Blood pressure estimation device, computer program, and storage medium

Abstract

To estimate blood pressure with higher accuracy.　Time-series data of an instantaneous heart rate of a subject group is analyzed, a first circulation dynamic index related to a long-term blood pressure adjustment mechanism and a second circulation dynamic index related to a short-term blood pressure adjustment mechanism are obtained in an arterial pressure-related frequency band, a database of a three-dimensional response curbed surface model in which the first circulation dynamic index and the second circulation dynamic index are used as explanatory variables and actual measurement data of the systolic blood pressure of each subject group is used as an objective variable is constructed, and the systolic blood pressure of a blood pressure estimation target subject is estimated by applying an estimation first circulation dynamic index and an estimation second circulation dynamic index of the blood pressure estimation target subject to the three-dimensional response curved surface model.

Description

Blood pressure estimation device, computer program, and recording medium

[0001] The present invention relates to a technique for estimating blood pressure based on information about heart rate.

[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 vibration information within the body caused by blood flow fluctuations from the ventricular filling period to the isovolumic contraction period, obtains an index showing the fluctuations by applying the Lorenz plot method, and estimates blood pressure values ​​from the index (Patent Document 1).

[0003] Specifically, this technology captures biosignals (acoustic pulse waves (APW)) from the surface of the body on the back using a biosignal detection sensor, filters the time-series waveform in a predetermined frequency band, and obtains a filtered waveform that reveals the cardiac cycle. This filtered waveform is compared with electrocardiogram waveform data obtained from an electrocardiograph at the same time, and the waveform components from the ventricular filling period to the isovolumic contraction period are identified. Then, the Lorenz plot method is applied, and using the slope of a large number of point clouds plotted at a predetermined measurement time as a reference, a new time series is constructed for the angle difference with the slope of point clouds at shorter measurement times. The time-series waveform is frequency-analyzed, and the frequency analysis results are displayed on a double logarithmic axis to find a regression line for the range from VLF to LF in the analyzed waveform. The slope of the regression line is defined as the fractal slope, and blood pressure is estimated from the fractal slope.

[0004] Patent Document 2 focuses on the fact that pulse wave velocity (PWV) depends on blood pressure and the elastic properties of arteries, and discloses a technique for measuring blood pressure by assuming that blood pressure is proportional to PWV. It also describes that PWV is calculated from PAT (Pulse Arrival Time). Because PWV is inversely proportional to the derivative of PAT, blood pressure is estimated by calculating the derivative value of PAT. It also describes measuring heart rate and oxygen saturation using an optical sensor used in mobile computers such as mobile phones.

[0005] Both Patent Documents 1 and 2 have in common the point that they are techniques capable of estimating blood pressure without using a cuff-based sphygmomanometer, i.e., an upper arm sphygmomanometer.

[0006] JP 2019-122502 A

[0007] Guyton Physiology, 13th edition, published March 20, 2018, author John E. Hall, general translation: Yoshihiro Ishikawa et al., publisher: Elsevier Japan Co., Ltd.

[0008] In Patent Document 1, the fractal slope is calculated as an index showing the fluctuation of heart rate variability. However, arterial pressure is not regulated by a single pressure control system, but by an integrated, multifaceted system that is interrelated. However, the fractal slope is an index that shows the characteristics of the analyzed waveform using only a single regression line, and it is difficult to say that it fully reflects the characteristics of the pressure control system for arterial pressure.

[0009] In Patent Document 2, blood pressure is calculated by detecting pulse wave velocity. Although pulse wave velocity is said to be correlated with blood pressure, the degree of correlation is not very high. Therefore, even if pulse wave velocity or pulse wave propagation delay time, which is used as a substitute for pulse wave velocity, is simply estimated by comparing it with blood pressure, there are concerns about the accuracy.

[0010] In view of the above, an object of the present invention is to provide a technique that can determine blood pressure (particularly systolic blood pressure) with higher accuracy.

[0011] To solve these problems, the present invention focuses on the following points. First, as described above, arterial pressure is not regulated by a single pressure control system but by an integrated, multifaceted system that is interconnected. Specifically, arterial pressure is determined by the cooperative action (Interaction with Additional Systems: IAAS) of the renin-angiotensin-aldosterone system, i.e., the kidney-body fluid system, which constitutes the long-term blood pressure regulation mechanism (LTM), and the neural regulatory mechanism responsible for short-term blood pressure regulation (STM). For example, baroreceptor resetting, a short-term regulation, affects renal sympathetic nerve activity and is therefore involved in long-term blood pressure regulation. A slight increase in cardiac output leads to an increase in total peripheral vascular resistance through some form of autoregulation, resulting in an increase in blood pressure. Therefore, we focused on the theory proposed by Guyton (see Non-Patent Document 1), which states that "if pulse pressure control works sufficiently strongly and there is a delay between the excitation of the feedback function and the subsequent arterial pressure response, the feedback gain becomes large, causing a delay in the reaction time of the pressure control mechanism and generating fluctuations in the low-frequency band. In other words, arterial pressure oscillates due to the excitation of the feedback function of the pressure control mechanism, and operates on the same principle as mechanical and electrical systems."

[0012] Next, we focused on chaotic time series analysis using a delayed coordinate system as a method for capturing reaction time delays (phase information), and with regard to fluctuations in the low-frequency band, we focused on the relationship between global and local fluctuations used in stochastic differential equations, as well as the fact that fluctuations in the arterial pressure regulation mechanism are reflected in fluctuations in heart rate variability.

[0013] To facilitate consideration of the mechanical and electrical system, the period of the arterial pressure regulation mechanism was converted to frequency. Frequency-related regulatory factors include the fundamental period of arterial pressure fluctuations (60 seconds, 1 / 60 ≒ 0.017 Hz) described in Non-Patent Document 1 and the baroreflex recovery time (21 seconds, 1 / 21 ≒ 0.0476 Hz) described in Guyton's previous study. The fundamental period of arterial pressure fluctuations is 0.017 Hz when linearity is maintained. When nonlinearity increases in hypotension or hypertension, harmonic components of 0.017 Hz, such as 0.034 Hz, 0.051 Hz, 0.068 Hz, 0.085 Hz, 0.102 Hz, etc., are generated. Furthermore, while the frequency of the LF component is generally defined as 0.04 to 0.15 Hz, actual measurements have shown that it is primarily distributed between 0.06 and 0.11 Hz. Therefore, the operating band for short-term blood pressure regulation (STM) was set to the frequency band up to 0.06 Hz, which includes the fundamental frequency of arterial pressure fluctuations of 0.017 Hz, its harmonic components of 0.034 Hz ​​and 0.051 Hz, and 0.0476 Hz related to the baroreceptor reflex.

[0014] On the other hand, frequencies below 0.017 Hz are the frequency band in which aldosterone functions, and are the operating band of the long-term blood pressure regulation mechanism (LTM). When arterial pressure drops, the baroreceptor reflex activates the sympathetic nervous system, followed by increased aldosterone secretion within a few minutes (1 / 60 ≒ 0.017 Hz to 1 / 180 ≒ 0.0056 Hz), which modifies the pressure control characteristics of the renal-humoral system. Fluctuations in the chemical reaction of aldosterone in the renal-humoral system are reflected in the frequency gradient of the frequency band below 0.017 Hz.

[0015] For these reasons, fluctuations in the long-term blood pressure regulation mechanism (LTM) will be examined mainly in the frequency band of 0.006 to 0.017 Hz, and fluctuations in the short-term blood pressure regulation mechanism (STM) will be examined mainly in the frequency band of 0.017 to 0.06 Hz. However, because this operating band is not constant due to individual differences, the frequency bands examined will be broadened as shown in Figure 9, i.e., LTM will be examined between 0.006 Hz and IP2 (IP2 is 0.017 Hz or higher), and STM will be examined between IP1 and 0.06 Hz (IP1 is 0.017 Hz or lower).

[0016] That is, the blood pressure estimation device of the present invention comprises: a database construction unit that analyzes time-series data of instantaneous heart rates of a predetermined number of subjects, determines a first circulatory dynamics index associated with the long-term blood pressure regulation mechanism (LTM) and a second circulatory dynamics index associated with the short-term blood pressure regulation mechanism (STM) in an arterial pressure-related frequency band, and constructs a database of a three-dimensional response surface model using the first circulatory dynamics index and the second circulatory dynamics index as explanatory variables and the measured systolic blood pressure data of each of the subjects as a response variable; an estimation index calculation unit that analyzes time-series data of instantaneous heart rates of subjects whose blood pressure is to be estimated, determines the first circulatory dynamics index and the second circulatory dynamics index of each subject whose blood pressure is to be estimated, and sets these as a first estimation circulatory dynamics index and a second estimation circulatory dynamics index, respectively; and a systolic blood pressure estimation unit that applies the first estimation circulatory dynamics index and the second estimation circulatory dynamics index to the three-dimensional response surface model constructed by the database construction unit, and estimates the systolic blood pressure of each subject whose blood pressure is to be estimated.

[0017] The database construction unit reconstructs a two-dimensional delay coordinate system by applying a Lorenz plot to the time series data of the instantaneous heart rate of the predetermined number of subjects, creates an attractor in a reference time range, and calculates reference angle information (θ ０ ), and an individual attractor in the two-dimensional delay coordinate system in an individual time range shorter than the reference time range is created, and individual angle information (θ ｉ ) and the difference (θ ｉ－０ ) is calculated sequentially while moving the individual time range, and the difference (θ ｉ－０ a time series data analysis unit for obtaining time series data of the difference (θ ｉ－０a frequency analysis unit that performs frequency analysis on time series data of the arterial pressure-related frequency band and displays a spectral diagram on a double logarithmic graph; an approximation curve calculation unit that sets two types of approximation curves, a quartic function and a quintic function, for the arterial pressure-related frequency band in the spectral diagram; a best approximation curve selection unit that selects the best approximation curve to be used for analysis from the two types of approximation curves, the quartic function and the quintic function; an inflection point / intersection extraction unit that extracts inflection points and tangent intersections for the best approximation curve; and an approximation gradient curve setting unit that sets two types of approximation curves, a cubic function and a quartic function, as approximate gradient curves for each frequency band within the arterial pressure-related frequency band, which corresponds to all combinations of the frequencies of the inflection points, the frequencies of the tangent intersections, and the two frequencies selected from the lower and upper limits of the arterial pressure-related frequency band, in each of the operating bands of the long-term blood pressure regulation mechanism and the short-term blood pressure regulation mechanism. It is preferable that the configuration include an optimized approximate gradient curve selection unit that selects an optimized approximate gradient curve to be used for analysis from the two types of approximate gradient curves, the cubic function and the quartic function, for each of the operating bands of the long-term blood pressure regulation mechanism and the short-term blood pressure regulation mechanism; a circulatory dynamics index determination unit that sets the value of the frequency gradient of the optimized approximate gradient curve for the operating band of the long-term blood pressure regulation mechanism as the first circulatory dynamics index and the value of the frequency gradient of the optimized approximate gradient curve for the operating band of the short-term blood pressure regulation mechanism as the second circulatory dynamics index; and a three-dimensional response surface model creation unit that creates the three-dimensional response surface model using the first circulatory dynamics index and the second circulatory dynamics index.

