [Problems that the invention aims to solve]

[0005] According to Patent Document 1, blood pressure can be estimated without restraint using biosignals collected from a person's back. Therefore, it has the advantage of being easy to measure without the need to wrap a cuff around the upper arm, and because it is non-restraining, blood pressure can be measured continuously. However, in the case of the technology described in Patent Document 1, while a certain level of accuracy can be achieved in a quiet environment, there were concerns regarding accuracy in environments subject to noise such as vibration, such as when taking measurements inside a moving vehicle.

[0006] In light of this, the inventors first investigated the correlation between the average values ​​of heart rate (HR) and pulse propagation delay time (PPDT) and brachial blood pressure (systolic blood pressure: SBP), as well as the correlation between the average values ​​of the amplitude ratios of the A and E waves of the apical impulse and brachial blood pressure (systolic blood pressure: SBP). It was found that the correlation with these individual physiological indicators was sometimes low, indicating a lack of stability in the accuracy of blood pressure estimation.

[0007] Therefore, we analyzed acoustic pulse waves (APW) obtained via the body surface during the atrial and ventricular contractions. We applied filtering to the APW at 10-30 Hz to create a time waveform that could capture extreme values ​​at time intervals of 0.03-0.1 seconds. Instead of the amplitude ratio of apical impulse waves in the frequency band from 0.5 Hz to the boundary frequency, we decided to correlate the amplitude variation (APW to be Filtered-variability: AFV) that captures the atrial and ventricular contractions during their respective phases with systolic blood pressure. Specifically, the amplitude fluctuations (AFV) of acoustic pulse waves obtained via the body surface during the atrial and ventricular contractions were processed using a Lorentz plot, and the derivative (standard slope) was calculated over 360 seconds. Next, the derivative with respect to the standard slope was calculated over 30 seconds, and a periodic function of the derivative (θi function) with a period of 3 seconds was created to capture the tachycardia of the sympathetic nervous system, which occurs slowly, and the bradycardia of the parasympathetic nervous system, which occurs rapidly. When the frequency analysis results of this θi function were expressed as a quartic function, the gradient of the tangent to the inflection point (GT) and the gradient between inflections (GI) were obtained. By calculating an arterial pressure regression line (SBP-Regression line: SR) that shows the correlation with brachial blood pressure values, using GT and GI values ​​obtained from the amplitude fluctuations of acoustic pulse waves obtained through the body surface as explanatory variables, and using this line to estimate blood pressure, the coefficient of determination was higher than when using an arterial pressure regression line based on the average value of the amplitude ratio of the A and E waves of the apical impulse wave. Therefore, the inventors proposed an invention related to blood pressure estimation based on this finding as Japanese Patent Application No. 2022-72790.

[0008] However, subsequent research has shown that while blood pressure estimation methods using GT and GI values ​​of amplitude fluctuations of acoustic pulse waves obtained via the body surface have improved in static environments, they are still not entirely sufficient in terms of estimation accuracy in dynamic environments such as while riding in a vehicle.

[0009] The present invention has been made in view of the above, and aims to provide a blood pressure estimation device, a computer program, and a recording medium that can obtain sufficient estimation accuracy not only in a static environment but also in a dynamic environment. [Means for solving the problem]

[0010] To solve the above problems, the inventors focused on the following points. The sympathetic nervous system regulates the contractility of the myocardium, i.e., stroke volume. Therefore, the regulation of heart rate and stroke volume, which are determinants of cardiac output, is under the control of the central nervous system. Heart rate is regulated by the sympathetic and parasympathetic nerves distributed in the sinoatrial node and atrioventricular node. At rest, the inhibitory effect of the parasympathetic nervous system is dominant. Physiologically, heart rate is regulated by the balance between the sympathetic and parasympathetic nervous systems. Heart rate variability responds more quickly to stimulation of the parasympathetic nervous system than to the sympathetic nervous system. Fluctuations in body temperature also affect heart rate; for example, a 1°C increase in body temperature due to fever increases heart rate by approximately 10 beats / min. Conversely, a decrease in body temperature decreases heart rate. Sympathetic nerve stimulation increases heart rate, contractility, and relaxation rate, and shortens diastole, the phase in which blood fills the heart. Stroke volume is influenced by two factors with opposing effects: one is the energy required for ventricular contraction, and the other is the arterial pressure required for blood ejection. Contractility is increased by raising end-diastolic pressure, which stretches the myocardium during quiescence. There is also an increase in contractility, where contractile force increases while muscle length remains constant, due to the heart's law known as "Starling's Law of the heart" and increased sympathetic nervous system activity and elevated adrenaline levels in the blood. When arterial pressure rises, stroke volume decreases because ejection does not begin until intraventricular pressure exceeds arterial pressure. If arterial pressure is high, a lot of energy is consumed to raise intraventricular pressure during isovolumetric systole, reducing the energy available for ejection. Pulse pressure is proportional to stroke volume divided by arterial compliance. The pulse wave propagation velocity depends on the stiffness of the arterial wall. Arterial walls become stiffer with increased arterial pressure and aging, so the velocity is high. Therefore, the distensibility of human arterial walls can be evaluated from the pulse wave propagation delay time. When the pulse wave propagation velocity is fast, the reflected wave returns to late systole, causing a further increase in late systolic blood pressure. Heart rate, arterial pressure, and pulse wave delay time are each influenced by two factors. The cardiovascular system functions through the coordinated action of individual organs interacting with each other, working and responding as a whole. The principle of adaptive integration applies, where the responses of individual organs are integrated to form the response of the entire body.

[0011] In other words, the contraction of the heart and blood vessels is reflexively controlled by the nervous and endocrine systems, and the brain harmonizes these two systems. One important reflex observed in the cardiovascular system is the arterial baroreceptor reflex (ABF), which regulates blood pressure to stabilize blood flow to the brain. Based on information from the baroreceptor reflex, the activity level of the autonomic nerves that control the heart and blood vessels is reflexively regulated. This changes the patterns of heart rate variability (HRV), which is related to cardiac output; acoustic pulse wave (APW), which is obtained via the body surface and is related to venous volume; amplitude variability (AFV), which is related to the acoustic pulse wave; and pulse wave delay time (PPDTV), which is related to peripheral resistance of blood vessels, thereby regulating blood pressure.

[0012] Therefore, by plotting these three physiological indicators (HRV, AFV, PPDTV) on the same graph, creating integrated correlation data, and calculating the regression equation (Integration-SBP-Regression line (ISR)) in this integrated correlation data, and using the GT and GI values ​​of heart rate variability, the GT and GI values ​​of amplitude variability of acoustic pulse waves obtained via the body surface, and the GT and GI values ​​of pulse wave propagation delay time individually, or the average of these three indicators, as explanatory variables, it is expected that the accuracy of blood pressure estimation can be improved.

[0013] In other words, the blood pressure estimation device of the present invention is A gradient value calculation unit analyzes time waveform data of multiple types of biometric indicators related to cardiovascular responses, obtains each analyzed waveform on a log-log axis, defines a quartic function for the analyzed waveform, and calculates the gradient value between two inflection points or the gradient value of the tangent line at an inflection point according to the number of extreme values, and outputs one of these as the analysis result. A type-specific correlation data calculation unit calculates type-specific correlation data by correlating the gradient value, which is the result of the analysis, with the measured upper arm blood pressure value for each type of biometric indicator. Using the time waveform data obtained from a plurality of subjects, the type-specific correlation data obtained for each type of the biological index is developed on the same coordinate to obtain integrated correlation data, and an integrated correlation data calculation unit for obtaining the integrated correlation data; For the subject whose blood pressure is to be estimated, the gradient value calculation unit analyzes the time waveform data of at least one biological index among the plurality of types of biological indexes, obtains the gradient value as the analysis result, and substitutes this gradient value into the regression formula of the integrated correlation data to obtain an estimated blood pressure value corresponding to the upper arm blood pressure value, and an estimated blood pressure calculation unit; It has.

[0014] The estimated blood pressure calculation unit For the subject whose blood pressure is to be estimated, the gradient value calculation unit analyzes the time waveform data of two or more biological indexes among the plurality of types of biological indexes, obtains the gradient value as the analysis result for each biological index, and substitutes the average value of the gradient values obtained for each biological index into the regression formula of the integrated correlation data to obtain an estimated blood pressure value corresponding to the upper arm blood pressure value, which is preferable. The estimated blood pressure calculation unit For the subject whose blood pressure is to be estimated, the gradient value calculation unit analyzes the time waveform data of two or more biological indexes among the plurality of types of biological indexes, obtains the gradient value as the analysis result for each biological index, substitutes the gradient value obtained for each biological index into the regression formula of the corresponding type-specific correlation data to obtain a type-specific estimated blood pressure value, and outputs the average value of the obtained plurality of type-specific estimated blood pressure values as the estimated blood pressure value corresponding to the upper arm blood pressure value, which is also preferable. The plurality of types of biological signals are preferably selected from fluctuations in heart rate, fluctuations in pulse wave propagation delay time, and amplitude fluctuations of acoustic pulse waves obtained through the body surface.

[0015] The gradient value calculation unit When there are three extreme values of the quartic function and it becomes a concave-convex function sandwiching two inflection points, the gradient value between the two inflection points is adopted as the analysis result, and when the number of extreme values is two or less, the gradient value of the tangent line at any inflection point is adopted as the analysis result, which is preferable. The gradient value calculation unit For each of the plurality of types of biological indicators, apply the Lorenz plot method to obtain the standard slope from the Lorenz plot in the reference time range, and sequentially obtain the relative slope, which is the difference between the slope of each Lorenz plot created at each predetermined time shorter than the reference time range and the standard slope, with a predetermined overlap rate and within a predetermined time window. Filter the time waveform of the obtained relative slope at a predetermined frequency, perform frequency analysis on the filtered time waveform, convert the frequency analysis result into an analysis waveform shown in the double logarithmic axis display, apply a regression line with a slope of 1 / f to the analysis waveform, obtain the standard deviation of the distances of each point constituting the analysis waveform with respect to the regression line, and preferably obtain the quartic function for the point group of the analysis waveform within the range of a predetermined frequency band extracted based on this standard deviation.

[0016] The gradient value calculation unit When using the "amplitude variation of the acoustic pulse wave obtained through the body surface" as the biological indicator, obtain the amplitude between two extreme values from the extreme values included in the phases of atrial contraction and ventricular contraction, apply the Lorenz plot method, obtain the standard slope from the Lorenz plot in the reference time range, and sequentially obtain the relative slope, which is the difference between the slope of each Lorenz plot created at each predetermined time shorter than the reference time range and the standard slope, with a predetermined overlap rate and within a predetermined time window. Filter the time waveform of the obtained relative slope at a predetermined frequency, perform frequency analysis on the filtered time waveform, convert the frequency analysis result into an analysis waveform shown in the double logarithmic axis display, apply a regression line with a slope of 1 / f to the analysis waveform, obtain the standard deviation of the distances of each point constituting the analysis waveform with respect to the regression line, and preferably obtain the quartic function for the point group of the analysis waveform within the range of a predetermined frequency band extracted based on this standard deviation.