[0018] It is preferable that the estimation index calculation unit analyzes time series data of the instantaneous heart rate of the subject whose blood pressure is to be estimated using the time series data analysis unit, the frequency analysis unit, the approximation curve calculation unit, the best approximation curve selection unit, the inflection point / intersection extraction unit, the approximation gradient curve setting unit, the optimized approximation gradient curve selection unit, and the circulatory dynamics index determination unit to obtain a first circulatory dynamics index and a second circulatory dynamics index of the subject whose blood pressure is to be estimated, and set these as the first circulatory dynamics index for estimation and the second circulatory dynamics index for estimation, respectively.

[0019] Preferably, when the optimization approximate curve is an approximate gradient curve represented by the cubic function, the circulatory dynamics index determination unit determines the gradient of the tangent at the inflection point as the frequency gradient, and when the optimization approximate curve is an approximate gradient curve represented by the quartic function, the circulatory dynamics index determination unit determines the gradient of the bitangent as the frequency gradient.

[0020] The best approximation curve selection unit is preferably configured to select the best approximation curve using at least one of the following evaluation indices regarding the residuals between each of the two types of approximation curves, the quartic function and the quintic function, and the spectrogram: standard deviation of the residuals, average value of the absolute values ​​of the residuals, best approximation error which is the difference between the standard deviation of the residuals and the average value of the absolute values ​​of the residuals, and distribution shape of the residuals.The optimized approximation gradient curve selection unit is preferably configured to select the optimized approximation gradient curve using at least one of the following evaluation indices regarding the residuals between each of the two types of approximation curves, the cubic function and the quartic function, and the spectrogram: standard deviation of the residuals, average value of the absolute values ​​of the residuals, best approximation error which is the difference between the standard deviation of the residuals and the average value of the absolute values ​​of the residuals, distribution shape of the residuals, and correlation between the standard deviation of the residuals and the best approximation error. The optimized approximate gradient curve selection unit is preferably configured to determine whether an overlap band (OLF) modified by both the long-term blood pressure regulation mechanism (LTM) and the short-term blood pressure regulation mechanism (STM) exists at a predetermined position in the selected optimized approximate gradient curve, and if an overlap band (OLF) does not exist, to exclude outliers included in the long-term blood pressure regulation mechanism (LTM) in the spectrogram and perform recalculation.

[0021] The computer program of the present invention is a computer program that causes a computer to function as a blood pressure estimation device, and causes the computer to execute the following steps: analyzing time-series data of instantaneous heart rates of a predetermined number of subjects, determining a first circulatory dynamics index associated with the long-term blood pressure regulation mechanism (LTM) and a second circulatory dynamics index associated with the short-term blood pressure regulation mechanism (STM) in an arterial pressure-related frequency band, and constructing a database of a three-dimensional response surface model using the first circulatory dynamics index and the second circulatory dynamics index as explanatory variables and the measured data of the systolic blood pressure of each of the subjects as a response variable; analyzing time-series data of instantaneous heart rates of subjects whose blood pressure is to be estimated, determining the first circulatory dynamics index and the second circulatory dynamics index of each subject whose blood pressure is to be estimated, and setting them as a first circulatory dynamics index for estimation and a second circulatory dynamics index for estimation, respectively; and applying the first circulatory dynamics index and the second circulatory dynamics index for estimation to the three-dimensional response surface model, thereby estimating the systolic blood pressure of each subject whose blood pressure is to be estimated.

[0022] In the procedure for constructing the database, a two-dimensional delay coordinate system is reconstructed by applying a Lorenz plot to the time series data of the instantaneous heart rate of the predetermined number of subjects, an attractor in a reference time range is created, and reference angle information (θ ０ ), and an individual attractor in the two-dimensional delay coordinate system in an individual time range shorter than the reference time range is created, and individual angle information (θ ｉ ) and the difference (θ ｉ－０ ) is calculated sequentially while moving the individual time range, and the difference (θ ｉ－０ ) time series data, and the difference (θ ｉ－０) and displaying a spectral diagram on a double logarithmic graph; setting two types of approximation curves, a quartic function and a quintic function, for the arterial pressure-related frequency band in the spectral diagram; selecting the best approximation curve to be used for analysis from the two types of approximation curves, the quartic function and the quintic function; extracting inflection points and tangent intersections for the best approximation curve; setting two types of approximation curves, a cubic function and a quartic function, as approximation gradient curves for each frequency band within the arterial pressure-related frequency band, corresponding to all combinations of the frequencies of the inflection points, the frequencies of the tangent intersections, and two frequencies selected from the lower and upper limits of the arterial pressure-related frequency band, in each of the operating bands of the long-term blood pressure regulation mechanism and the short-term blood pressure regulation mechanism; selecting an optimized approximation gradient curve to be used for analysis from the two types of approximation gradient curves, the cubic function and the quartic function, in each of the operating bands of the long-term blood pressure regulation mechanism and the short-term blood pressure regulation mechanism. It is preferable to execute the steps of: setting the value of the frequency gradient of the optimized approximate gradient curve for the operating band of the long-term blood pressure regulation mechanism as the first circulatory dynamics index; and setting the value of the frequency gradient of the optimized approximate gradient curve for the operating band of the short-term blood pressure regulation mechanism as the second circulatory dynamics index; and creating the three-dimensional response surface model using the first circulatory dynamics index and the second circulatory dynamics index.

[0023] In the procedure for obtaining the first estimated circulatory dynamics index and the second estimated circulatory dynamics index of the subject whose blood pressure is to be estimated, it is preferable to execute the following steps: a procedure for obtaining time series data of the instantaneous heart rate of the subject whose blood pressure is to be estimated, a procedure for performing the frequency analysis and displaying a spectrogram, a procedure for setting the approximation curve, a procedure for selecting the best approximation curve, a procedure for extracting the inflection points and tangent intersections, a procedure for setting the approximation gradient curve, and a procedure for selecting the optimized approximation gradient curve; and obtain the first estimated circulatory dynamics index and the second estimated circulatory dynamics index of the subject whose blood pressure is to be estimated, respectively.

[0024] In the procedure for determining the first circulatory dynamics index and the second circulatory dynamics index, when the optimization approximate curve is an approximate gradient curve represented by the cubic function, it is preferable that the gradient of the tangent at the inflection point of the optimization approximate curve is the frequency gradient, and when the optimization approximate curve is an approximate gradient curve represented by the quartic function, the gradient of the bitangent is the frequency gradient.

[0025] In the procedure for selecting the best approximation curve, it is preferable to select the best approximation curve using at least one of the following evaluation indices regarding the residuals between each of the two types of approximation curves, the quartic function and the quintic function, and the spectrogram: standard deviation of the residuals, average value of the absolute values ​​of the residuals, best approximation error which is the difference between the standard deviation of the residuals and the average value of the absolute values ​​of the residuals, distribution shape of the residuals, and correlation between the standard deviation of the residuals and the best approximation error.

[0026] In the step of selecting the optimized approximate gradient curve, it is preferable to select the optimized approximate gradient curve using at least one of the following evaluation indexes regarding the residuals between each of the two types of approximate curves, the cubic function and the quartic function, and the spectrogram: standard deviation of the residuals, average value of the absolute values ​​of the residuals, best approximation error which is the difference between the standard deviation of the residuals and the average value of the absolute values ​​of the residuals, and distribution shape of the residuals. In the step of selecting the optimized approximate gradient curve, it is preferable to determine whether or not an overlap band (OLF) modified by both the long-term blood pressure regulation mechanism (LTM) and the short-term blood pressure regulation mechanism (STM) exists at a predetermined position in the selected optimized approximate gradient curve, and if not, to exclude outliers included in the long-term blood pressure regulation mechanism (LTM) in the spectrogram and perform recalculation.

[0027] The present invention also provides a computer-readable recording medium on which the computer program is recorded.

[0028] According to the present invention, time-series data on instantaneous heart rates of a group of subjects is analyzed to determine a first circulatory dynamics index associated with the long-term blood pressure regulatory mechanism and a second circulatory dynamics index associated with the short-term blood pressure regulatory mechanism in the arterial pressure-related frequency band, and a database of three-dimensional response surface models is constructed using the first and second circulatory dynamics indexes as explanatory variables and the measured systolic blood pressure data of each group of subjects as a response variable. The first and second circulatory dynamics indexes for estimation of blood pressure of a subject are applied to this three-dimensional response surface model to estimate the systolic blood pressure of the subject. Because the first circulatory dynamics index (first circulatory dynamics index for estimation) and the second circulatory dynamics index (second circulatory dynamics index for estimation) correspond to the long-term and short-term blood pressure regulatory mechanisms, respectively, the estimated systolic blood pressure reflects the influence of both mechanisms, improving the accuracy of blood pressure estimation.