[0017] In addition, the present invention analyzes the time waveform data of a plurality of types of biological indicators related to the response of the cardiovascular system, obtains each analysis waveform in the double logarithmic axis display, defines a quartic function for the analysis waveform, and according to the number of extreme values, obtains the gradient value between two inflection points or the gradient value of the tangent line at the inflection point, and outputs either of them as the analysis result, and A procedure for obtaining type-specific correlation data by correlating the aforementioned gradient value, which is the result of the analysis, with the measured upper arm blood pressure value, for each type of biometric indicator, A procedure for obtaining integrated correlation data by using the time waveform data obtained from multiple subjects, unfolding the multiple correlation data for each type of biometric on the same coordinate system, and For individuals whose blood pressure is to be estimated, the procedure involves analyzing the time waveform data of at least one of the multiple types of biometric indicators, obtaining the gradient value resulting from the analysis, and substituting this gradient value into the regression equation of the integrated correlation data to obtain an estimated blood pressure value corresponding to the brachial blood pressure value. A computer program that runs and makes the computer function as a blood pressure estimation device. To provide.

[0018] In the procedure for determining the estimated blood pressure value, For the individuals whose blood pressure is to be estimated, it is preferable to analyze the time waveform data of two or more of the multiple types of biometric indicators, obtain the gradient value which is the result of the analysis for each biometric indicator, and substitute the average value of the gradient values ​​obtained for each biometric indicator into the regression equation of the integrated correlation data to obtain an estimated blood pressure value corresponding to the brachial blood pressure value. In the procedure for determining the estimated blood pressure value, It is also preferable to analyze the time waveform data of two or more of the multiple types of bioindicators for the person whose blood pressure is to be estimated, obtain the gradient value which is the result of the analysis for each bioindicator, substitute the gradient value obtained for each bioindicator into the regression equation of the corresponding type-specific correlation data to obtain the type-specific estimated blood pressure value, and output the average of the multiple type-specific estimated blood pressure values ​​obtained as the estimated blood pressure value corresponding to the brachial blood pressure value. Preferably, the aforementioned multiple types of biosignals are selected from variations in heart rate, variations in pulse wave propagation delay time, and variations in the amplitude of acoustic pulse waves obtained via the body surface. In the procedure for determining the gradient value, When the quartic function has three extrema and is a concave-convex function with two inflection points in between, it is preferable to use the gradient value between the two inflection points as the analysis result. When the extrema is two or less, it is preferable to use the gradient value of the tangent line at any of the inflection points as the analysis result. In the procedure for determining the gradient value, Preferably, the Lorentz plot method is applied to each of the aforementioned multiple types of biometric indicators, the standard slope is determined from the Lorentz plot in a reference time range, the relative slope, which is the difference between the slope of each Lorentz plot created at predetermined time intervals shorter than the reference time range and the standard slope, is successively determined with a predetermined overlap rate and a predetermined time window, the time waveform of the obtained relative slope is filtered at a predetermined frequency, frequency analysis is performed on the filtered time waveform, the result of the frequency analysis is converted into an analysis waveform shown on a log-log axis, a regression line with a slope of 1 / f is applied to the analysis waveform, the standard deviation of the distance of each point constituting the analysis waveform with respect to the regression line is determined, and the quartic function is determined for the point cloud of analysis waveforms in a predetermined frequency band range extracted based on this standard deviation.

[0019] In the procedure for determining the gradient value, When using the "amplitude fluctuation of acoustic pulse waves obtained via the body surface" as the bioindicator, it is preferable to determine the amplitude between two extreme values ​​from the extreme values ​​included in the time phases of atrial and ventricular contraction, apply the Lorentz plot method, determine the standard slope from the Lorentz plot in a reference time range, sequentially determine the relative slope, which is the difference between the slope of each Lorentz plot created at predetermined time intervals shorter than the reference time range and the standard slope, with a predetermined overlap rate and a predetermined time window, filter the time waveform of the obtained relative slope at a predetermined frequency, perform frequency analysis on the filtered time waveform, convert the frequency analysis result into an analysis waveform shown on a log-log axis, apply a regression line with a slope of 1 / f to the analysis waveform, determine the standard deviation of the distance of each point constituting the analysis waveform with respect to the regression line, and determine the quartic function for the point cloud of analysis waveforms in a predetermined frequency band range extracted based on this standard deviation.

[0020] Furthermore, the present invention provides a computer-readable recording medium on which the aforementioned computer program is recorded. [Effects of the Invention]

[0021] According to the present invention, blood pressure is estimated by using integrated correlation data obtained by integrating type-specific correlation data from multiple types of biomarkers related to the cardiovascular system response, preferably three indices: heart rate variability (HRV), amplitude variability of acoustic pulse waves obtained via the body surface (AFV), and pulse wave delay time variability (PPDTV). The gradient values ​​(GT values ​​or GI values) obtained from each physiological indicator are then applied to the regression equation of this integrated correlation data, or the average value of the gradient values ​​of each physiological indicator is applied. As described above, these physiological indicators work in coordination as a whole cardiovascular system, responding and functioning together, and blood pressure is regulated by them in combination. Therefore, using integrated correlation data leads to an improvement in the accuracy of blood pressure prediction.

[0022] Furthermore, the integrated correlation data allows for the individual application of gradient values ​​for each physiological indicator, or the application of their mean values. For example, in dynamic environments such as during travel, by utilizing heart rate, which can detect relatively stable data even in dynamic environments, and using the gradient value obtained from heart rate fluctuations as an explanatory variable, the accuracy of blood pressure estimation in dynamic environments can be improved. [Brief explanation of the drawing]

[0023] [Figure 1] Figure 1(a) is an external perspective view showing a biosignal detection sensor (4SR) used in one embodiment of the present invention, Figure 1(b) is an external perspective view showing the air pack and gel pack separated, and Figure 1(c) is a cross-sectional view. [Figure 2] Figure 2 schematically shows the mounting position of the biosignal detection sensor when collecting acoustic pulse waves (R-APW) from the posterior chest. [Figure 3] Figure 3 is a block diagram showing a schematic configuration of a blood pressure estimation device according to one embodiment of the present invention. [Figure 4]Figure 4 is a block diagram showing the schematic configuration of the first gradient value calculation unit. [Figure 5] Figure 5 is a flowchart illustrating the analysis process in the first gradient value calculation unit. [Figure 6] Figure 6 shows the data output at each step in Figure 5, following the flowchart in Figure 5. [Figure 7] Figure 7 is a flowchart illustrating the analysis process in the second gradient value calculation unit. [Figure 8] Figure 8 shows the data output at each step in Figure 7, following the flowchart in Figure 7. [Figure 9] Figure 9 is a flowchart illustrating the analysis process in the third gradient value calculation unit. [Figure 10] Figure 10 shows the data output at each step in Figure 9, following the flowchart in Figure 9. [Figure 11] Figure 11 is a flowchart showing the process for determining the estimated blood pressure value of a person whose blood pressure is to be estimated. [Figure 12] Figures 12(a) and 12(b) illustrate how to find an approximate curve represented by a quartic function. [Figure 13] Figure 13 shows the correlation between heart rate and measured systolic blood pressure (upper arm blood pressure values) measured by an upper arm blood pressure monitor for 76 subjects in Experiment Example 1. [Figure 14] Figure 14 shows the correlation between the standard slope calculated by the second gradient calculation unit and the systolic blood pressure value for each subject, using the data from Figure 13. [Figure 15] Figure 15 shows the correlation between the gradient values ​​of heart rate variability (GT value, GI value) and brachial blood pressure values. [Figure 16] Figure 16 shows the correlation between pulse wave delay time and brachial blood pressure values ​​for 76 subjects in Experiment Example 1. [Figure 17] Figure 17 shows the correlation between the standard slope calculated by the third gradient calculation unit and the systolic blood pressure value for each subject, using the data from Figure 16. [Figure 18] Figure 18 shows the correlation between the gradient values ​​(GT and GI values) of the variation in pulse wave propagation delay time and brachial blood pressure. [Figure 19] Figure 19 shows the correlation between the gradient value of the amplitude fluctuation of the acoustic pulse wave and the brachial blood pressure value for 76 subjects in Experiment Example 1. [Figure 20] Figure 20 shows the integrated correlation data, plotted on the same coordinate system all of the data that make up the three types of correlation data shown in Figures 15, 18, and 19. [Figure 21] Figure 21 shows the correlation between heart rate and measured systolic blood pressure (upper arm blood pressure values) measured by an upper arm blood pressure monitor for 26 subjects in Experiment Example 2. [Figure 22] Figure 22 shows the correlation between the standard slope calculated by the second gradient calculation unit and the systolic blood pressure value for each subject, using the data from Figure 21. [Figure 23] Figure 23 shows the correlation between the gradient values ​​of heart rate variability (GT value, GI value) and brachial blood pressure values. [Figure 24] Figure 24 shows the correlation between pulse wave delay time and brachial blood pressure values ​​for 76 subjects in Experiment Example 2. [Figure 25] Figure 25 shows the correlation between the standard slope calculated by the third gradient calculation unit and the systolic blood pressure value for each subject, using the data from Figure 24. [Figure 26] Figure 26 shows the correlation between the gradient values ​​(GT and GI values) of the variation in pulse wave propagation delay time and brachial blood pressure. [Figure 27] Figure 27 shows the correlation between the gradient value of the amplitude variation of the acoustic pulse wave and the brachial blood pressure value for 26 subjects in Experiment Example 2. [Figure 28] Figure 28 shows the integrated correlation data, plotted on the same coordinate system all of the data constituting the three types of correlation data shown in Figures 23, 26, and 27. [Figure 29]Figure 29 shows the correlation data by type, which is a combination of the correlation data by type for 102 subjects based on heart rate variability shown in Figure 15 of Experimental Example 1 and Figure 23 of Experimental Example 2. [Figure 30] Figure 30 shows the correlation data by type for 102 subjects, which is a combination of the correlation data by type based on pulse wave propagation delay time shown in Figure 18 of Experimental Example 1 and Figure 26 of Experimental Example 2. [Figure 31] Figure 31 shows the correlation data by type, which is a combination of the correlation data by type for 102 subjects based on the amplitude fluctuations of acoustic pulse waves shown in Figure 19 of Experimental Example 1 and Figure 27 of Experimental Example 2. [Figure 32] Figure 32 shows the integrated correlation data, which is a compilation of the type-specific correlation data from Figures 29 to 31. [Figure 33] Figures 33(a) to (f) show the time waveforms of various biosignal data in a sleep experiment involving hypertensive subjects. [Figure 34] Figures 34(a) to (c) are Lorentz plots of heart rate variability in a sleep experiment with hypertensive subjects. (a) is a plot using data from the entire sleep experiment (40 minutes), (b) is a plot using data from 1500 to 1860 seconds (6 minutes) during the sleep experiment, and (c) is a plot using data measured in a seated position after waking. Figures 34(d) to (f) are Lorentz plots of acoustic pulse wave amplitude variability. (d) is a plot using data from the entire sleep experiment (40 minutes), (e) is a plot using data from 1500 to 1860 seconds (6 minutes) during the sleep experiment, and (f) is a plot using data measured in a seated position after waking. [Figure 35] Figures 35(a) to (d) show the analysis results displayed on log-log axes, obtained using the regression lines of the Lorenz plots shown in Figures 34(a) to (f), and the gradient values ​​(GT and GI values) obtained by applying a quartic function to them. Figure 35(e) shows the estimated blood pressure values ​​obtained by substituting the integrated correlation data and the gradient values ​​from Figures 35(a) to (d) into Figure 32. [Figure 36]Figure 36 shows correlation data by type, illustrating the correlation between the gradient values ​​(GT and GI values) of acoustic pulse wave amplitude variation (AF) and systolic blood pressure values ​​(measured values) in 21 healthy subjects aged 21-65 years (average age: 43 years), including 2 women. [Figure 37] Figure 37 shows the correlation data by type, with all the data from Figures 19, 27, and 36 combined into a single coordinate system. [Modes for carrying out the invention]

[0024] The present invention will now be described in more detail based on embodiments of the present invention shown in the drawings. In this embodiment, three types of bioindicators related to the response of the cardiovascular system are used: heart rate variability, pulse wave delay time variability, and amplitude variability of the acoustic pulse wave (APW) obtained via the body surface. As described above, based on information from the baroreceptor reflex, the activity level of the autonomic nerves that innervate the heart and blood vessels is reflexively regulated, which changes the patterns of heart rate variability (HRV) related to cardiac output, APW amplitude variability (AFV) related to venous volume, and pulse wave delay time variability (PPDTV) related to the amplitude variability of APW (AFV) and peripheral resistance of blood vessels, thereby regulating blood pressure. Therefore, it is preferable to use these three.

[0025] Heart rate and acoustic pulse waves (APW) obtained via the body surface can be measured using the acoustic pulse wave detection sensor 1, which uses a three-dimensional knitted fabric and a microphone, as described later. Heart rate can also be measured using an electrocardiograph. Portable electrocardiographs and wearable heart rate monitors such as wristwatches exist, so it is preferable to use such portable or wearable types for measurements in dynamic environments, such as during riding. Pulse wave propagation delay time can be measured using an electrocardiograph and a fingertip volume plethysmometer. It is also possible to measure by placing the acoustic pulse wave detection sensor 1 at two locations at a predetermined interval, such as on the back of the chest and near the periphery.

[0026] Next, the configuration of the acoustic pulse wave detection sensor 1 for measuring acoustic pulse waves (APW) will be described based on Figures 1(a) to 1(c). The acoustic pulse wave detection sensor 1 of this embodiment consists of a laminated structure of an air pack 1A and a gel pack 1B. The air pack 1A is composed of a three-dimensional knitted fabric (3D net) 10 and a housing film 20 that airtightly houses the three-dimensional knitted fabric (3D net) 10. In the gel pack 1B, a microphone 30 is fixedly placed inside a case 40, and gel 50 is filled around the microphone 30.

[0027] The three-dimensional knitted fabric 10 is formed by joining a pair of ground knitted fabrics, spaced apart from each other, with connecting yarns. Each ground knitted fabric can be formed, for example, from twisted yarn into a flat knitted structure (fine stitches) that is continuous in either the wale direction or the course direction, or into a knitted structure having a honeycomb (hexagonal) mesh. The connecting yarns impart a predetermined rigidity to the three-dimensional knitted fabric so that one ground knitted fabric and the other ground knitted fabric maintain a predetermined distance from each other. Therefore, by applying tension in the planar direction, it becomes possible to cause string vibrations in the yarns of the opposing ground knitted fabrics constituting the three-dimensional knitted fabric, or in the connecting yarns that connect the opposing ground knitted fabrics. As a result, string vibrations are generated by the sound and vibrations of the cardiovascular system, which are biological signals, and these vibrations are propagated in the planar direction of the three-dimensional knitted fabric.

[0028] Various materials can be used as the yarn or connecting yarn that forms the ground fabric of a three-dimensional knitted fabric. Examples include synthetic fibers such as polypropylene, polyester, polyamide, polyacrylonitrile, and rayon, as well as regenerated fibers and natural fibers such as wool, silk, and cotton. These materials may be used individually or in combination as desired. Preferably, the materials are polyester fibers such as polyethylene terephthalate (PET) and polybutylene terephthalate (PBT), polyamide fibers such as nylon 6 and nylon 66, polyolefin fibers such as polyethylene and polypropylene, or combinations of two or more of these fibers. Furthermore, the shape of the ground yarn or connecting yarn is not limited and may be round cross-section yarn, irregular cross-section yarn, hollow yarn, etc. In addition, carbon yarn, metallic yarn, etc., can also be used.

[0029] Examples of usable three-dimensional knitted fabrics include the following: (a) Product number: 49013D (manufactured by Suminoe Textile Co., Ltd.), thickness 10mm Material: Front side ground knit fabric... 2 strands of 450 ds / 108 f polyethylene terephthalate fiber false-twisted yarn The reverse side's ground knit fabric... 2 strands of 450 dsitex / 108f polyethylene terephthalate fiber false-twisted yarn Connecting yarn: 350 ds / 1f polytrimethylene terephthalate monofilament (b) Product number: AKE70042 (manufactured by Asahi Kasei Corporation), thickness 7 mm (c) Product number: T28019C8G (manufactured by Asahi Kasei Corporation), thickness 7mm

[0030] The three-dimensional knitted fabric 10 is covered by a containment film 20. In this embodiment, the containment film 20 consists of two films 21 and 22 made of synthetic resin, which are arranged to cover the front and back surfaces of the three-dimensional knitted fabric 10, and their periphery is fixed by welding or the like. As a result, the three-dimensional knitted fabric 10 is sealed and contained within the containment film 20. When fixing the periphery of the films 21 and 22, it is preferable to fix them so that the films 21 and 22 slightly press the three-dimensional knitted fabric 10 in the thickness direction. This increases the tension of the three-dimensional knitted fabric 10, making it easier for string vibrations to occur in the yarns that make up the three-dimensional knitted fabric 10.

[0031] A case 40 is attached to the outside of the containment film 20, and a microphone 30 is arranged inside the case 40. Inside the case 40, the area around the microphone 30 is filled with gel 50, which acts as a disturbance suppression material. The case 40 is made of synthetic resin and has the function of preventing the diffusion of acoustic vibrations transmitted to the microphone 30 to the outside, and the gel 50 suppresses the microphone 30 from capturing external vibrations. A code 30a that carries the detected acoustic vibration data as an electrical signal is connected to the microphone 30.

[0032] The acoustic pulse wave detection sensor 1 has a configuration that includes a three-dimensional knitted fabric, and can measure acoustic pulse waves (APW) amplified by stochastic resonance. The acoustic pulse wave detection sensor 1 is used by contacting various parts of the body to be measured, such as the back, chest, and waist. Vibrations on the body surface are transmitted to the containment film 20 and the three-dimensional knitted fabric 10 and captured by the microphone 30. It can be used not only by directly attaching it to the skin surface, but also by attaching it to the surface of clothing or the back of a chair, etc. Figure 2 shows an example of attachment when collecting acoustic pulse waves (R-APW) transmitted from the back of a person's chest. On clothing or a backrest, for example, the microphone 30 is positioned about 10 cm to the left of the spine in the back of a person's chest.

[0033] It should be noted that the acoustic pulse wave detection sensor 1 does not include a gel, and instead has a microphone placed within a film together with a three-dimensional knitted fabric. The type (3SR) previously proposed by the applicant can also be used. However, in terms of measurement accuracy, the above-described acoustic pulse wave detection sensor 1 (4SR) is preferred.

[0034] • Blood pressure estimation device Next, a blood pressure estimation device 100, which is equipped with a computer program that processes data obtained from an acoustic pulse wave detection sensor 1, an electrocardiograph, a heart rate monitor, and a fingertip plethysmometer (hereinafter collectively referred to as "biometric indicator sensors 1000"), will be described based on Figures 3 and 4.

[0035] The blood pressure estimation device 100 processes and analyzes time waveform data of biometric indicators acquired by the biometric indicator sensor 1000 to estimate blood pressure. The blood pressure estimation device 100 consists of a computer (including personal computers, microcomputers incorporated into devices, etc.) and receives biosignal data transmitted from the biometric indicator sensor. It then performs predetermined processing using the received time waveform data.

[0036] More specifically, the blood pressure estimation device 100 stores a computer program in its storage unit (including not only the storage medium such as the hard disk built into the computer (blood pressure estimation device 100), but also various removable storage media and storage media of other computers connected via communication means) that executes procedures for the functions of the gradient value calculation unit 200, the type-specific correlation data calculation unit 300, the integrated correlation data calculation unit 400, and the estimated blood pressure calculation unit 500. Furthermore, it has a database 600 that stores the analysis results of the gradient value calculation unit 200, the type-specific correlation data calculation unit 300, and the integrated correlation data calculation unit 400, and is referenced when the estimated blood pressure calculation unit 500 is executed. The database 600 is also stored in storage media such as the hard disk built into the computer constituting the blood pressure estimation device 100, as well as various removable storage media and storage media of other computers connected via communication means. The blood pressure estimation device 100 can also be implemented using an electronic circuit having one or more memory circuits into which a computer program is incorporated that functions as a gradient value calculation unit 200, a type-specific correlation data calculation unit 300, an integrated correlation data calculation unit 400, and an estimated blood pressure calculation unit 500.

[0037] Furthermore, computer programs can be provided by storing them on a recording medium. The recording medium storing the computer program may be a non-transient recording medium. While non-transient recording media are not particularly limited, examples include flexible disks, hard disks, CD-ROMs, MO (magneto-optical disks), DVD-ROMs, and memory cards. It is also possible to transmit and install computer programs to a computer via a communication line.