[0029] FIG. 1 is a block diagram showing the overall configuration of a blood pressure estimation device according to a first embodiment of the present invention. FIG. 2 is a block diagram showing the configuration of a database construction unit. FIG. 3 is a flowchart illustrating the analysis procedure performed by the time-series data analysis unit. FIG. 4(a) shows electrocardiogram waveform data to be analyzed by the time-series data analysis unit. FIGS. 4(b) to 4(d) show output data obtained according to the analysis procedure performed by the time-series data analysis unit. FIG. 4(e) is a spectrum diagram obtained by the frequency analysis unit. FIG. 5 is a flowchart illustrating the procedure for setting an approximation curve on the spectrum diagram, obtaining a best approximation curve, and extracting inflection points and tangent intersections. FIG. 6 is a flowchart illustrating the procedure for obtaining a first circulatory dynamics index (GT of LTM) and a second circulatory dynamics index (GT of STM) following the procedure in FIG. 5. FIGS. 7(a) to 7(d) show examples of data obtained in Experimental Example 1 by setting an approximation curve on the spectrum diagram, obtaining a best approximation curve, and extracting inflection points and tangent intersections according to the flowchart in FIG. 5. Figures 8(a) to 8(d) show other examples of data obtained by setting approximation curves on the spectrum diagram, obtaining the best approximation curve, and extracting the inflection points and tangent intersections according to the flowchart of Figure 5 in Experimental Example 1. Figure 9 shows the frequency band of the long-term blood pressure regulation mechanism (LTM) from which the first circulatory dynamics index (GT of LTM) is obtained, and the frequency band of the short-term blood pressure regulation mechanism (STM) from which the second circulatory dynamics index (GT of STM) is obtained. Figures 10(a) and 10(b) show data obtained by calculating the optimized approximation slope curve for the long-term blood pressure regulation mechanism (LTM) from the best approximation curve obtained in Figure 7(d). Figures 11(a) and 11(b) show data obtained by calculating the optimized approximation slope curve for the short-term blood pressure regulation mechanism (STM) from the best approximation curve obtained in Figure 7(d). Figure 12 shows the SD-εn correlation diagram for the residuals of the approximate gradient curves obtained in Figures 10(a), (b) and 11(a), (b). Figures 13(a) and 13(b) show data for the optimized approximate gradient curve for the long-term blood pressure regulation mechanism (LTM) obtained from the best approximate curve obtained in Figure 8(d).Figures 14(a) and 14(b) show data obtained by calculating the optimized approximate gradient curve for the short-term blood pressure regulation mechanism (STM) from the best approximate curve obtained in Figure 8(d). Figure 15 shows the SD-εn correlation diagram for the residuals of the approximate gradient curves calculated in Figures 13(a) and 13(b) and 14(a) and 14(b). Figure 16(a) shows the best approximation curve obtained in Figure 7(d), as well as the optimized approximation slope curve for the long-term blood pressure regulation mechanism (LTM) and the optimized approximation slope curve for the short-term blood pressure regulation mechanism (STM) obtained in Figures 10(a), (b) and 11(a), (b). Figure 16(b) shows the frequency slope (GT of LTM), which serves as the first circulatory dynamics index obtained from the optimized approximation slope curve for the long-term blood pressure regulation mechanism (LTM), and the frequency slope (GT of STM), which serves as the second circulatory dynamics index obtained from the optimized approximation slope curve for the short-term blood pressure regulation mechanism (STM). Fig. 17(a) shows the best approximation curve obtained in Fig. 8(d), as well as the optimized approximate gradient curves for the long-term blood pressure regulation mechanism (LTM) and the short-term blood pressure regulation mechanism (STM) obtained in Figs. 13(a), (b) and 14(a), (b). Fig. 17(b) shows the frequency gradient (GT of LTM), which is the first circulatory dynamics index obtained from the optimized approximate gradient curve for the long-term blood pressure regulation mechanism (LTM), and the frequency gradient (GT of STM), which is the second circulatory dynamics index obtained from the optimized approximate gradient curve for the short-term blood pressure regulation mechanism (STM). Fig. 18 shows the three-dimensional response surface model obtained in Experimental Example 1. FIG. 19(a) is a correlation diagram between GT of LTM and systolic blood pressure in the three-dimensional response surface model of FIG. 18 , and FIG. 19(b) is a correlation diagram between GT of STM and systolic blood pressure in the three-dimensional response surface model of FIG. 18 . FIG. 20(a) is a correlation diagram between GT of LTM and GT of STM, and FIG. 20(b) is a correlation diagram between estimated systolic blood pressure (eSBP) estimated using the three-dimensional response surface model of FIG. 18 and each measured systolic blood pressure (SBP). FIG. 21(a) shows data from a 34-year-old subject in Experimental Example 2, and FIG. 21(b) shows data from a 67-year-old subject in Experimental Example 2. FIGS. 22(a) to 22(c) are diagrams illustrating the relationship between long-term blood pressure regulation (LTM), short-term blood pressure regulation (STM), and overlap band (OLF).FIG. 23 is a diagram showing a configuration including an OLF requirement determination unit 171 according to the second embodiment. FIG. 24 is a flowchart showing the processing procedure of the second embodiment. FIGS. 25(a) to 25(c) are diagrams showing an example of the processing procedure of the second embodiment. FIGS. 26(a) to 26(f) are examples of approximation curves (approximation gradient curves) of a cubic or quartic function in the operating band of the long-term blood pressure regulation mechanism (LTM) among the best approximation curves of the quintic function in FIG. 25(c). FIGS. 27(g) to 27(m) are examples of approximation curves (approximation gradient curves) of a cubic or quartic function in the operating band of the short-term blood pressure regulation mechanism (STM) among the best approximation curves of the quintic function in FIG. 25(c). FIG. 28 is an SD-εn correlation diagram for each of the approximation gradient curves of FIGS. 26 and 27. Figure 29(a) shows the approximate gradient curve of a cubic function for the LTM frequency band from 0.006 Hz to IP2, the approximate gradient curve of a quartic function for STM from IT2 to 0.06 Hz, and the best approximation curve of a quintic function. Figure 29(b) shows the LTM frequency band, the STM frequency band, and the OLF frequency band. Figure 30 shows the three-dimensional response surface model obtained in Experimental Example 3. Figure 31(a) shows the correlation between GT of LTM and systolic blood pressure in the three-dimensional response surface model of Figure 30, and Figure 31(b) shows the correlation between GT of STM and systolic blood pressure in the three-dimensional response surface model of Figure 30. FIG. 32(a) is a correlation diagram between GT of LTM and GT of STM, and FIG. 32(b) is a correlation diagram between the estimated systolic blood pressure (eSBP) estimated using the three-dimensional response surface model of FIG. 30 and each measured systolic blood pressure (SBP). FIGS. 33(a) to 33(c) are diagrams showing analysis examples of other subjects in Experimental Example 3. FIG. 34 is a diagram showing an analysis example of subject (c-1) in Experimental Example 3. FIG. 35 is a diagram showing an analysis example of subject (c-2) in Experimental Example 3. FIG. 36 is a diagram showing an analysis example of subject (c-3) in Experimental Example 3. FIG. 37 is a diagram showing an analysis example of subject (c-5) in Experimental Example 3. FIG. 38 is a diagram showing an analysis example of subject (c-6) in Experimental Example 3. FIG. 39 shows data for a 67-year-old subject in Experimental Example 4.

[0030] First Embodiment [Blood Pressure Estimation Device 1] FIG. 1 shows a schematic configuration of the blood pressure estimation device 1 of this embodiment. The blood pressure estimation device 1 of this embodiment is configured by a computer (the type of computer is not limited, and includes a personal computer, a microcomputer, a mobile information terminal such as a smartphone, etc.) and executes predetermined processing. Specifically, the blood pressure estimation device 1 is configured by storing computer programs that execute procedures functioning as a database construction unit 10, an estimation index calculation unit 20, and a systolic blood pressure estimation unit 30 in a storage unit (including a storage medium such as a hard disk built into the blood pressure estimation device 1, various removable storage media, and storage media of other computers connected via communication means). The blood pressure estimation device 1 can also be realized using an electronic circuit having one or more storage circuits incorporating the computer programs. That is, it can be configured by processing circuits functioning as the database construction unit 10, the estimation index calculation unit 20, and the systolic blood pressure estimation unit 30.

[0031] The computer program can also be provided by being stored on a recording medium. The recording medium storing the computer program may be a non-transitory recording medium. Examples of non-transitory recording media include, but are not limited to, flexible disks, hard disks, CD-ROMs, MOs (magneto-optical disks), DVD-ROMs, and memory cards. The computer program can also be transmitted to a computer via a communication line and installed.

[0032] [Database Construction Unit 10] The database construction unit 10 analyzes time series data of instantaneous heart rates of a predetermined number of subjects, determines a first circulatory dynamics index (GT of LTM) related to the long-term blood pressure regulation mechanism (LTM) and a second circulatory dynamics index (GT of STM) related to the short-term blood pressure regulation mechanism (STM) in the arterial pressure-related frequency band, and constructs a database of a three-dimensional response surface model in which the first circulatory dynamics index and the second circulatory dynamics index are used as explanatory variables and the measured data of the systolic blood pressure of each of the subjects is used as a response variable.

[0033] More specifically, the computer program functioning as the database construction unit 10 executes procedures that function as a time-series data analysis unit 11, a frequency analysis unit 12, an approximate curve calculation unit 13, a best approximate curve selection unit 14, an inflection point / intersection extraction unit 15, an approximate gradient curve setting unit 16, an optimized approximate gradient curve selection unit 17, a circulatory dynamics index determination unit 18, and a three-dimensional response surface model creation unit 19, as shown in FIG. 2 .

[0034] (Time Series Data Analysis Unit 11) The time series data analysis unit 11 reconstructs a two-dimensional delay coordinate system for the time series data of the instantaneous heart rate (HR = 60 / Δt) using a Lorenz plot, and finds an attractor related to heart rate fluctuations. The time series data of the instantaneous heart rate can be, for example, electrocardiogram data obtained from an electrocardiograph (ECG) (S1 in Fig. 3, Fig. 4(a)). The Lorenz plot uses two points from the time series data of the instantaneous heart rate (HR), and plots them while moving the range. Specifically, the x-axis is plotted as HR. ｉ＋１ (HR ｉ＋２ , H.R. ｉ＋３ ...), HR on the y-axis ｉ (HR ｉ＋１ , H.R. ｉ＋２ ...) are taken and their intersections are plotted sequentially. This is performed for a reference time range (360 seconds in this embodiment) to create an attractor for the reference time range formed by the plots. Next, a regression line of the attractor for this reference time range is obtained. This regression line indicates the ensemble average of the plot group for the reference time range, and the slope of the regression line with respect to the x-axis is used as an index showing its characteristics, and the reference angle information (θ ０ ) (S2 in FIG. 3, FIG. 4(b)).

[0035] Next, the time series data analysis unit 11 reconstructs a two-dimensional delay coordinate system using a Lorenz plot in the same manner as described above for an individual time range (30 seconds in this embodiment) shorter than the reference time range, and creates an individual attractor. Then, as an index showing the characteristics of the slope of the regression line, which is the ensemble average of the plot group that constitutes the individual attractor, the slope of the regression line with respect to the x-axis is calculated as individual angle information (θ ｉ ) is calculated as individual angle information (θ ｉ) is calculated sequentially while moving the individual time range (30 seconds) at a predetermined overlap rate (for example, 90%) (S2 in FIG. 3, FIG. 4(c)). Next, the time series data analysis unit 11 calculates the reference angle information (θ ０ ) and each individual angle information (θ ｉ ) and the difference (θ ｉ－０ ) are calculated sequentially, and the difference (θ ｉ－０ ) time series data (θ ｉ－０ (time series data) is calculated (S3 in FIG. 3, FIG. 4(d)). ｉ－０ The time series data can be considered as an index showing the general fluctuation of the individual attractors obtained every 3 seconds in the case of a 90% overlap rate.

[0036] θ ｉ－０ As shown in FIG. 4(d), the time series data has the horizontal axis as time and the vertical axis as angle information showing an increasing or decreasing tendency (a negative sign indicates an increasing tendency of the instantaneous heart rate and an increasing tendency of the sympathetic nervous system, and a positive sign indicates a decreasing tendency of the instantaneous heart rate and an inhibiting tendency of the sympathetic nervous system). In the case of a 90% overlap rate, the difference (θ ｉ－０ ) is plotted as a time series graph. ｉ－０ ) should preferably be plotted every 3 seconds. This is the average time over which the effects of parasympathetic nervous activation and inhibition are reflected, and 30 seconds is approximately twice the average time of 12.5 seconds over 10 to 15 seconds over which the effects of sympathetic nervous activation and inhibition are reflected. 360 seconds corresponds to 1 / 4 cycle of fluctuations in alertness, drowsiness, and mental work ability in the ultradian cycle. Note that a -+ display indicates a tendency to suppress blood pressure, while a + indicates a tendency to suppress blood pressure.

[0037] (Frequency analysis unit 12) The frequency analysis unit 12 analyzes the θ ｉ－０ This is a means for performing frequency analysis on time series data and displaying it on a double logarithmic axis (S4 and S5 in FIG. 3, and FIG. 4(e)). FIG. 4(e) shows an example of a double logarithmic graph display (spectral diagram) of the power spectrum as a result of frequency analysis, with many points plotted.