[0038] As shown in Figure 3, the gradient value calculation unit 200 calculates gradient values ​​for each type of biometric indicator. In this embodiment, three biometric indicators are used: amplitude variation of acoustic pulse wave, variation of heart rate, and variation of pulse wave propagation delay time. Therefore, a first gradient value calculation unit 210, a second gradient value calculation unit 220, and a third gradient value calculation unit 230 are configured.

[0039] The first gradient value calculation unit 210 determines the gradient value from the amplitude fluctuation of the acoustic pulse wave, and as shown in Figure 4, it has a filtering processing unit 2110 and a relative slope waveform calculation unit 2120. More specifically, the filtering processing unit 2110 and the relative slope waveform calculation unit 2120 are constructed as a program shown in steps S501 to S515 in Figure 5 and S601 to S611 in Figure 6.

[0040] The filtering processing unit 2110 is a means of filtering the time waveform data obtained from the acoustic pulse wave detection sensor 1 (acoustic pulse wave (R-APW) amplified by stochastic resonance and preferably taken from the back of the chest) using, for example, a bandpass filter with a center frequency near 20 Hz, preferably a bandpass filter with a frequency band of 10 to 30 Hz. This yields a filtered waveform of 10 to 30 Hz, i.e., amplitude variation (APW to be Filtered-variability: AFV) with a probability distribution centered at 20 Hz. The standard heart rate is around 1 to 1.5 Hz, but as shown in step S601 of Figure 6, the AFV captures waveform components with a relatively large total amplitude with a period of about 1 second, thus revealing the cardiac cycle.

[0041] As shown in Figure 4, the relative slope waveform calculation unit 2120 further includes a first calculation unit 2121, a second calculation unit 2122, a third calculation unit 2123, a fourth calculation unit 2124, a fifth calculation unit 2125, and a sixth calculation unit 2126. The first calculation unit 2121 selects three extreme values ​​in the time waveform data of amplitude fluctuation of the acoustic pulse wave during the time phase from atrial contraction to ventricular contraction, and determines two amplitude points between these extreme values ​​(S501 in Figure 5, S601 in Figure 6). The three extreme values ​​are centered on the extreme value immediately preceding the time phase of the R wave of the electrocardiogram, and consist of this central extreme value and the extreme values ​​before and after it. From these, two amplitude points between adjacent extreme values ​​are determined. The time phase of the R wave of the electrocardiogram can be obtained by measuring electrocardiogram waveform data simultaneously with the acoustic pulse wave detection sensor 1 of the above embodiment, but in that case, it is also necessary to attach an electrocardiograph. Therefore, by using the time waveform data obtained from the acoustic pulse wave detection sensor 1 to capture the phase of the atrioventricular valve closing sound, i.e., the first heart sound, which corresponds to the phase of the R wave in an electrocardiogram, the phase of the R wave can be identified without using an electrocardiograph.

[0042] As a means of capturing the temporal phase of the first heart sound, for example, the method proposed by the applicant in Japanese Patent Application Nos. 2020-180964 and 2020-180963 can be used. This method focuses on determining the boundary frequency (BF) between vibrations caused by apical impulse and vibrations caused by heart sounds from the frequency analysis results of biosignal data (temporal waveform data of bioindicators). By using this boundary frequency, vibrations caused by heart sounds can be extracted from the temporal waveform data of bioindicators, thereby identifying the temporal phase corresponding to the first heart sound.

[0043] The second calculation unit 2122 is a means of using the Lorentz plot method and executes steps S501 to S506 in Figure 5 and S601 to S606 in Figure 6. It performs Lorentz plotting on the amplitude fluctuations of the time waveform of the acoustic pulse wave during the time phase from atrial contraction to ventricular contraction. Specifically, it plots the total amplitudes of two amplitude points identified by the first calculation unit 2121 on the vertical and horizontal axes (S501 in Figure 5, S601 in Figure 6). At this time, the slope of the point cloud plotted over a predetermined measurement time (for example, the total measurement time) is determined as the standard slope (S502 in Figure 5, S602 in Figure 6). Furthermore, the Lorentz plotting method is applied to plot points for a time period shorter than a predetermined measurement time (for example, the total measurement time) (for example, if the total measurement time is 360 seconds, for a measurement time shorter than that (for example, 30 seconds)). Next, the slope of the point cloud plotted during this short time window is determined (short-time point cloud slope), and the difference between this short-time point cloud slope and the standard slope is determined as the relative slope (θi) (S503 in Figure 5, S603 in Figure 6). The short-time point cloud slope is Then, the values ​​are sequentially calculated for each time window corresponding to the short measurement time mentioned above, with a predetermined overlap rate. For example, if the time window is 30 seconds and the overlap rate is 90%, the time windows are set with a 3-second shift in each instance, and the short-time point cloud slope and relative slope (θi) are calculated for each time window. Then, the time waveform of the relative slope (θi) is calculated (S504 in Figure 5, S604 in Figure 6), and further, the time waveform is recalculated using the average value of the relative slope as the reference value (S505 in Figure 5, S605 in Figure 6).

[0044] Furthermore, it is preferable to apply a 0.08Hz low-pass filter to the time waveform of the relative slope. This is because, with a time window of 30 seconds and a 90% overlap rate, one point is plotted every 3 seconds, so the higher frequency is 1 / 3Hz, and the Nyquist frequency is 1 / 6Hz (0.17Hz). For the lower frequency, since the total measurement time is 6 minutes, with a 30-second window and sliding every 3 seconds, 110 points are plotted, and the frequency resolution (Δf) is (1 / 3) / 110 = 0.003Hz. However, since the average of three points (0Hz, 0.003Hz, 0.006Hz) is applied to the frequency axis, 0.006Hz is the minimum value. If we define that a waveform is identified by 5 points, the time required for the waveform is 12 seconds. Therefore, 1 / 12 seconds = 0.08Hz is the cutoff frequency, and waveforms in the frequency band above 0.08Hz can be said to have poor reliability. Therefore, as described above, a 0.08 Hz low-pass filter is applied to the time waveform of the relative slope (S506 in Figure 5, S606 in Figure 6).

[0045] The third calculation unit 2123 performs frequency analysis. Specifically, it performs frequency analysis on the θi function obtained by the second calculation unit 2122 (with a 0.08 Hz low-pass filter applied (time waveform shown in S606 in Figure 6)) (S507 in Figure 5) and displays it on a log-log scale. The analysis waveform, which is the result of the frequency analysis displayed on a log-log scale, is smoothed by applying a moving average of three points (one before and one after the arbitrary point) (S508 in Figure 5, S607 in Figure 6).

[0046] The fourth calculation unit 2124 draws a regression line between 0.006 and 0.08 Hz on the smoothed analysis waveform (θ1-Spectrum) obtained by the third calculation unit 2123. Specifically, this regression line is first automatically drawn with a slope of 1 / f, as shown in Figure 12(a). Next, as shown in Figure 12(b), the standard deviation of the distance between each point of the analysis waveform and the 1 / f regression line is calculated, and a quartic function is calculated for the point cloud of analysis waveforms within a predetermined frequency band range extracted based on this standard deviation. In this embodiment, the point closest to the regression line (the point with the lowest deviation value) near 0.006 Hz and the point closest to the regression line (the point with the lowest deviation value) near 0.08 Hz are extracted, and the area between these two points is extracted as the frequency band for which the approximation curve of the quartic function is calculated (S509 in Figure 5, S608 in Figure 6). Furthermore, the points to be extracted are not limited to those with low standard scores. For example, especially on the low-frequency side, it may be more appropriate to extract points with high standard scores. The fourth calculation unit 2124 is configured to extract the appropriate frequency band depending on how the standard scores are distributed.

[0047] The fifth calculation unit 2125 corresponds to steps S510 in Figure 5 and S609 in Figure 6, and calculates an approximate curve of a quartic function in the frequency band extracted by the fourth calculation unit 2124 for the smoothed analysis waveform obtained by the third calculation unit 2123. The fifth calculation unit 2125 also finds two inflection points from the calculated quartic function.

[0048] The sixth calculation unit 2126 corresponds to the steps from S511 onwards in Figure 5, and calculates the number of extrema of the quartic function obtained by the fifth calculation unit 2125, and outputs the final analysis result of the first gradient value calculation unit 210. Specifically, the sixth calculation unit 2126 creates an increasing / decreasing table and a concavity / convexity table of the quartic function, and the inflection points are Within the target bandwidth Determine whether there are two (S511 in Figure 5). If the determination in S511 in Figure 5 is "Yes", further, the extreme values Within the target bandwidthThe system determines whether there are three extrema within the target band (S512 in Figure 5). If it determines there are three extrema ("Yes" in S12 in Figure 5), it calculates the gradient of inflection (GI) between the two inflection points of the quartic function and outputs this as the analysis result (S513 in Figure 5, S610 in Figure 6). If it determines that there are not three extrema within the target band ("No" in S512 in Figure 5), it further determines whether there are one or two extrema within the target band (S514 in Figure 5). If the result in S514 in Figure 5 is "Yes", it calculates the gradient of tangent (GT) to the inflection point (S515 in Figure 5, S611 in Figure 6) and outputs this as the analysis result.

[0049] If a quartic function has two or fewer extrema, the gradient of the tangent line is determined as follows: If there are two extrema, the choice of which inflection point's tangent line to use is determined by considering whether the inflection point in the low-frequency band is higher or lower than the maximum value, and whether both positive and negative gradients exist for the tangent line. If there is only one extrema and a concavity-concave function exists on either side of the inflection point, the tangent line to the inflection point that is not an extremum is used.

[0050] If the two inflection points of the quartic function do not exist within the target bandwidth (in the case of "No" in the judgment of S511 in Figure 5), and if the extrema of the quartic function does not exist within the target bandwidth (in the case of "No" in the judgment of S514 in Figure 5), in both cases the process returns to step S509, the target bandwidth is reset, the quartic function is calculated again, and the above process is repeated.