[0038] In this embodiment, frequency analysis was performed with a sampling frequency of 1000 Hz and a time resolution of 0.001 seconds. Therefore, the effective band in the calculation is up to 1000 / 2.5 = 400 Hz. ｎWhen performing spectral analysis using 360 seconds of time series data, ７ = 128 is close to this condition. As mentioned above, the moving average in the time series data analysis unit 11 is performed with a 30-second window shifted by 3 seconds, and the first point can be calculated after 30 seconds have passed. The heart rate regulation function and respiratory regulation function are reflected in the moving average performed at 30-second intervals. Therefore, (360 - 30) / 3 + 1 = 111 data are output in 360 seconds. In order to perform the spectrum calculation, 2 ７ Since 128 data points are required, the missing 17 data points are folded back. Assuming that the response of the sympathetic nervous system to stimulation is captured in 12.5 seconds, low-pass filtering is performed at 1 / 12.5 = 0.08 Hz. The result is 1 / 360 = 0.0028 Hz, but because the power spectrum near 0.0028 Hz contains drift and offset components, high-pass filtering is performed at 0.006 Hz, above 0.0028 Hz x 2 = 0.0056 Hz. As a result, the frequency band related to the long-term blood pressure regulation mechanism (LTM) covers approximately 70% of the VLF band, which is the frequency band from 0.003 to 0.04 Hz.

[0039] The spectrum diagram obtained by the frequency analysis unit 12 is expressed as a double logarithmic graph with the horizontal axis representing frequency and the vertical axis representing the power spectrum, so the graph slopes downward to the right. ｎ The periodic component creates a large spectrum and produces extreme values ​​on the graph.

[0040] (Approximation curve calculation unit 13) The approximation curve calculation unit 13 performs a calculation to fit an approximation curve to the spectrum diagram of the power spectrum plotted on a double logarithmic graph obtained by the frequency analysis unit 12. Here, the frequency analysis unit 12 performs processing using a high-pass filter at 0.006 Hz and a low-pass filter at 0.08 Hz. However, as described above, the operating band of the long-term blood pressure regulation mechanism (LTM) is centered on the frequency band of 0.006 to 0.017 Hz, and the operating band of the short-term blood pressure regulation mechanism (STM) is centered on the frequency band of 0.017 to 0.06 Hz, so 0.006 to 0.06 Hz is defined as the "arterial pressure-related frequency band."

[0041] The approximation curve calculation unit 13 calculates and sets two types of approximation curves, a quartic function and a quintic function, for the spectrogram in the arterial pressure-related frequency band of 0.006 to 0.06 Hz (S11 in FIG. 5). In the analysis example described below, approximation curve A in FIGS. 7(a) and 8(a) is an example of an approximation curve represented by a quartic function, and approximation curve B (B-1, B-2) is an example of an approximation curve represented by a quintic function. The arterial pressure-related frequency band of 0.006 to 0.06 Hz covers the operating bands of the long-term blood pressure regulation mechanism (LTM) and the short-term blood pressure regulation mechanism (STM), and this range indicates the overall tendency of arterial pressure control. Furthermore, the operating bands of the long-term blood pressure regulation mechanism (LTM) and the short-term blood pressure regulation mechanism (STM) are not constant due to individual differences and the influence of physical condition at the time of measurement. The lower limit frequency of 0.006 Hz and the upper limit frequency of 0.06 Hz can be limited in consideration of the accuracy of frequency analysis as described above, but the boundary between the two operating bands is not necessarily 0.017 Hz. Therefore, within each operating band of the long-term blood pressure regulation mechanism (LTM) or short-term blood pressure regulation mechanism (STM), the frequency band to be analyzed for determining each circulatory dynamics index is not constant due to differences between individuals and differences in physical condition.

[0042] Therefore, the approximation curve calculation unit 13 sets two types of approximation curves, a quartic function and a quintic function, for the entire range of 0.006 to 0.06 Hz, which is the arterial pressure-related frequency band, so that a more appropriate band can be selected as the analysis target frequency band for determining each circulatory dynamics index for each data to be analyzed, and selects one of them as the best approximation curve.

[0043] This is because two or more inflection points appear, making it easy to find characteristic points (inflection points and tangent intersections, described below) for limiting the frequency band to be analyzed. However, the basic form of the approximation curve is a quartic function in which the coefficient term to the fourth power is negative, and when the graph shape is reversed, the coefficient becomes positive. When the coefficient is positive, the gradient of the tangent line to the inflection point (frequency gradient) becomes positive, suggesting enhanced accommodation function. Furthermore, the graph of a quartic function has diagonal symmetry, and the basic form has three extreme values. It can be divided into cases where two extreme values ​​are close to each other and the number of extreme values ​​decreases, and cases where two extreme values ​​exist outside the frequency band to be analyzed. Generally, if the approximation curve of an nth-order function does not follow a normal distribution in terms of the residuals described below, increasing the order increases the likelihood that it will follow a normal distribution. Furthermore, if outliers occur in the residuals between the power spectrum and the approximation curve, the frequency band to be analyzed is changed or the order is increased to 5th, and an approximation curve is created, and the inflection point IP and the intersection point IT are calculated.

[0044] However, in this embodiment, rather than fitting an approximate curve of a quartic function and then finding an approximate curve of a quintic function in light of the distribution of residuals, the two types of functions, a quartic function and a quintic function, are compared from the beginning to quickly obtain an approximate curve (best approximate curve) that is appropriate for capturing the overall trend of arterial blood pressure control.

[0045] (Best Approximation Curve Calculation Unit 14) The best approximation curve selection unit 14 selects an approximation curve (best approximation curve) to be used for analysis from the two types of approximation curves, the quartic function and the quintic function, calculated by the approximation curve calculation unit 13 (S12 in FIG. 5). As the selection criteria, it is preferable to use an evaluation index related to the residual between each of the two types of approximation curves, the quartic function and the quintic function, and the spectrogram (the difference between the power spectrum plot and each approximation curve). The evaluation index related to the residual can be at least one of the following: the standard deviation (SD) of the residuals calculated for each power spectrum plot in the spectrogram; the average value (Mean(abs)) of the absolute values ​​of the residuals; the best approximation error (εn), which is the difference (SD-Mean(abs)) between the standard deviation (SD) of the residuals and the average value (Mean(abs)) of the absolute values ​​of the residuals; and the distribution shape of the residuals (whether the distribution is symmetrical or has a wider tail on either the left or right).

[0046] In this embodiment, the standard deviation (SD) of the residuals, the best approximation error (εn), and the distribution shape of the residuals are used as selection criteria among the above evaluation indices. Specifically, first, an SD-εn correlation diagram showing the correlation between the standard deviation (SD) of the residuals and the best approximation error (εn) is created, and a histogram of the residuals is also created. Then, as shown in S12 and S13 of FIG. 5, an approximation curve of either a quartic or quintic function in which the histogram has a symmetrical distribution or a distribution with a broad tail on either the left or right, and in which the best approximation error εn→0 (in this specification, εn is 0 or close to 0) is selected as the best approximation curve (see FIGS. 7(b) and (c) and FIGS. 8(b) and (c)).

[0047] Here, we will explain why the best approximation error εn is included in the selection criteria. In the interval (a, b), the approximation function by a polynomial that satisfies the normal distribution and is easy to handle numerically is set as g(x). i} i=1 n The value at {gx i} i=1 n Select an n-th order function f(x) for

[0048] For a function g(x) that is continuous in the interval (a, b), the k-th degree polynomial f k Made using (x) is the best approximation error, and the n-th order function f(x) is called the best approximation function. Note that εn→0 and Weierstrass's theorem are equivalent.

[0049] Also, if the random variable X follows a normal distribution, then the probability density of the best approximation function f(x) is This is a state in which the above conditions are satisfied.

[0050] Since a function that is continuous in a closed interval follows the Weierstrass theorem, which is the fundamental theorem of approximation theory, and the normal distribution, we select the n-th order function f(x) as {x i} i=1 nFrom the residuals, which are the errors between the two, the standard deviation (SD) of the residuals and the average value of the absolute values ​​of the residuals (Mean (abs)) were calculated, SD-Mean (abs) was defined as εn, and the selection condition for the best approximation curve was that εn → 0 and SD be as small as εn.

[0051] (Inflection Point / Intersection Extraction Unit 15) The inflection point / intersection extraction unit 15 extracts inflection points and tangent intersections for the best approximation curve. That is, the selected best approximation curve is differentiated to calculate inflection points (IP), and the intersections of tangents (IT) at the inflection points are obtained. If it is determined in S14 of FIG. 5 that a quartic function has been selected as the best approximation curve, the IP and IT will be IP1, IT1, and IP2, in order from the low frequency side (see S15 of FIG. 5 and FIG. 8(d)). If it is determined in S14 of FIG. 5 that a quintic function has been selected as the best approximation curve, the IP and IT will be IP1, IT1, IP2, IT2, and IP3, in order from the low frequency side (see S16 of FIG. 5 and FIG. 7(d)).

[0052] (Approximate gradient curve setting unit 16) The approximate gradient curve setting unit 16 determines two types of approximate curves, a cubic function and a quartic function, for each frequency band corresponding to all combinations of the frequency of the inflection point (IPn), the frequency of the tangent intersection (ITn), and two frequencies selected from the lower limit 0.006 Hz and the upper limit 0.06 Hz of the arterial pressure-related frequency band in each of the operating bands of the long-term blood pressure regulation mechanism (LTM) and the short-term blood pressure regulation mechanism (STM) (S21 in FIG. 6). These approximate curves are referred to as "approximate gradient curves" in this specification.

[0053] As mentioned above, the operating bands of the long-term blood pressure control mechanism (LTM) and the short-term blood pressure control mechanism (STM) are separated by approximately 0.017 Hz, but this is not constant due to individual differences and physical condition. Therefore, the operating band of the long-term blood pressure control mechanism (LTM) is considered to be the range from the low-frequency side to the second inflection point (IP2) in the case of a quartic function, from 0.006 Hz to 0.017 Hz, and in the case of a quintic function, from the low-frequency side to the second tangent intersection point (IT2) located next to the second inflection point (IP2). The operating band of the short-term blood pressure control mechanism (STM) is considered to be the range from the lowest-frequency inflection point (IP1) to 0.06 Hz, for both the quartic and quintic functions, from 0.017 Hz to 0.06 Hz.

[0054] Figure 9 is a diagram conceptually showing the relationship between the inflection points (IPn) and the tangent intersection points (ITn) in the operating bands of the long-term blood pressure control mechanism (LTM) and the short-term blood pressure control mechanism (STM). The specific frequency bands to be considered for application of approximate gradient curves expressed by cubic and quartic functions are as follows:

[0055] (1) Operating band of long-term blood pressure regulation mechanism (LTM) (1-a) When the best approximation curve is a quartic function: 0.006 Hz to IP1, 0.006 Hz to IT1, 0.006 Hz to IP2, IP1 to IT1, IP1 to IP2, IT1 to IP2 (number of frequency bands considered: 6) (1-b) When the best approximation curve is a quintic function: 0.006 Hz to IP1, 0.006 Hz to IT1, 0.006 Hz to IP2, 0.006 Hz to IT2, IP1 to IT1, IP1 to IP2, IP1 to IT2, IT1 to IP2, IT1 to IP2, IP2 to IT2 (number of frequency bands considered: 10)

[0056] (2) Operating band of short-term blood pressure regulation mechanism (STM) (2-a) When the best approximation curve is a quartic function: IP1 to IT1, IP1 to IP2, IP1 to 0.06 Hz, IT1 to IP2, IT1 to 0.06 Hz, IP2 to 0.06 Hz (number of frequency bands considered: 6) (2-b) When the best approximation curve is a quintic function: IP1 to IT1, IP1 to IP2, IP1 to IT2, IP1 to IP3, IP1 to 0.06 Hz, IT1 to IP2, IT1 to IT2, IT1 to IP3, IT1 to 0.06 Hz, IP2 to IT2, IP2 to IP3, IP2 to 0.06 Hz, IT2 to IP3, IT2 to 0.06 Hz, IP3 to 0.06 Hz (number of frequency bands considered: 15)

[0057] (Optimized approximate gradient curve selection unit 17) The optimized approximate gradient curve selection unit 17 selects an optimized approximate gradient curve, which is an approximate gradient curve for calculating the first circulatory dynamics index and the second circulatory dynamics index, from all of the two types of approximate gradient curves, namely, cubic functions and quartic functions, calculated in each of the above frequency bands (S22 in Figure 6).

[0058] As a selection criterion for the optimized approximate gradient curve, it is preferable to use an evaluation index related to the residual between each approximate gradient curve and the spectrogram (the difference between the plot of the power spectrum and each approximate gradient curve). As an evaluation index related to the residual, similar to the case of selecting the best approximate curve described above, at least one of the following can be used: "the standard deviation (SD) of the residuals obtained for each plot of the power spectrum in the spectrogram, the average value (Mean(abs)) of the absolute values ​​of the residuals, the best approximation error (εn) which is the difference (SD-Mean(abs)) between the standard deviation (SD) of the residuals and the average value (Mean(abs)) of the absolute values ​​of the residuals, and the distribution shape of the residuals (whether the distribution is symmetrical or has a wider tail on either the left or right)."

[0059] In this embodiment, the standard deviation (SD) of the residuals, the best approximation error (εn), and the distribution shape of the residuals are used as selection criteria among the above evaluation indices. Specifically, first, an SD-εn correlation diagram showing the correlation between the standard deviation (SD) of the residuals and the best approximation error (εn) is created, and a histogram of the residuals is also created. Then, an approximation gradient curve of either a cubic or quartic function where the histogram has a symmetrical distribution or a distribution with a broad tail on either the left or right, and where εn → 0, is selected as the optimized approximation gradient curve (S22 in FIG. 6).

[0060] (Circulatory dynamics index determination unit 18) The circulatory dynamics index determination unit 18 determines the frequency gradient of the optimized approximate gradient curve selected in the operating band of the long-term blood pressure regulation mechanism (LTM) and sets it as the first circulatory dynamics index (GT of LTM), and determines the frequency gradient of the optimized approximate gradient curve selected in the operating band of the short-term blood pressure regulation mechanism (STM) and sets it as the second circulatory dynamics index (GT of STM) (S23 to S26 in FIG. 6). Note that, in the double logarithmic graph of frequency and power spectrum, the frequency gradient is the gradient of the tangent when a cubic function is selected as the optimized approximate gradient curve (S23 and S24 in FIG. 6), and the gradient of the bitangent when a quartic function is selected as the optimized approximate gradient curve (S23 and S25 in FIG. 6).

[0061] (Three-dimensional response surface model creation unit 19) The three-dimensional response surface model creation unit 19 uses the first circulatory dynamics index (GT of LTM) and the second circulatory dynamics index (GT of STM) determined by the circulatory dynamics index determination unit 18 as explanatory variables, and applies the measured systolic blood pressure data of each subject group as a response variable to create a three-dimensional response surface model (see Figure 2). Figure 18 shows an example of a three-dimensional response surface model created in this way, represented by a fourth-order polynomial. The database construction unit 10 stores the obtained three-dimensional response surface model as a database in the memory unit of the blood pressure estimation device 1.

[0062] 1, the estimation index calculation unit 20 obtains and analyzes time-series data of the instantaneous heart rate of a person whose blood pressure is to be estimated (a blood pressure estimation subject), and determines a first circulatory dynamics index (GT of LTM) and a second circulatory dynamics index (GT of STM) of the subject, which are designated as the first circulatory dynamics index (estimation GT of LTM) and the second circulatory dynamics index (estimation GT of STM), respectively. The first circulatory dynamics index (GT of LTM) and the second circulatory dynamics index (GT of STM) are determined by the time-series data analysis unit 11, the frequency analysis unit 12, the approximation curve calculation unit 13, the best approximation curve selection unit 14, the inflection point / intersection extraction unit 15, the approximation gradient curve setting unit 16, the optimized approximation gradient curve selection unit 17, and the circulatory dynamics index determination unit 18, which are included in the database construction unit 10. That is, time series data of the instantaneous heart rate of the subject for whom blood pressure is to be estimated is sent to the time series data analysis unit 11, and processing is executed in order from the frequency analysis unit 12 to the circulatory dynamics index determination unit 18, and the first circulatory dynamics index (GT of LTM) and the second circulatory dynamics index (GT of STM) are determined.

[0063] [Systolic Blood Pressure Estimation Unit 30] As shown in FIG. 1 , the systolic blood pressure estimation unit 30 applies the first estimation circulatory dynamics index (estimation GT of LTM) and the second estimation circulatory dynamics index (estimation GT of STM) obtained by the estimation index calculation unit 20 to the three-dimensional response surface model constructed by the database construction unit 10, estimates the systolic blood pressure of the subject whose blood pressure is to be estimated, and outputs the estimated value.

[0064] In the above description, data obtained from an electrocardiograph (ECG) is used as the time-series data of instantaneous heart rate analyzed by the time-series data analysis unit 11. However, this is not limited to this, as long as similar data can be obtained. For example, a wristwatch-type electrocardiograph known as a smart watch or other wearable device can also be used. Furthermore, time-series waveform data showing a period approximating the R-R interval (RRI) can also be used, which is obtained by capturing biosignals (acoustic vibration data (Acoustic Pulse Wave: APW)) from the back of the body surface using a biosignal detection sensor built into a Sleep Buster product manufactured by Delta Tooling Co., Ltd. and filtering the time-series waveform in a predetermined frequency band.

[0065] (Experimental Example 1) The subjects were in their 20s to 90s, sitting quietly and breathing naturally. The experiment consisted of a total of 171 subjects, including a group of healthy individuals (95 individuals) who were not receiving medical care, and a group of outpatients (76 individuals) with cardiovascular diseases or risk factors for these diseases, such as heart disease, high blood pressure, gout, diabetes, and dyslipidemia.

[0066] Heart rate data for 360 seconds was collected from 171 subjects (260 data points) in a seated, resting position using an electrocardiograph, and systolic blood pressure (SBP) was measured using a brachial blood pressure monitor. Measurements were conducted at three locations: Delta Tooling's research facility, Shiga University of Medical Science, and Fukui Prefectural Saiseikai Hospital. At Delta Tooling's research facility and Shiga University of Medical Science, measurements were conducted using a car seat, while at Fukui Prefectural Saiseikai Hospital, measurements were conducted using a relaxation chair. The experimenters, subjects, analysts, and seasons were all different. Brachial systolic blood pressure was recorded as the average of two measurements. However, if the difference between the two measurements was 5 mmHg or more, measurements were conducted three or four times, and the average of the two closest measurements was recorded.

[0067] 260 pieces of heart rate data (time series data of instantaneous heart rate) were analyzed by the time series data analysis unit 11 and frequency analyzed by the frequency analysis unit 12, and for the obtained spectrogram, an approximate curve of a quartic function and an approximate curve of a quintic function were calculated by the approximate curve calculation unit 13. The approximate curve A in Fig. 7(a) and the approximate curve A in Fig. 8(a) are examples of approximate curves of quartic functions, and the approximate curves B-1 and B-2 in Fig. 7(a) and the approximate curve B in Fig. 8(a) are examples of approximate curves of quintic functions.

[0068] The best approximation curve selection unit 14 selects the best approximation curve from each of the approximation curves shown in FIGS. 7(a) and 8(a). In both cases, the best approximation error εn→0 and the residual histogram distribution shape are selected in accordance with S12 of FIG. 5, with a symmetrical distribution or a distribution that is slanted to the left or right. In the case of the approximation curve shown in FIG. 7(a), as shown in FIG. 7(b), the residual histogram approximation curve A, which is a quartic function, exhibits a symmetrical distribution, while the quintic function approximation curves B-1 and B-2 both exhibit slanted to the left and slanted to the right. Therefore, both histogram distribution shape conditions are satisfied. Next, comparing the best approximation errors εn, the SD-εn correlation diagram shown in FIG. 7(c) shows that the εn value closest to 0 is 0.0137. As a result, the quintic function approximation curve B-2 is selected as the best approximation curve.

[0069] Next, since the selected approximate curve B-2 is a quintic function, the inflection point / intersection extraction unit 15 extracts inflection points IP1 and IP2 and tangent intersection points IT1 and IT2 as shown in Fig. 7(d) in accordance with S14 and S16 of Fig. 5. Note that since the approximate curve B-2 in Fig. 7(d) is a quintic function, an inflection point IP3 also exists, but is not shown because it is outside the range of the graph.

[0070] On the other hand, in the case of the example of the approximation curve A of the quartic function and the approximation curve B of the quintic function in Fig. 8(a), the histograms of the residuals both have a symmetrical distribution as shown in Fig. 8(b). Therefore, by referring to the SD-εn correlation diagram shown in Fig. 8(c), it is determined that the εn value closest to 0 is 0.0221, and the approximation curve A of the quartic function is selected as the best approximation curve. Because the selected approximation curve A is a quartic function, the inflection point / intersection extraction unit 15 extracts the inflection points IP1 and IP2 and the intersection point IT1 of the tangents as shown in Fig. 8(d).

[0071] Next, the approximation gradient curve setting unit 16 sets approximation curves of cubic and quartic functions in the above frequency bands corresponding to the characteristic points of the best approximation curve, such as the inflection points IPn (n = 1, 2 for a quartic function, and n = 1, 2, 3 for a quintic function), the tangent intersection points ITn (n = 1 for a quartic function, and n = 1, 2 for a quintic function), and two combinations of frequencies between the lower limit of 0.006 Hz and the upper limit of 0.06 Hz in the arterial pressure-related frequency band. Figure 10(a) shows an example of the approximation curve (approximation gradient curve) of the cubic function in the operating band of the long-term blood pressure regulation mechanism (LTM) from the best approximation curve of the quintic function in Figure 7(d). Figure 10(b) shows an example of the approximation curve (approximation gradient curve) of the quartic function in the operating band of the long-term blood pressure regulation mechanism (LTM) from the best approximation curve of the quintic function in Figure 7(d).

[0072] The approximate gradient curves shown in FIGS. 10A and 10B show only data that can be candidates for the optimized approximate gradient curve, excluding data in the frequency band under consideration where the histogram of residuals is neither symmetrical nor spreads out to the left or right and where the best approximation error εn is large.

[0073] Figure 11(a) is an example of a best fit approximation curve of the quintic function in Figure 7(d) in which an approximation curve (approximation gradient curve) of a cubic function was obtained in the operating range of the short-term blood pressure control mechanism (STM). Figure 11(b) is an example of a best fit approximation curve of the quintic function in Figure 7(d) in which an approximation curve (approximation gradient curve) of a quartic function was obtained in the operating range of the short-term blood pressure control mechanism (STM).