[0051] Next, we will explain the case where time waveform data of heart rate obtained from an electrocardiograph, which is a biometric sensor 1000, is used. In this case, the calculation is performed by the second gradient value calculation unit 220. First, the heart rate (HR) is obtained from the RRI of the electrocardiogram (S701 in Figure 7, S801 in Figure 8). Next, the Lorentz plot method is applied, and the (i+1)th heart rate is plotted sequentially on the horizontal axis and the ith heart rate on the vertical axis to create a Lorentz plot diagram for a predetermined measurement time (for example, 360 seconds, which corresponds to the total measurement time), and the standard slope of the plotted point cloud is determined (S702 in Figure 7, S802 in Figure 8). Furthermore, the Lorentz plot method is applied to plot points for a time period shorter than a predetermined measurement time (for example, the total measurement time) (for example, if the total measurement time is 360 seconds, a measurement time shorter than that (for example, 30 seconds)), and the slope of the point cloud plotted for this short time window is determined (short-time point cloud slope). The difference between this short-time point cloud slope and the standard slope is then calculated as the relative slope (θi) (S703 in Figure 7, S803 in Figure 8). The short-time point cloud slope is calculated sequentially for each time window corresponding to a short measurement time with a predetermined overlap rate, similar to the analysis in the first gradient value calculation unit 210 (for example, every 3 seconds if the time window is 30 seconds and the overlap rate is 90%). Then, the short-time point cloud slope and the relative slope (θi) are calculated for each time window, and a time waveform of the relative slope (θi) is created (S704 in Figure 7, S804 in Figure 8).

[0052] Subsequently, similar to the third calculation units 2123 to the sixth calculation units 2126 of the first gradient value calculation unit 210, frequency analysis is performed on the time waveform of the relative slope (θi), a quartic function is defined for the analyzed waveform shown on a log-log axis, and the GT value or GI value as a gradient value related to heart rate variability is determined from the position of the inflection point and the number of extreme values.

[0053] Next, we will explain the case where time waveform data of pulse wave propagation delay time obtained from an electrocardiograph and a fingertip volume plethysmograph, which are used as biometric indicator sensors 1000. In this case, the calculation is performed by the third gradient value calculation unit 230. First, the time waveform data of the electrocardiogram and the fingertip volume plethysmograph are compared to obtain the pulse wave propagation delay time, which is the time from the R wave of the electrocardiogram to the appearance of the peripheral pulse wave (S901 in Figure 9, S1001 in Figure 10). The pulse wave propagation delay time is usually related to changes in blood pressure, as it shortens when blood pressure rises and the blood vessel walls harden, and lengthens when blood pressure falls and the blood vessel walls soften.

[0054] Next, the Lorentz plot method is applied, and the (i+1)th pulse wave propagation delay time is plotted sequentially on the horizontal axis and the i-th pulse wave propagation delay time on the vertical axis to create a Lorentz plot for a predetermined measurement time (for example, 360 seconds, which corresponds to the total measurement time), and the standard slope of the plotted point cloud is determined (S902 in Figure 9, S1002 in Figure 10). Furthermore, the Lorentz plot method is applied to a time shorter than the predetermined measurement time (for example, if the total measurement time is 360 seconds, a measurement time shorter than that (for example, 30 seconds)), and the points are plotted in the same way. The slope of the point cloud plotted for this short time window (short-time point cloud slope) is determined, and the difference between this short-time point cloud slope and the standard slope is determined as the relative slope (θi) (S903 in Figure 9, S1003 in Figure 10). The short-time point cloud slope is determined sequentially for each time window corresponding to a short measurement time with a predetermined overlap rate, similar to the analysis in the first gradient value calculation unit 210 (for example, every 3 seconds if the time window is 30 seconds and the overlap rate is 90%). Then, the short-time point cloud slope and relative slope (θi) are determined for each time window, and a time waveform of the relative slope (θi) is created (S904 in Figure 9, S1004 in Figure 10).

[0055] Subsequently, similar to the third calculation units 2123 to the sixth calculation units 2126 of the first gradient value calculation unit 210, frequency analysis is performed on the time waveform of the relative slope (θi), a quartic function is defined for the analyzed waveform shown on a log-log axis, and the GT value or GI value is obtained as a gradient value relating to the variation in pulse wave propagation delay time based on the position of the inflection point and the number of extrema.

[0056] The type-specific correlation data calculation unit 300 correlates the gradient values ​​(GT values, GI values) for each biometric indicator obtained above with the brachial blood pressure values ​​obtained from the blood pressure monitor (upper arm type) of the subject from whom they were calculated, obtains a correlation diagram as type-specific correlation data, and records it in the first recording unit 610 of the database 600 (see Figure 4). By recording the correlations of multiple subjects, type-specific correlation data between brachial blood pressure values ​​obtained from blood pressure monitors and gradient values ​​(GT values, GI values) is constructed. In this embodiment, three types of correlation data are obtained: those relating to the amplitude variation of the acoustic pulse wave, those relating to the variation of the heart rate, and those relating to the variation of the pulse wave propagation delay time.

[0057] The integrated correlation data calculation unit 400 unfolds the multiple types of type-specific correlation data obtained by the type-specific correlation data calculation unit 300—in this embodiment, the type-specific correlation data relating to amplitude fluctuations of acoustic pulse waves, fluctuations in heart rate, and fluctuations in pulse wave propagation delay time—on the same coordinate system to obtain a correlation diagram, which becomes the integrated correlation data. The integrated correlation data calculation unit 400 then calculates a regression line (regression equation) for the point cloud plotted on this correlation diagram, which is the integrated correlation data. The integrated correlation data calculation unit 400 records the correlation diagram and regression line (regression equation) data obtained in this way as integrated correlation data in the second recording unit 620 of the database 600 (see Figure 4).

[0058] The blood pressure estimation unit 500 estimates the blood pressure of the person whose blood pressure is to be estimated. Specifically, for the person whose blood pressure is to be estimated, time waveform data relating to at least one biometric indicator such as acoustic pulse wave, heart rate, and pulse wave propagation delay time, measured using the acoustic pulse wave detection sensor 1, electrocardiograph, commercially available heart rate monitor, fingertip plethysmometer, etc., is sent to the gradient value calculation unit 200, which receives this information (S1101 in Figure 11). The gradient value calculation unit 200 uses this time waveform data to calculate gradient values ​​(GT value, GI value) for each of the biometric indicators of the person whose blood pressure is to be estimated, namely the acoustic pulse wave, heart rate, and pulse wave propagation delay time (S1102 in Figure 11). The analysis process of the gradient value calculation unit 200 to obtain the gradient values ​​of the person whose blood pressure is to be estimated is the same as the process shown in Figures 4 to 6 for the acoustic pulse wave, the process shown in Figures 7 and 8 for the heart rate, and the process shown in Figures 9 and 10 for the pulse wave propagation delay time.

[0059] The blood pressure estimation unit 500 receives gradient value information (GT value, GI value) from the gradient value calculation unit 200 (S1103 in Figure 11). The blood pressure estimation unit 500 accesses the second recording unit 620 of the database 600 to obtain integrated correlation data (S1104 in Figure 11). The GT value and GI value of the blood pressure estimation target are applied to the integrated correlation data (GT value and GI value are substituted into the regression equation) (S1105 in Figure 11) to obtain the estimated blood pressure value (S1106 in Figure 11).

[0060] In this embodiment, the gradient values ​​(GT and GI values) of the blood pressure estimation target may be calculated using any one of the following bioindicators: heart rate, acoustic pulse wave, or pulse wave delay time. As described above, the integrated correlation data is a combination of multiple types of individual correlation data. Therefore, as shown in the experimental results described later, the coefficient of determination of the regression line (regression equation) of the integrated correlation data is often higher than that of the regression line (regression equation) obtained from each type of individual correlation data. Conversely, even when the coefficient of determination is lower, the difference is small. Therefore, regardless of which bioindicator the gradient values ​​(GT and GI values) are based on, applying them to the integrated correlation data allows for the estimation of blood pressure with generally high estimation accuracy. Thus, the integrated correlation data used in this embodiment has the characteristic of being highly versatile, as it can correspond to multiple types of bioindicators while being a single correlation diagram.

[0061] Among biometric indicators, acoustic pulse waves, for example, are obtained from an acoustic pulse wave detection sensor 1 placed on the seat. While this is not a problem in a static environment, when placed on a car seat, vibrations are constantly input during driving, requiring processing to remove these external vibrations. Although it is certainly possible to obtain highly accurate acoustic pulse wave data by removing external vibrations, this is computationally intensive. Therefore, for biometric indicators such as those used during driving, it is preferable to obtain heart rate information from a wearable heart rate monitor attached to the chest or arm. Such a heart rate monitor allows for relatively high-accuracy measurement even inside a moving car.

[0062] Therefore, in dynamic environments where external noise input is a problem, high-accuracy blood pressure estimation can be achieved by using, for example, gradient values ​​(GT values, GI values) obtained from heart rate data. Thus, the estimation accuracy can be improved by selecting gradient values ​​(GT values, GI values) used in the regression equation of the integrated correlation data that are derived from biometric indicators suitable for the measurement environment.

[0063] Furthermore, it is preferable to obtain the estimated blood pressure value by substituting the average of the gradient values ​​of two or more biometric indicators—in this embodiment, heart rate, acoustic pulse wave, and pulse wave propagation delay time—into the regression equation of the integrated correlation data, as this improves the estimation accuracy.

[0064] In the above explanation, the blood pressure estimation unit 500 estimates blood pressure using integrated correlation data. However, if multiple types of bioindicators are available, the gradient values ​​obtained from each bioindicator can be substituted into the regression equation of the correlation data for each type to obtain estimated blood pressure values ​​for each type, and the average of the obtained multiple estimated blood pressure values ​​for each type can be used as the final output estimated blood pressure value.

[0065] Next, we will describe an experiment conducted using the blood pressure estimation device 100 of this embodiment. [Experimental Example 1: Construction of correlation data using data from inpatients or outpatients] Correlation data was constructed using biometric information from 76 subjects aged 40-95 who were hospitalized or receiving outpatient treatment for some kind of illness. (Calculation of correlation data by type using heart rate variability) Figure 13 shows the correlation between heart rate and systolic blood pressure measured by an upper arm blood pressure monitor for each of the 76 subjects mentioned above. The regression equation for the regression line (arterial pressure regression line (SBP-Regression line: SR)) in this correlation figure was y = -0.1208x + 128.74. Coefficient of determination: R 2 The correlation was extremely low, at 0.0115. Figure 14 shows the correlation between the standard slope of the Lorentz plot (S702 in Figure 7, S802 in Figure 8) and the systolic blood pressure value, calculated using the data from Figure 13 by the second gradient calculation unit 220 for each of the 76 subjects. The regression equation for the arterial pressure regression line (SR) in this correlation figure was y = 1.8558x + 119.7. Coefficient of determination: R 2 The correlation was extremely low, at 0.0021.

[0066] Figure 15 shows the type-specific correlation data for heart rate variability obtained by applying a Lorentz plot to each of the 76 subjects in the second gradient calculation unit 220 to obtain the gradient values ​​(GT value, GI value) of heart rate variability, and then correlating the gradient values ​​with brachial blood pressure values ​​using the type-specific correlation data calculation unit 300. The arterial pressure regression line (SR) in this correlation diagram was y = 15.192x + 126.31. Coefficient of determination: R 2 The value was 0.7001, indicating a high correlation. Regarding heart rate variability information, the correlation with systolic blood pressure was low in Figures 13 and 14, but as shown in Figure 15, it can be seen that the correlation with systolic blood pressure increases when gradient values ​​(GT value, GI value) are used.

[0067] (Calculation of correlation data by type using variations in pulse wave propagation delay time) Figure 16 shows the correlation between pulse wave delay time and systolic blood pressure measured by an upper arm blood pressure monitor for each of the 76 subjects mentioned above. The regression equation for the arterial pressure regression line (SR) in this correlation figure was y = -282.2x + 176.45. Coefficient of determination: R 2 The correlation was low, at 0.162. Figure 17 shows the correlation between the standard slope of the Lorentz plot (S902 in Figure 9, S1002 in Figure 10) and the systolic blood pressure value, calculated using the data from Figure 16 by the third gradient calculation unit 230 for each of the 76 subjects. The regression equation for the arterial pressure regression line (SR) in this correlation figure was y = 17.707x + 112.16. Coefficient of determination: R 2 The value was 0.1075, indicating that the correlation was not very high.

[0068] Figure 18 shows the type-specific correlation data regarding the variation in pulse wave propagation delay time, obtained by applying a Lorentz plot to each of the 76 subjects in the third gradient calculation unit 230 to obtain gradient values ​​(GT value, GI value) of the variation in pulse wave propagation delay time, and then correlating the gradient values ​​with brachial blood pressure values ​​using the type-specific correlation data calculation unit 300. The regression equation for the arterial pressure regression line (SR) in this correlation diagram was y = 14.884x + 126.8. Coefficient of determination: R 2The correlation was 0.5547, indicating a relatively high correlation. Regarding the variation information of pulse wave propagation delay time, the correlation with systolic blood pressure was low in Figures 16 and 17, but as shown in Figure 18, it can be seen that the correlation with systolic blood pressure increases when gradient values ​​(GT value, GI value) are used.

[0069] (Calculation of correlation data by type using amplitude fluctuations of acoustic pulse waves) Figure 19 shows the correlation data for the amplitude variation of acoustic pulse waves obtained by the first gradient value calculation unit 210, which applies a Lorentz plot to the time waveform data obtained from the acoustic pulse wave detection sensor 1 for the 76 subjects mentioned above. The gradient values ​​(GT value, GI value) of the amplitude variation of the acoustic pulse wave are obtained by the type-specific correlation data calculation unit 300, which correlates the gradient values ​​with the brachial blood pressure values. The regression equation for the arterial pressure regression line (SR) in this correlation diagram is y = 21.558x + 138.58. 2 The coefficient of determination was 0.7606, indicating a high correlation.

[0070] (Calculation of integrated correlation data) The integrated correlation data calculation unit 400 plotted all the data constituting the three types of correlation data shown in Figures 15, 18, and 19 on the same coordinate system. The correlation diagram shown in Figure 20 represents the plotted coordinates (integrated correlation data). The regression line of this integrated correlation data (integration-SBP-Regression line (ISR)) is given by the regression equation y = 15.486x + 129.01, with the coefficient of determination being R 2 The result was 0.6121.

[0071] (Estimation of blood pressure by the blood pressure calculation unit 500) For subjects whose measured upper arm blood pressure reading using an upper arm blood pressure monitor was 119 mmHg, which falls within the normal blood pressure range, blood pressure estimation was performed. The gradient value obtained from the subject's heart rate variability was -0.62, the gradient value obtained from the variability in pulse wave propagation delay time was -0.93, and the gradient value obtained from the amplitude variability of the acoustic pulse wave was -0.55.

[0072] First, the estimated blood pressure value obtained by substituting the gradient value of -0.62 obtained from the heart rate variability into x in the regression equation for the type-specific correlation data obtained from the heart rate variability in Figure 15 was 117 mmHg. The estimated blood pressure value obtained by substituting the gradient value of -0.93 obtained from the pulse wave propagation delay time variability into x in the regression equation for the type-specific correlation data obtained from the pulse wave propagation delay time variability in Figure 18 was 113 mmHg. The estimated blood pressure value obtained by substituting the gradient value of -0.55 obtained from the acoustic pulse wave amplitude variability into x in the regression equation for the type-specific correlation data obtained from the acoustic pulse wave amplitude variability in Figure 19 was 127 mmHg.

[0073] On the other hand, when substituting the integrated arterial pressure regression line (ISR) shown in Figure 20, which is the integrated correlation data, into the regression equation, the estimated blood pressure values ​​were 119 mmHg when substituting the gradient value of -0.62 obtained from heart rate variability, 115 mmHg when substituting the gradient value of -0.93 obtained from pulse wave propagation delay time variability, and 121 mmHg when substituting the gradient value of -0.55 obtained from acoustic pulse wave amplitude variability. The maximum difference in estimated blood pressure values ​​was 6 mmHg. When substituting each gradient value into the individual correlation data in Figures 15, 18, and 19, the estimated blood pressure values ​​were a maximum of 14 mmHg (127 mmHg - 113 mmHg). However, when using integrated correlation data, high estimation accuracy can be obtained even with data obtained from different measurement devices such as heart rate, pulse wave propagation delay time, and acoustic pulse wave, demonstrating its versatility and ability to handle multiple types of bioindicators. Therefore, for example, data acquisition by the acoustic pulse wave detection sensor 1 can be performed with high accuracy in a quiet environment. In a dynamic environment, processes such as noise reduction are necessary to obtain high accuracy, but heart rate information can be acquired with relatively high accuracy even in a dynamic environment. Thus, in such an environment, blood pressure can be estimated with high accuracy by fitting the heart rate information to the integrated arterial pressure regression line (ISR) in Figure 20, which is integrated correlation data. In other words, the integrated correlation data of this embodiment allows for blood pressure estimation using easily obtainable biometric indicators that correspond to the measurement environment.

[0074] Also, when calculating the average value of the three gradient values (-0.62, -0.93, -0.55), the result is -0.7. When this average value is substituted into the regression equation x representing the integrated arterial pressure regression line (ISR) in Figure 20, the estimated blood pressure value is 118 mmHg. Therefore, by using the average value of the gradient values obtained from multiple biological indicators for estimation, the result can be closer to the measured value and the estimation accuracy can be improved.

[0075] Also, the estimated blood pressure value (category-specific estimated blood pressure value) obtained by substituting the gradient value -0.62 obtained from the fluctuation of the heart rate into the x of the regression equation of the category-specific correlation data obtained from the fluctuation of the heart rate in Figure 15: 117 mmHg, the gradient value -0.93 obtained from the fluctuation of the pulse wave propagation delay time is substituted into the x of the regression equation of the category-specific correlation data obtained from the fluctuation of the pulse wave propagation delay time in Figure 18, and the estimated blood pressure value (category-specific estimated blood pressure value): 113 mmHg, and the gradient value -0.55 obtained from the amplitude fluctuation of the acoustic pulse wave is substituted into the x of the regression equation of the category-specific correlation data obtained from the amplitude fluctuation of the acoustic pulse wave in Figure 19, and the estimated blood pressure value (category-specific estimated blood pressure value): 127 mmHg. The average of these values is 119 mmHg. Thus, when averaging after obtaining the category-specific estimated blood pressure values, the influence of data with relatively large differences in the coefficient and constant term values of the regression equation among the category-specific correlation data can be more reflected.

[0076] [Experimental Example 2: Construction of Correlation Data Using Biological Information of Healthy Subjects] Correlation data was constructed using the data of 26 healthy subjects in their 20s. (Calculation of Category-Specific Correlation Data Using Heart Rate Fluctuations) Regarding each of the 26 subjects, Figure 21 shows the correlation between the heart rate and the systolic blood pressure value measured by a sphygmomanometer (upper arm type), and Figure 22 shows the correlation between the standard slope of the Lorenz plot obtained by the second gradient value calculation unit 220 and the systolic blood pressure value. The coefficient of determination is R 2 = 0.0386 in Figure 21 and R 2 = 0.0095 in Figure 22, and the correlations were both low. In contrast, the second gradient calculation unit 220 applies a Lorentz plot to obtain the gradient values ​​(GT value, GI value) of heart rate variability, and the type-specific correlation data calculation unit 300 correlates the gradient values ​​with the brachial blood pressure values ​​to obtain type-specific correlation data for heart rate variability. In the correlation diagram shown in Figure 23, the regression equation for the arterial pressure regression line (SR) is y = 6.7468x + 118.72, and the coefficient of determination is R 2 The result was 0.4045, indicating a certain degree of correlation.

[0077] (Calculation of correlation data by type using variations in pulse wave propagation delay time) For each of the 26 subjects mentioned above, Figure 24 shows the correlation between pulse wave propagation delay time and systolic blood pressure values ​​measured by a blood pressure monitor (upper arm type), and Figure 25 shows the correlation between the standard slope of the Lorentz plot obtained by the second gradient calculation unit 220 and the systolic blood pressure values. The coefficient of determination is R in Figure 24. 2 =0.1677, in Figure 25 R 2 The correlation was low in all cases, with a value of 0.0013. In contrast, the second gradient value calculation unit 220 applies a Lorentz plot to obtain gradient values ​​(GT value, GI value) of heart rate variability, and the type-specific correlation data calculation unit 300 correlates the gradient values ​​with brachial blood pressure values ​​to obtain type-specific correlation data regarding pulse wave propagation delay time. In the correlation diagram shown in Figure 26, the regression equation for the arterial pressure regression line (SR) is y = 7.0151x + 120.15, and the coefficient of determination is R 2 The result was 0.438, indicating a certain degree of correlation.

[0078] (Calculation of correlation data by type using amplitude fluctuations of acoustic pulse waves) Figure 27 shows the correlation data for the amplitude variation of acoustic pulse waves obtained by the first gradient calculation unit 210, which applies a Lorentz plot to the time waveform data obtained from the acoustic pulse wave detection sensor 1 for the 26 subjects mentioned above. The gradient values ​​(GT value, GI value) of the amplitude variation of the acoustic pulse wave are obtained by the type-specific correlation data calculation unit 300, which correlates the gradient values ​​with the brachial blood pressure values. The regression equation for the arterial pressure regression line (SR) in this correlation diagram is y = 14.782x + 125.29. 2The coefficient of determination was 0.6184, indicating a high correlation.

[0079] (Calculation of integrated correlation data) The integrated correlation data calculation unit 400 plotted all the data constituting the three types of correlation data shown in Figures 23, 26, and 27 on the same coordinate system. The correlation diagram shown in Figure 28 represents the plotted coordinate system (integrated correlation data). The regression line (integrated arterial pressure regression line (ISR)) of this integrated correlation data is given by the regression equation y = 7.5967x + 119.7, with the coefficient of determination being R 2 The result was 0.4159.