[0074] As in the case of FIGS. 10(a) and 10(b), the approximate gradient curves shown in FIGS. 11(a) and 11(b) show only data that can be candidates for the optimized approximate gradient curve, excluding data in the frequency band under consideration where the histogram of residuals is neither symmetrical nor spreads out to either the left or right, and where the best approximation error εn is large.

[0075] Figure 12 is an SD-εn correlation diagram for Figures 10(a) and 10(b) and Figures 11(a) and 11(b). From Figure 12, the optimized approximate gradient curve selection unit 17 selects data that satisfies the following conditions: a low standard deviation SD, a best approximation error εn close to 0, and a regression line close to the operating band of the long-term blood pressure regulation mechanism (LTM). As a result, the frequency band to be analyzed is selected from Figure 10(a), from 0.006 Hz to IP2, where the best approximation error εn is 0.005, and the approximate gradient curve of the cubic function found in this analysis frequency band is selected as the optimized approximate gradient curve.

[0076] 12, data in the operating band of the short-term blood pressure control mechanism (STM) is selected from Figures 11(a) and 11(b) that satisfy the following conditions: the standard deviation SD is low, the best approximation error εn is close to 0, and the data is close to the regression line in the operating band of the short-term blood pressure control mechanism (STM). As a result, the frequency band from IT2 to 0.06 Hz in Figure 11(a), where the best approximation error εn is 0.017, is selected as the frequency band to be analyzed, and the approximate gradient curve of the cubic function found in this analysis frequency band is selected as the optimized approximate gradient curve.

[0077] 13 and 14 show examples in which approximate curve A of a quartic function is selected as the best approximate curve of FIG. 8(d), and the approximate gradient curve setting unit 16 sets approximate curves of a cubic function and a quartic function in the above frequency bands relating to the combination of the inflection points IP1 and IP2, the intersection point IT1 of the tangent lines, and two of the frequencies of the lower limit 0.006 Hz and the upper limit 0.06 Hz of the arterial pressure-related frequency band.

[0078] Figure 13(a) is an example of the approximate curve (approximate gradient curve) of a cubic function obtained in the operating range of the long-term blood pressure regulation mechanism (LTM) from the best approximation curve of the quartic function in Figure 8(d), Figure 13(b) is an example of the approximate curve (approximate gradient curve) of a quartic function obtained in the operating range of the long-term blood pressure regulation mechanism (LTM) from the best approximation curve of the quartic function in Figure 8(d), Figure 14(a) is an example of the approximate curve (approximate gradient curve) of a cubic function obtained in the operating range of the short-term blood pressure regulation mechanism (STM) from the best approximation curve of the quartic function in Figure 8(d), and Figure 14(b) is an example of the approximate curve (approximate gradient curve) of a quartic function obtained in the operating range of the short-term blood pressure regulation mechanism (STM) from the best approximation curve of the quartic function in Figure 8(d).

[0079] The approximate gradient curves shown in FIGS. 13(a) and 13(b) and 14(a) and 14(b) show only data that can be candidates for the optimized approximate gradient curve, excluding data in the frequency band under consideration where the histogram of residuals does not show a symmetrical distribution or a distribution that spreads out to either the left or right and where the value of the best approximation error εn is large.

[0080] FIG. 15 shows the SD-εn correlation diagrams for FIGS. 13(a) and 13(b) and 14(a) and 14(b). The optimized approximate gradient curve selector 17 selects data from FIG. 15 that meets the following conditions: a low standard deviation SD, a best approximation error εn close to 0, and a regression line close to the operating band of the long-term blood pressure regulation mechanism (LTM). As a result, the frequency band from 0.006 Hz to IP2 in FIG. 13(b), where the best approximation error εn is 0.0107, is selected as the frequency band to be analyzed, and the approximate gradient curve of the cubic function calculated in this analyzed frequency band is selected as the optimized approximate gradient curve. Similarly, for the operating band of the short-term blood pressure regulation mechanism (STM), the frequency band from IT1 to 0.06 Hz in FIG. 14(b), where the best approximation error εn is 0.0096, is selected as the frequency band to be analyzed, and the approximate gradient curve of the cubic function calculated in this analyzed frequency band is selected as the optimized approximate gradient curve.

[0081] Fig. 16(a) is a graph showing the best approximation curve of the quintic function shown in Fig. 7(d), the optimized approximate gradient curve of the long-term blood pressure regulation mechanism (LTM) with the frequency band of 0.006 Hz to IP2 analyzed shown in Fig. 10(a), and the optimized approximate gradient curve of the short-term blood pressure regulation mechanism (STM) with the frequency band of IT2 to 0.06 Hz analyzed shown in Fig. 11(a). The optimized approximate gradient curves of the long-term blood pressure regulation mechanism (LTM) and the short-term blood pressure regulation mechanism (STM) are all expressed as cubic functions.

[0082] Therefore, the circulatory dynamics index determination unit 18 determines the gradient of the tangent to each inflection point as the frequency gradient (first circulatory dynamics index, second circulatory dynamics index). Figure 16(b) shows the gradient of the tangent to each inflection point. The frequency gradient (first circulatory dynamics index): GT of LTM of the long-term blood pressure regulation mechanism (LTM) in the frequency band to be analyzed from 0.006 Hz to IP2 is determined to be 1.048, and the frequency gradient (second circulatory dynamics index): GT of STM of the short-term blood pressure regulation mechanism (STM) in the frequency band to be analyzed from IT2 to 0.06 Hz is determined to be 1.9366.

[0083] Fig. 17(a) is a graph showing the best approximation curve of the quartic function shown in Fig. 8(d), the optimized approximate gradient curve of the long-term blood pressure regulation mechanism (LTM) with the frequency band of 0.006 Hz to IP2 analyzed shown in Fig. 13(b), and the optimized approximate gradient curve of the short-term blood pressure regulation mechanism (STM) with the frequency band of IT1 to 0.06 Hz analyzed shown in Fig. 14(b). The optimized approximate gradient curves of the long-term blood pressure regulation mechanism (LTM) and the short-term blood pressure regulation mechanism (STM) are all represented by quartic functions.

[0084] Therefore, the circulatory dynamics index determination unit 18 determines the gradients of the bitangents as the frequency gradients (first circulatory dynamics index, second circulatory dynamics index). Figure 17(b) shows the gradients of the bitangents. The frequency gradient (first circulatory dynamics index): GT of LTM of the long-term blood pressure regulation mechanism (LTM) in the frequency band of 0.006 Hz to IP2 is determined to be -0.508, and the frequency gradient (second circulatory dynamics index): GT of STM of the short-term blood pressure regulation mechanism (STM) in the frequency band of IT1 to 0.06 Hz is determined to be -0.3648.

[0085] Fluctuations in the frequency band to be analyzed occur due to factors such as aging and stress. When noise such as aging and stress is added to the chaotic data in the two-dimensional delay coordinate system described above, the data becomes more orderly, which becomes a source of error. The frequency gradients of the long-term blood pressure regulation mechanism (LTM) and short-term blood pressure regulation mechanism (STM) exhibit a white noise state when close to 0, a pink noise state showing 1 / f fluctuations when close to -1, and a Brownian noise state with strong linearity when -2 or above.

[0086] Once the first circulatory dynamics index (GT of LTM) and the second circulatory dynamics index (GT of STM) are obtained for all of the data for the subjects, the three-dimensional response surface model creation unit 19 creates a three-dimensional response surface model using the first circulatory dynamics index (GT of LTM) on the X-axis, the second circulatory dynamics index (GT of STM) on the Y-axis, and the systolic blood pressure (SBP) values ​​corresponding to the Z-axis, and stores the model in the database (DB) 40. The resulting three-dimensional response surface model is shown in FIG. 18. FIG. 19(a) is a correlation diagram illustrating the three-dimensional response surface model of FIG. 18 with GT of LTM on the horizontal axis and systolic blood pressure on the vertical axis, and FIG. 19(b) is a correlation diagram illustrating the three-dimensional response surface model of FIG. 18 with GT of STM on the horizontal axis and systolic blood pressure on the vertical axis.

[0087] The obtained three-dimensional response surface model was approximated by the following fourth-order polynomial: f(x,y) = -5.8097x 4 -3.32x 3 y -5.5567x 3 -0.91069x 2 y 2-4.4718x 2 y + 6.0328x 2 -0.14457xy 3 +2.6996xy 2 +3.6018xy +15.1034x -0.31076y 4 -1.3092y 3 + 5.8705y 2 + 23.4022y + 135.6673

[0088] 18 and 19 show that a saddle point was formed on the three-dimensional response surface model. The saddle point was located within the normal blood pressure range of 100 to 120 mmHg. In the normal blood pressure state located on the saddle point, it is believed that both LTM and STM regulation minimize energy and function efficiently with 1 / f fluctuations. A sudden change in the contour of the response surface appears near 80 mmHg, and this sudden change is thought to indicate a breakdown in homeostasis.

[0089] The distribution of the plots is white noise in the correlation diagram between SBP and GT of LTM in FIG. 19(a), and R ２ A moderate correlation of 0.4541 was observed.

[0090] FIG. 20(a) shows a correlation diagram between GT of LTM and GT of STM. However, different types of plots are shown for different blood pressure ranges according to the classification defined in the "Guidelines for the Treatment of Hypertension 2019 (JSH2019)" (Published by the Japanese Society of Hypertension). In accordance with the WHO criteria, systolic blood pressure of 100 mmHg or less was classified as "hypotension." The distribution of the plots is a white noise state overall, but the hypertension state and the normotensive state are approximated by a quadratic function curve, and the hypotension, high-normal blood pressure, and high-normal blood pressure are approximated by a linear line. There is a moderate correlation, R ２= 0.3806-0.4858. Therefore, it can be said that SBP is determined by GT of LTM and GT of STM and shows a correlation with a three-dimensional response surface. Furthermore, it is under integrated regulation by the long-term blood pressure regulatory mechanism (LTM) and the short-term blood pressure regulatory mechanism (STM). In normotensive individuals, the principle of minimum energy action applies when the GT of LTM value is near -0.5, and in hypertensive individuals, the GT of LTM value is near 0, suggesting the possibility of an efficient frequency gradient (fluctuation) for arterial pressure regulation.

[0091] 20(b) is a correlation diagram between each estimated systolic blood pressure (eSBP) calculated by substituting x = first circulatory dynamics index: GT of LTM and y = second circulatory dynamics index: GT of STM into the fourth-order polynomial equation representing the three-dimensional response surface model of FIG. 18 for each of the 171 subjects, and each measured systolic blood pressure (SBP). ２ = 0.7468, showing a high correlation between the two.

[0092] As described above, the estimation index calculation unit 20 obtains time-series data on the instantaneous heart rate of the subject of blood pressure estimation, and executes a computer program comprising the time-series data analysis unit 11 to the circulatory dynamics index determination unit 18 of the database construction unit 10 to determine the first circulatory dynamics index: GT of LTM and the second circulatory dynamics index: GT of STM. The systolic blood pressure estimation unit 30 applies these to the three-dimensional response surface model (a polynomial representing the three-dimensional response surface model) as the first circulatory dynamics index for estimation: GT of LTM and the second circulatory dynamics index for estimation: GT of STM, respectively, thereby making it possible to obtain a highly accurate estimate of the systolic blood pressure for the subject of blood pressure estimation.

[0093] (Experimental Example 2) A blood pressure measurement experiment was conducted over a one-year period on a 34-year-old subject who maintained normal blood pressure and a 67-year-old subject who maintained normal to elevated blood pressure through medication. The 34-year-old subject was measured 38 times, and the 67-year-old subject was measured 121 times. Systolic blood pressure (SBP) was measured using an upper arm sphygmomanometer, and an estimated systolic blood pressure (eSBP) was calculated using the fourth-order polynomial representing the three-dimensional response surface model shown in Figure 18 by executing the estimation index calculation unit 20 and the systolic blood pressure estimation unit 30. The results are shown in Figures 21(a) and 21(b).

[0094] As shown in Figures 21(a) and 21(b), for a 34-year-old subject, R ２ = 0.7667, for a 67-year-old subject, R ２ = 0.9013, indicating a high correlation. Therefore, it was found that the three-dimensional response surface model shown in Figure 18 (a polynomial representing the three-dimensional response surface model) can be used in place of actual measurement data to estimate fluctuations in systolic blood pressure of a specific individual over a specified period.

[0095] Second Embodiment In this embodiment, we focused on the existence of an overlap frequency (OLF) between the long-term blood pressure regulation mechanism (LTM) and the short-term blood pressure regulation mechanism (STM), where the frequency bands of the two mechanisms overlap. Figures 22(a) to 22(c) show images of the overlap frequency (OLF), and Figure 22(b) is the same as Figure 9. However, as shown in Figure 22(a), the modification degrees (vertical axis) of the long-term blood pressure regulation mechanism (LTM) and the short-term blood pressure regulation mechanism (STM) from the frequency (horizontal axis) decrease in opposite directions from the center of 0.017 Hz, and the long-term blood pressure regulation mechanism (LTM) has an upper limit (Upper Limit of LTM) and the short-term blood pressure regulation mechanism (STM) has a lower limit (Lower Limit of STM). More specifically, the inflection point IP on the approximate slope curve is the change point caused by the modification by the long-term blood pressure regulation mechanism (LTM) and the short-term blood pressure regulation mechanism (STM), and the intersection point IT is related to the overlap zone (OLF).

[0096] Therefore, it is essential that an overlap band (OLF) exists between the frequency band related to the long-term blood pressure regulation mechanism (LTM) and the frequency band related to the short-term blood pressure regulation mechanism (STM); if not, the data is presumed to contain outliers. Here, the phrase "an overlap band (OLF) exists between the frequency band related to the long-term blood pressure regulation mechanism (LTM) and the frequency band related to the short-term blood pressure regulation mechanism (STM) (OLF requirement)" specifically refers to the following: within the combined frequency band of the LTM and STM, the LTM and STM frequency bands are contiguous and straddle the boundary between the two frequency bands (see the analysis example of (c-1) in Figure 39(c)); the OLF frequency band exists from the boundary with either the LTM or STM frequency band (see the analysis example of (c-2) and (c-3) in Figure 39(c)); or the OLF frequency band exists contiguous between the LTM and STM frequency bands (see the analysis example of (c-4) in Figure 39(c)). For example, if the OLF frequency band is not contiguous with the LTM frequency band and is separated from it, the OLF requirement is not met.

[0097] When GT of LTM and GT of STM that do not satisfy the OLF requirements are selected, the GT of STM is used as the reference and the frequency band to be analyzed for the GT of LTM is reselected so that it is a frequency band that leads to OLF. Specifically, if the OLF requirements are not satisfied, the approximate curve of a quartic or quintic function calculated by the approximate curve calculation unit 13 may be inappropriate, or the approximate curve of a cubic or quartic function calculated by the approximate gradient curve setting unit 16 may be inappropriate. This is due to the inclusion of outliers in the plot display of the spectrogram obtained by the frequency analysis unit 12. In the spectrogram of the double logarithmic graph from 0.006 to 0.06 Hz used in the first and second embodiments, outliers are likely to occur on the low-frequency side. In particular, outliers are likely to occur at the three end points on the low-frequency side. If the three end points are connected by a parabola, it will not be an outlier, but if the approximate curve is a parabola passing through two end points (such as the example of approximate curve B-1 in Figure 38), the end point on the lowest frequency side is likely to be an outlier. If the approximate curve passing through the two end points is almost a straight line, it is not an outlier, but may reflect the action of a lower frequency regulatory system than LTM. Therefore, among the plots displayed in the spectrogram, the plot located on the lowest frequency side, which is likely to be an outlier, is excluded, and processing to determine the approximate curve is performed by the approximate curve calculation unit 13 or the approximate gradient curve setting unit 16.

[0098] Specifically, as shown in Fig. 23 , after the optimization approximate gradient curve selection unit 17 selects the frequency band to be analyzed for the optimized approximate gradient curve, or after the circulatory dynamics index determination unit 18 determines the first circulatory dynamics index (GT of LTM) and the second circulatory dynamics index (GT of STM), it is confirmed whether or not the overlap band (OLF) exists at a predetermined position. That is, whether or not the above-mentioned OLF requirement is met is determined by an OLF requirement determination unit 171 provided in the optimization approximate gradient curve selection unit 17. If it is met (if the overlap band (OLF) exists at a predetermined position), that is, if S30 of the flowchart in Fig. 24 is Yes, the first circulatory dynamics index (GT of LTM) and the second circulatory dynamics index (GT of STM) determined by the circulatory dynamics index determination unit 18 in S23 to S25 are used as they are. If the answer is No in S30, the process returns to S11, where the plot located at the lowest frequency side in the spectrogram is excluded, and the approximate curve calculation unit 13 calculates approximate curves for a quartic function and a quintic function, or the process returns to S22, where the plot located at the lowest frequency side in the spectrogram is excluded, and the approximate gradient curve setting unit 16 calculates approximate curves for a cubic function and a quartic function. Thereafter, the process is performed in the same manner as in the above embodiment, and the first circulatory dynamics index (GT of LTM) and the second circulatory dynamics index (GT of STM) are calculated by the circulatory dynamics index determination unit 18. It is then determined again whether the recalculation requirement is met, and the same process is repeated. Note that in the flowchart of FIG. 24, S11 to S16 indicate the execution of the same steps as S11 to S16 in the flowchart of FIG. 5, and S21 to S26 indicate the execution of the same steps as S21 to S26 in the flowchart of FIG. 6. If the answer is No in S30, it is optional to return to either the step of S11 where the approximate curve is calculated by the approximate curve calculation unit 13 or the step of S22 where the approximate curve is calculated by the approximate gradient curve setting unit 16. However, in terms of shortening the calculation time, it is preferable to give priority to returning to the step of S22 where the approximate curve is calculated by the approximate gradient curve setting unit 16. Note that the configuration of the second embodiment is exactly the same as that of the first embodiment except for the OLF requirement determination unit 171.

[0099] Here, we will discuss the relationship between inflection points, long-term vasopressor mechanisms (LTM), short-term vasopressor mechanisms (STM), and overlap bands (OLF). First, the slope of y = f(x) changes, and the slope of the curve at an inflection point is the slope of the tangent at that point. The average slope between two inflection points on a function y = f(x) is expressed in the form of a difference coefficient. The limiting slope where this difference becomes infinitesimally small is the derivative, or differential coefficient, of the function y = f(x) at the inflection point. If we plot the slope of the curve y = f(x) with the ordinate y', we obtain the first derivative of the original curve y = f(x), the slope curve y' = f'(x'). Furthermore, the derivative of y' = f'(x') gives the second derivative y'' = f''(x) of the original curve y = f(x). Since y'' = 0 at the inflection point of the curve y = f(x), the sign of f''(x) changes before and after the inflection point. The inflection point is the switching frequency of accommodation between LTM and STM. Furthermore, the extreme values ​​on the curve represent an equilibrium state of accommodation function when y' = 0 and the distance between the extreme values ​​approaches infinity. The equilibrium state of accommodation function is captured by the frequency at the intersection of the two tangent slopes, and this intersection frequency also becomes the switching frequency between LTM and STM. Starting from these switching frequencies, OLF appears between LTM and STM.

[0100] (Experimental Example 3) Analysis was performed using data from the same subject as in Experimental Example 1. FIGS. 25 to 27 show analysis examples in which a determination was made by the OLF requirement determination unit 171 of this embodiment. The subject in this analysis example was a subject with high blood pressure (subject (c-4) described below). First, as shown in FIG. 25(a), frequency analysis was performed by the frequency analysis unit 12, and for the obtained spectrogram, an approximate curve of a quartic function and an approximate curve of a quintic function were calculated by the approximate curve calculation unit 13. Approximate curve A in FIG. 25(a) is the approximate curve of the quartic function, and approximate curve B in FIG. 25(a) is the approximate curve of the quintic function.

[0101] Next, the best approximation curve selection unit 14 selects the best approximation error εn→0, and the distribution shape of the residual histogram is either a bilaterally symmetrical distribution or a distribution that is inclined to widen to the left or right. For each of the approximation curves in FIG. 25(a), as shown in FIG. 25(b), the residual histograms all have bilaterally symmetrical distributions, and all satisfy the conditions for the histogram distribution shape. Next, comparing the best approximation errors εn, the SD-εn correlation diagram in FIG. 25(b) shows that the εn closest to 0 is 0.0191. As a result, the quintic function approximation curve B is selected as the best approximation curve.

[0102] Next, as shown in FIG. 25(c), the inflection point / intersection extraction unit 15 extracts inflection points IP1, IP2, IP3 and tangent intersection points IT1, IT2 because the best approximation curve is a quintic function.

[0103] Next, the approximation gradient curve setting unit 16 sets approximation curves of a cubic function and a quartic function in the above frequency band corresponding to the combination of the inflection points IP1, IP2, and IP3, which are characteristic points, and the intersection points IT1 and IT2 of the tangents, which are the respective characteristic points of the best approximation curve, as well as two frequencies out of the lower limit of 0.006 Hz and the upper limit of 0.06 Hz of the arterial pressure-related frequency band. Figures 26(a) to 26(d) show examples of approximation curves (approximation gradient curves) of a cubic function in the operating band of the long-term blood pressure regulation mechanism (LTM) among the best approximation curves of the quintic function in Figure 25(c), and Figures 26(e) to 26(f) show examples of approximation curves (approximation gradient curves) of a quartic function in the operating band of the long-term blood pressure regulation mechanism (LTM) among the best approximation curves of the quintic function in Figure 25(c).

[0104] Figures 27(g) to (j) are examples of the best approximation curve of the quintic function in Figure 25(c) where an approximation curve (approximation gradient curve) of a cubic function is obtained in the operating range of the short-term blood pressure regulation mechanism (STM), and Figures 27(k) to (m) are examples of the best approximation curve of the quintic function in Figure 25(c) where an approximation curve (approximation gradient curve) of a quartic function is obtained in the operating range of the short-term blood pressure regulation mechanism (STM).

[0105] The approximate gradient curves shown in FIGS. 26 and 27 show only data that can be candidates for the optimized approximate gradient curve, excluding data in the frequency band under consideration where the histogram of residuals is neither symmetrical nor spreads out to the left or right and where the best approximation error εn is large.