[0080] (Estimation of blood pressure by the blood pressure calculation unit 500) For subjects whose measured upper arm blood pressure reading using an upper arm blood pressure monitor was 112 mmHg, which falls within the normal blood pressure range, blood pressure estimation was performed. The gradient value obtained from the subject's heart rate variability was -0.8, the gradient value obtained from the variability in pulse wave propagation delay time was -0.7, and the gradient value obtained from the amplitude variability of the acoustic pulse wave was -0.92.

[0081] First, the estimated blood pressure value obtained by substituting the gradient value of -0.8 obtained from the heart rate variability into x in the regression equation for the type-specific correlation data obtained from the heart rate variability in Figure 23 was 113 mmHg. The estimated blood pressure value obtained by substituting the gradient value of -0.7 obtained from the pulse wave propagation delay time variability into x in the regression equation for the type-specific correlation data obtained from the pulse wave propagation delay time variability in Figure 26 was 115 mmHg. The estimated blood pressure value obtained by substituting the gradient value of -0.92 obtained from the acoustic pulse wave amplitude variability into x in the regression equation for the type-specific correlation data obtained from the acoustic pulse wave amplitude variability in Figure 27 was 112 mmHg.

[0082] On the other hand, when substituting the integrated arterial pressure regression line (ISR) shown in Figure 28, which is the integrated correlation data, into the x of the regression equation, the estimated blood pressure values ​​were 114 mmHg when substituting the gradient value of -0.8 obtained from heart rate variability, 114 mmHg when substituting the gradient value of -0.7 obtained from pulse wave propagation delay time variability, and 113 mmHg when substituting the gradient value of -0.92 obtained from acoustic pulse wave amplitude variability. The maximum difference in estimated blood pressure values ​​was 1.68 mmHg. When substituting each gradient value into the individual correlation data in Figures 23, 26, and 27, the estimated blood pressure values ​​were a maximum of 3 mmHg (115 mmHg - 112 mmHg). However, even in this experimental example, it can be seen that when using integrated correlation data, high estimation accuracy can be obtained even with data obtained from different measurement devices such as heart rate, pulse wave propagation delay time, and acoustic pulse wave, demonstrating high versatility and the ability to handle multiple types of bioindicators.

[0083] Furthermore, the average of the three gradient values ​​(-0.8, -0.7, -0.92) is -0.8. When this average value is substituted into the regression equation x, which shows the integrated arterial pressure regression line (ISR) in Figure 28, the estimated blood pressure value is 114 mmHg. In this experimental example, there was no substantial difference between using the three gradient values ​​individually and using the average value, but considering the results of Experimental Example 1, it can be said that using the average value tends to bring the result closer to the measured value. On the other hand, the estimated blood pressure values ​​for each type were 113 mmHg, 115 mmHg, and 112 mmHg, respectively. The estimated blood pressure value calculated from the average of these values ​​was 113 mmHg, which is closer to the actual measured blood pressure value.

[0084] [Examination of correlation data by type] Figure 29 contains correlation data for 102 subjects based on heart rate variability, as shown in Figure 15 of Experimental Example 1 and Figure 23 of Experimental Example 2. Figure 30 contains correlation data for 102 subjects based on pulse wave propagation delay time, as shown in Figure 18 of Experimental Example 1 and Figure 26 of Experimental Example 2. Figure 31 contains correlation data for 102 subjects based on acoustic pulse wave amplitude variability, as shown in Figure 19 of Experimental Example 1 and Figure 27 of Experimental Example 2. Figure 32 shows the integrated correlation data obtained by combining the type-specific correlation data from Figures 29 to 31, and also shows the regression lines y1, y2, and y3 from Figures 29 to 31.

[0085] Comparing the three regression lines shown in Figure 32, the slopes and y-intercepts of the regression line y1 for heart rate variability and the regression line y2 for pulse wave propagation delay time are all similar, while the slope and y-intercept of the regression line y3 based on acoustic pulse wave amplitude variation are relatively different. This difference is due to the fact that heart rate variability and pulse wave propagation delay time use periodic components as parameters, while acoustic pulse wave amplitude variation uses amplitude components as parameters. However, despite these regression lines being measured using different instruments and calculated using different parameters, a slope of -1.3 corresponds to a systolic blood pressure of 110 mmHg in all cases. This indicates a high correlation between the slope value (GT value, GI value) and systolic blood pressure, and can be considered one piece of evidence demonstrating the correctness of the systolic blood pressure estimation method using the slope value. Furthermore, the slope of the regression line y3 based on the amplitude fluctuations of the acoustic pulse wave and the y-intercept of 136.58 closely coincide with 140 mmHg, which is the boundary between normal and hypertensive systolic blood pressure. This suggests that the fluctuations in intraventricular pressure captured by the amplitude fluctuations of the acoustic pulse wave are roughly centered around 140 mmHg.

[0086] [Estimating blood pressure in hypertensive subjects] Sleep experiments were conducted on subjects diagnosed with hypertension and gout, and suspected of having myocardial infarction, cerebral infarction, and aortic stenosis. Blood pressure was estimated using the analyzed data. During the experiment, the subjects' electrocardiograms were measured, and heart rate was determined from the RRI of the electrocardiogram (Figure 33(a)). Autonomic nervous system activity was captured by wavelet analysis of the RRI (Figure 33(b)). Furthermore, the acoustic pulse wave detection sensor 1 was placed in contact with the left side of the back of the chest (left-side sensor) and the head (head sensor), and the time-series waveform of the frequency was obtained from the data obtained from each using zero-crossing points, as described in Japanese Patent Application Publication No. 2016-112144 by the present applicant. The time-series waveform of the frequency slope was then obtained by sliding calculation of this time-series waveform of frequency (Figures 33(c), (e)). In addition, from the time-series waveform of the frequency slope, the following frequency components were extracted as frequency components indicating fluctuations of the cardiovascular system: a functional adjustment signal with a frequency lower than the frequency at which the characteristics of fluctuations of the cardiovascular system switch (0.0017 Hz), a fatigue reception signal with a frequency higher than the functional adjustment signal (0.0035 Hz), and an activity adjustment signal with an even higher frequency than the fatigue reception signal (0.0053 Hz). The distribution rate of each of these frequency components was then determined in time series (Figures 33(d), (f)).

[0087] The sleep experiment was conducted in a supine position. The subject was fatigued and fell asleep quickly. As shown in Figure 33(a), heart rate variability was small, at 60-65 / min, for the first 24 minutes (approximately 1440 seconds), and then fluctuated relatively large, at approximately 55-70 / min, for the following 12 minutes. The subject woke up approximately 36 minutes (approximately 2160 seconds) after the start of the experiment, and data was continued to be measured using an electrocardiograph and acoustic pulse wave detection sensor 1 for 6 minutes while sitting. In the sitting position, the heart rate was relatively stable at approximately 55-60 / min. This pattern of heart rate variability can also be interpreted as sympathetic nervous system stimulation. The sympathetic nervous system becomes slightly dominant after 1440 seconds, and the LF / HF index spikes at approximately 2100-2200 seconds (approximately 35-37 minutes), indicating increased sympathetic nervous system activity and wakefulness.

[0088] The frequency slope and frequency slope distribution waveforms in Figures 33(c) to (f) also show similar characteristic changes in the time periods corresponding to those described above.

[0089] Figures 34(a) to (c) are Lorentz plots relating to heart rate variability, and Figures 34(d) to (f) are Lorentz plots relating to the amplitude variability of acoustic pulse waves. As shown in Figures 34(d) and (e), in this subject, in addition to a large cluster of plots concentrated around 0.01 (V) on the horizontal axis, there is a smaller cluster of plots around the value 0 (v) on the horizontal axis. This is thought to be a phenomenon that appears based on the presence of cardiovascular disease, and since this phenomenon is not considered if the smaller cluster is excluded, it is considered appropriate to calculate the regression line considering both clusters.

[0090] Figures 35(a) to (d) show the gradient values ​​(GT and GI values) obtained using the regression lines of the Lorenz plots shown in Figures 34(a) to (f). Figures 35(a) and (b) show the gradient values ​​calculated using data from 1500 to 1860 seconds of sleep, while Figures 35(c) and (d) show the gradient values ​​calculated using data from 6 minutes of sitting posture after waking.

[0091] Figure 35(e) shows the regression equation of the regression line for the integrated correlation data in Figure 32: y = 14.6 + 128.19, R 2 The systolic blood pressure estimated by substituting the respective gradient values ​​from Figures 35(a) to (d) into =0.6076 is shown. The actual measured systolic blood pressure of this subject, measured using an upper arm blood pressure monitor, was 142 mmHg during sleep (1500-1860 seconds) and 150 mmHg in a seated position.

[0092] The estimated blood pressure during sleep (supine position) calculated from the gradient value of the amplitude variation (AF) of the acoustic pulse wave (Figure 35(a)) of 0.3554 was 133 mmHg, and the estimated blood pressure during sleep (supine position) calculated from the gradient value of the heart rate variation (Figure 35(b)) of 0.4347 was 135 mmHg. The estimated blood pressure value while awake (sitting) was 144 mmHg, calculated from the gradient value of the acoustic pulse wave amplitude variation (AF) (1.0575) in Figure 35(c), and the estimated blood pressure value while awake (sitting) was 144 mmHg, calculated from the gradient value of the heart rate variation (1.069) in Figure 35(d).

[0093] Therefore, the estimation of systolic blood pressure using gradient values ​​(GT values, GI values) and the integrated correlation data shown in Figure 32 accurately captures the blood pressure of hypertensive subjects during sleep, and also accurately captures the phenomenon of increased blood pressure due to elevated sympathetic nervous system activity after wakefulness. Furthermore, by applying gradient values ​​to the integrated correlation data shown in Figure 32, it can be seen that the estimated blood pressure values ​​closely match those obtained from either heart rate variability or acoustic pulse wave amplitude variability.

[0094] Here, Figure 36 shows the correlation data by type between the gradient values ​​(GT and GI values) of acoustic pulse wave amplitude variation (AF) and systolic blood pressure values ​​(measured values) for 21 healthy subjects aged 21-65 years (average age: 43 years), including 2 women (number of analysis cases: 398). The regression equation for this regression line is y = 18.031x + 136.14, and the coefficient of determination is R 2 The result was 0.8167. On the other hand, the correlation data between the gradient values ​​(GT values, GI values) of the amplitude fluctuations of acoustic pulse waves and systolic blood pressure values ​​for each type of subject in Experimental Example 1, "76 subjects aged 40-95 years who were hospitalized or receiving outpatient treatment for some disease," is shown in Figure 19, and the regression equation was y = 21.558x + 138.58. Furthermore, the correlation data between the gradient values ​​(GT values, GI values) of the amplitude fluctuations of acoustic pulse waves and systolic blood pressure values ​​for 26 healthy subjects in their 20s in Experimental Example 2 are shown in Figure 27, and the regression equation was y = 14.782x + 125.29.

[0095] These results show that in the group of healthy subjects in their 20s, the coefficient of x in the regression equation was 14.782 (Figure 27), in the group of healthy subjects with an average age of 43, the coefficient of x in the regression equation was 18.031 (Figure 36), and in the group of elderly subjects aged 40 to 95 with diseases, the coefficient of x in the regression equation was 21.558 (Figure 19). This confirms a tendency for the slope to be smaller in younger groups and larger in older individuals and those with diseases. This is thought to be related to the fact that with age, the elasticity of blood vessels is lost, and hormonal regulation makes blood vessels more prone to constriction, making them more sensitive to the contractile force of the heart. Therefore, the regression equation of the type-specific correlation data obtained in relation to the gradient values ​​(GT value, GI value) in this embodiment can also be used as an indicator to estimate aging and health status. Furthermore, Figure 37 is a type-based correlation data that combines all the data from Figures 19, 27, and 36 into a single coordinate system. A larger number of cases increases the reliability of blood pressure estimation. [Explanation of symbols]

[0096] 1. Acoustic pulse wave detection sensor 100 Blood pressure estimation device 200 Gradient Value Calculation Unit 210 First gradient value calculation unit 220 Second Gradient Value Calculation Unit 230 Third Gradient Value Calculation Unit 300-type correlation data calculation unit 400 Integrated Correlation Data Calculation Unit 500 Estimated blood pressure calculation unit 600 databases 610 First Record Section 620 Second Records Section 1000 Biometric Sensors

Claims

1. A gradient value calculation unit analyzes time waveform data of multiple types of biometric indicators related to cardiovascular responses, obtains each analysis waveform on a log-log axis, defines a quartic function for the analysis waveform, finds two inflection points from the quartic function, outputs the gradient value between the two inflection points as an analysis result if it is determined that there are three extrema of the quartic function, and outputs the gradient value of the tangent line at either inflection point as an analysis result if it is determined that there is one or two extrema of the quartic function. A type-specific correlation data calculation unit calculates type-specific correlation data by correlating the gradient value, which is the result of the analysis, with the measured upper arm blood pressure value, according to the type of biometric indicator. An integrated correlation data calculation unit that uses the time waveform data obtained from multiple subjects, unfolds the multiple type-specific correlation data obtained for each type of biometric on the same coordinate system, and calculates integrated correlation data, For individuals whose blood pressure is to be estimated, the estimated blood pressure calculation unit analyzes the time waveform data of at least one of the multiple types of biometric indicators using the gradient value calculation unit, obtains a gradient value as a result of the analysis, and substitutes this gradient value into the regression equation of the integrated correlation data to obtain an estimated blood pressure value corresponding to the brachial blood pressure value. A blood pressure estimation device having the following features.

2. The aforementioned estimated blood pressure calculation unit, The blood pressure estimation device according to claim 1, wherein, for the person whose blood pressure is to be estimated, the gradient value calculation unit analyzes the time waveform data of two or more biological indicators from among the multiple types of biological indicators, obtains the gradient value which is the result of the analysis for each biological indicator, and substitutes the average value of the gradient values ​​obtained for each biological indicator into the regression equation of the integrated correlation data to obtain an estimated blood pressure value corresponding to the brachial blood pressure value.

3. The aforementioned estimated blood pressure calculation unit, The blood pressure estimation device according to claim 1, wherein, for the person whose blood pressure is to be estimated, the gradient value calculation unit analyzes the time waveform data of two or more biological indicators from among the multiple types of biological indicators, obtains the gradient value which is the result of the analysis for each biological indicator, substitutes the gradient value obtained for each biological indicator into the regression equation of the corresponding type-specific correlation data to obtain type-specific estimated blood pressure values, and outputs the average value of the multiple type-specific estimated blood pressure values ​​obtained as an estimated blood pressure value corresponding to the brachial blood pressure value.

4. The blood pressure estimation device according to claim 1, wherein the plurality of types of biological signals are selected from variations in heart rate, variations in pulse wave propagation delay time, and variations in the amplitude of acoustic pulse waves obtained via the body surface.

5. The gradient value calculation unit, For each of the aforementioned multiple types of biometric indicators, the Lorentz plot method is applied to determine the standard slope from the Lorentz plot in a reference time range. The relative slope, which is the difference between the slope of each Lorentz plot created at predetermined time intervals shorter than the reference time range and the standard slope, is successively determined with a predetermined overlap rate and a predetermined time window. The time waveform of the obtained relative slope is filtered at a predetermined frequency. Frequency analysis is performed on the filtered time waveform. The results of the frequency analysis are converted into an analysis waveform shown on a log-log axis. A regression line with a slope of 1 / f is applied to the analysis waveform. The standard deviation of the distance of each point constituting the analysis waveform with respect to the regression line is determined. The quartic function is calculated for the point cloud of analysis waveforms in a predetermined frequency band range extracted based on this standard deviation. The blood pressure estimation device according to claim 1.

6. The gradient value calculation unit, When using the "amplitude fluctuation of acoustic pulse waves obtained via the body surface" as the bioindicator, the amplitude between two extreme values ​​is determined from the extreme values ​​included in the time phases of atrial and ventricular contraction, the Lorentz plot method is applied, the standard slope is determined from the Lorentz plot in a reference time range, the relative slope, which is the difference between the slope of each Lorentz plot created at predetermined time intervals shorter than the reference time range and the standard slope, is successively determined with a predetermined overlap rate and a predetermined time window, the time waveform of the obtained relative slope is filtered at a predetermined frequency, frequency analysis is performed on the filtered time waveform, the result of the frequency analysis is converted into an analysis waveform shown on a log-log axis, a regression line with a slope of 1 / f is applied to the analysis waveform, the standard deviation of the distance of each point constituting the analysis waveform with respect to the regression line is determined, and the quartic function is determined for the point cloud of analysis waveforms in a predetermined frequency band range extracted based on this standard deviation. The blood pressure estimation device according to claim 4.

7. The procedure involves analyzing time waveform data of multiple types of biometric indicators related to cardiovascular responses, obtaining each analysis waveform on a log-log scale, defining a quartic function for the analysis waveform, finding two inflection points from the quartic function, outputting the gradient value between the two inflection points as the analysis result if the quartic function is determined to have three extrema, and outputting the gradient value of the tangent line at either inflection point as the analysis result if the quartic function is determined to have one or two extrema. A procedure for obtaining type-specific correlation data by correlating the aforementioned gradient value, which is the result of the analysis, with the measured brachial blood pressure value, according to the type of biometric indicator, A procedure for obtaining integrated correlation data by using the time waveform data obtained from multiple subjects, unfolding the multiple correlation data for each type of biometric on the same coordinate system, and For individuals whose blood pressure is to be estimated, the procedure involves analyzing the time waveform data of at least one of the multiple types of biometric indicators, obtaining the gradient value resulting from the analysis, and substituting this gradient value into the regression equation of the integrated correlation data to obtain an estimated blood pressure value corresponding to the brachial blood pressure value. A computer program that executes a command to make a computer function as a blood pressure estimation device.

8. In the procedure for determining the estimated blood pressure value, The computer program according to claim 7, which analyzes the time waveform data of two or more of the multiple types of biological indicators for the person whose blood pressure is to be estimated, obtains the gradient value which is the result of the analysis for each biological indicator, and substitutes the average value of the gradient values ​​obtained for each biological indicator into the regression equation of the integrated correlation data to obtain an estimated blood pressure value corresponding to the brachial blood pressure value.

9. In the procedure for determining the estimated blood pressure value, The computer program according to claim 7, wherein for the person whose blood pressure is to be estimated, time waveform data of two or more of the multiple types of biological indicators are analyzed, the gradient value which is the result of the analysis is obtained for each biological indicator, the gradient value obtained for each biological indicator is substituted into the regression equation of the corresponding type-specific correlation data to obtain type-specific estimated blood pressure values, and the average value of the multiple type-specific estimated blood pressure values ​​obtained is output as an estimated blood pressure value corresponding to the brachial blood pressure value.

10. The computer program according to claim 7, wherein the plurality of types of biological signals are selected from variations in heart rate, variations in pulse wave propagation delay time, and variations in the amplitude of acoustic pulse waves obtained via the body surface.

11. In the procedure for determining the gradient value, For each of the aforementioned multiple types of biometric indicators, the Lorentz plot method is applied to determine the standard slope from the Lorentz plot in a reference time range. The relative slope, which is the difference between the slope of each Lorentz plot created at predetermined time intervals shorter than the reference time range and the standard slope, is successively determined with a predetermined overlap rate and a predetermined time window. The time waveform of the obtained relative slope is filtered at a predetermined frequency. Frequency analysis is performed on the filtered time waveform. The results of the frequency analysis are converted into an analysis waveform shown on a log-log axis. A regression line with a slope of 1 / f is applied to the analysis waveform. The standard deviation of the distance of each point constituting the analysis waveform with respect to the regression line is determined. The quartic function is calculated for the point cloud of analysis waveforms in a predetermined frequency band range extracted based on this standard deviation. The computer program according to claim 7.

12. In the procedure for determining the gradient value, When using the "amplitude fluctuation of acoustic pulse waves obtained via the body surface" as the bioindicator, the amplitude between two extreme values ​​is determined from the extreme values ​​included in the time phases of atrial and ventricular contraction, the Lorentz plot method is applied, the standard slope is determined from the Lorentz plot in a reference time range, the relative slope, which is the difference between the slope of each Lorentz plot created at predetermined time intervals shorter than the reference time range and the standard slope, is successively determined with a predetermined overlap rate and a predetermined time window, the time waveform of the obtained relative slope is filtered at a predetermined frequency, frequency analysis is performed on the filtered time waveform, the result of the frequency analysis is converted into an analysis waveform shown on a log-log axis, a regression line with a slope of 1 / f is applied to the analysis waveform, the standard deviation of the distance of each point constituting the analysis waveform with respect to the regression line is determined, and the quartic function is determined for the point cloud of analysis waveforms in a predetermined frequency band range extracted based on this standard deviation. The computer program according to claim 10.

13. A computer-readable recording medium on which a computer program according to any one of claims 7 to 12 is recorded.

Images