[0106] Figure 28 shows an SD-εn correlation diagram for each of the approximate gradient curves shown in Figures 26 and 27. The optimized approximate gradient curve selection unit 17 selects an optimized approximate gradient curve using this SD-εn correlation diagram. First, the optimized approximate gradient curve selection unit 17 obtains three regression lines (No. 1) in the operating band of the long-term blood pressure regulation mechanism (LTM), the regression line (No. 2) in the operating band of the short-term blood pressure regulation mechanism (STM), and a regression line combining the two operating bands (No. 3) from the SD-εn correlation diagram, and checks whether each plot is close to one of the regression lines. The optimized approximate gradient curve selection unit 17 then selects data that meets the following conditions: a low standard deviation SD, a best approximation error εn close to 0, and proximity to each regression line. As a result, for the operating band of the long-term blood pressure regulation mechanism (LTM), the optimized approximate gradient curve selector 17 selects the frequency band from 0.006 Hz to IP2 in Figure 26(a), where the best approximation error εn is 0.0154, and selects the approximate gradient curve of a cubic function found in that analysis frequency band as the optimized approximate gradient curve.For the operating band of the short-term blood pressure regulation mechanism (STM), the optimized approximate gradient curve selector 17 selects the frequency band from IT2 to 0.06 Hz in Figure 27(m), where the best approximation error εn is 0.0059, and selects the approximate gradient curve of a quartic function found in that analysis frequency band as the optimized approximate gradient curve.

[0107] The OLF requirement determination unit 171 determines whether the analysis target frequency band of 0.006 Hz to IP2 for LTM and the analysis target frequency band of IT2 to 0.06 Hz for STM, which are selected as the optimized approximate gradient curves, satisfy the OLF requirement. Figure 29(a) shows the approximate gradient curve of a cubic function for the analysis target frequency band of 0.006 Hz to IP2 for LTM, the approximate gradient curve of a quartic function for IT2 to 0.06 Hz for STM, and the best approximate curve of a quintic function. Figure 29(b) shows the analysis target frequency band of LTM, the analysis target frequency band of STM, and the frequency band of OLF. As shown in Figure 29(b), in this analysis example, the analysis target frequency band of LTM, the frequency band of OLF, and the frequency band of STM are consecutive, from the lowest frequency, and satisfy the above-mentioned OLF requirement.

[0108] Therefore, for the long-term blood pressure regulation mechanism (LTM), the circulatory dynamics index determination unit 18 determines the frequency gradient (first circulatory dynamics index): GT of LTM, which is the gradient of the tangent to the inflection point of the optimized approximate gradient curve represented by a cubic function, as 1.4192, and for the short-term blood pressure regulation mechanism (STM), the frequency gradient (second circulatory dynamics index): GT of STM, which is the gradient of the bitangent to the optimized approximate gradient curve represented by a quartic function, as -0.0065.

[0109] Once the first circulatory dynamics index (GT of LTM) and the second circulatory dynamics index (GT of STM) have been obtained for all of the data for the subjects, the three-dimensional response surface model creation unit 19 creates a three-dimensional response surface model using the first circulatory dynamics index (GT of LTM) on the X-axis, the second circulatory dynamics index (GT of STM) on the Y-axis, and the systolic blood pressure (SBP) values ​​corresponding to the Z-axis, and stores the model in the database (DB) 40. The resulting three-dimensional response surface model is shown in FIG. 30. FIG. 31(a) is a correlation diagram illustrating the three-dimensional response surface model of FIG. 30 with GT of LTM on the horizontal axis and systolic blood pressure on the vertical axis, and FIG. 31(b) is a correlation diagram illustrating the three-dimensional response surface model of FIG. 30 with GT of STM on the horizontal axis and systolic blood pressure on the vertical axis.

[0110] The obtained three-dimensional response surface model was approximated by the following fourth-order polynomial: f(x,y) = -5.1399x 4-2.6893x 3 y -4.5494x 3 -0.82825x 2 y 2 -4.827x 2 y + 4.7529x 2 + 0.091724xy 3 + 2.207xy 2 + 0.55227xy + 12.9027x -0.18196y 4 -1.0256y 3 + 5.0227y 2 + 21.9606y + 135.0414

[0111] Figure 32(a) shows a correlation diagram between GT of LTM and GT of STM. Figure 32(b) shows a correlation diagram between each estimated systolic blood pressure (eSBP) calculated by substituting x = first circulatory dynamics index: GT of LTM and y = second circulatory dynamics index: GT of STM for each of the 171 subjects into the fourth-order polynomial equation showing the three-dimensional response surface model of Figure 30, and each measured systolic blood pressure (SBP). ２ = 0.8048, indicating a high correlation between the two. Furthermore, the coefficient of determination is higher than the results of the first embodiment shown in FIG. 20( b), reflecting the effect of the process of excluding outliers performed by the OLF requirement determination unit 171. Furthermore, as shown in the lower column of FIG. 32, the coefficient of determination for the hypertensive subjects is also slightly higher than the results of the first embodiment shown in FIG. 20.

[0112] Figures 33(a) to (c) show other analysis examples of this embodiment used to create the response surface model. Of these, Figure 33(c) shows analysis examples of one hypotensive subject (c-1), one normotensive subject (c-2), one high-normal blood pressure subject (c-3), one high-normal blood pressure subject (c-4), and two medicated hypertensive subjects (c-5, c-6), all classified according to the JSH2019HO classification and WHO criteria. As shown in the SD-εn correlation diagram in Figure 33(b) and Figure 33(c), the data shown in Figure 33 are all data that satisfy the selection conditions of the optimized approximate gradient curve selector 17 and the OLF requirements determined by the OLF requirement determiner 171. Figure 33(a) is the SBP-eSBP correlation diagram shown in Figure 32(b), with the data for subjects (c-1) to (c-6) in Figure 33(c) highlighted.

[0113] Figures 34 to 38 show detailed analysis examples of subjects (c-1), (c-2), (c-3), (c-5), and (c-6), respectively, and each shows the process of creating a spectrogram by the frequency analysis unit 12, calculating approximation curves of quartic and quintic functions by the approximation curve calculation unit 13, selecting the best approximation curve by the best approximation curve selection unit 14, extracting inflection points IPn and tangent intersection points ITn by the inflection point / intersection extraction unit 15, setting approximation curves of cubic and quartic functions in the operating bands of the long-term blood pressure regulation mechanism (LTM) and short-term blood pressure regulation mechanism (STM) by the approximation gradient curve setting unit 16, selecting an optimized approximation gradient curve by the optimization approximation gradient curve selection unit 17, and calculating each frequency gradient (GT of LTM, which is the first circulatory dynamics index, and GT of STM, which is the second circulatory dynamics index) in the long-term blood pressure regulation mechanism (LTM) and short-term blood pressure regulation mechanism (STM) by the circulatory dynamics index determination unit 18. In either case, the OLF requirement determination unit 171 determines whether the OLF requirements are satisfied in the optimized approximate gradient curve selection unit 17, and data that satisfies the OLF requirements is used. The analysis data of the high blood pressure subject (c-4) is the data shown in Figures 25 to 29.

[0114] (Experimental Example 4) Data from the 67-year-old subject in Experimental Example 2 was analyzed based on the processing by the OLF requirement determination unit 171 of the second embodiment. The results are shown in FIG. 39 . FIG. 39 (c) shows an example of analysis of 121 measurement data from the same subject, showing normal blood pressure (c-1), high-normal blood pressure (c-2), high blood pressure (c-3), and high blood pressure (c-4). This subject maintained normal to high blood pressure during drug treatment, but during drug suspension, the subject experienced hypertension of approximately 150 mmHg.

[0115] Fig. 39(a) shows a correlation diagram between the estimated systolic blood pressure (eSBP) obtained by substituting the first circulatory dynamics index (GT of LTM) and the second circulatory dynamics index (GT of STM) obtained from Fig. 39(c) into the polynomial equation of the second embodiment and each measured systolic blood pressure (SBP). The coefficient of determination of the regression line of this correlation diagram was 0.8974, indicating a significant correlation.

[0116] Figure 39(b) is an SD-εn correlation diagram using the data in Figure 39(c). These figures show that there is a difference in the distribution relationship between SD and εn during medication and during drug-free periods. That is, when receiving medication, the distribution of SD and εn is equivalent to that of normotensive individuals, and the best-fit curve is represented by a function with few extreme values ​​and little unevenness. On the other hand, during drug-free periods, the best-fit curve shows large unevenness. The best-fit curve for a hypertensive state, in which extreme values ​​occur in the frequency band of the short-term blood pressure regulation mechanism (STM), is thought to be a state in which the approximation function passes through the spectrum and εn → 0.

[0117] These findings support Guyton's hypothesis that the best fit curve changes in response to blood pressure values, and that arterial pressure operates according to the same principles as mechanical and electrical systems. Furthermore, the frequency gradients (GT of LTM, GT of STM) calculated in each of the above embodiments can be considered as mechanical and electrical parameters of the arterial pressure regulation system, and the optimized best fit gradient curve can be said to capture the general characteristics of the best fit curve in terms of the tangent gradient. Therefore, systolic blood pressure can be estimated using a fourth-order polynomial obtained from a three-dimensional response surface model, which is visualization information of the stochastic differential equation of the arterial pressure regulation system.

[0118] 1 Blood pressure estimation device 10 Database construction unit 11 Time series data analysis unit 12 Frequency analysis unit 13 Approximation curve calculation unit 14 Best approximation curve selection unit 15 Inflection point / intersection extraction unit 16 Approximation gradient curve setting unit 17 Optimization approximation gradient curve selection unit 171 OLF requirement determination unit 18 Cardiovascular dynamics index determination unit 19 Three-dimensional response surface model creation unit 20 Estimation index calculation unit 30 Systolic blood pressure estimation unit 40 Database (DB)

Claims

1. A blood pressure estimation device comprising: a database construction unit that analyzes time-series data of instantaneous heart rates of a predetermined number of subjects, determines a first circulatory dynamics index related to the long-term blood pressure regulation mechanism (LTM) and a second circulatory dynamics index related to the short-term blood pressure regulation mechanism (STM) in the arterial pressure-related frequency band, and constructs a database of a three-dimensional response surface model using the first circulatory dynamics index and the second circulatory dynamics index as explanatory variables and the measured data of the systolic blood pressure of each of the subjects as a target variable; an estimation index calculation unit that analyzes time-series data of instantaneous heart rates of subjects whose blood pressure is to be estimated, determines the first circulatory dynamics index and the second circulatory dynamics index of the subjects whose blood pressure is to be estimated, and sets these as a first estimation circulatory dynamics index and a second estimation circulatory dynamics index, respectively; and a systolic blood pressure estimation unit that applies the first estimation circulatory dynamics index and the second estimation circulatory dynamics index to the three-dimensional response surface model constructed by the database construction unit, and estimates the systolic blood pressure of the subjects whose blood pressure is to be estimated.

2. The database construction unit reconstructs a two-dimensional delay coordinate system by applying a Lorenz plot to the time series data of the instantaneous heart rate of the predetermined number of subjects, creates an attractor in a reference time range, and calculates reference angle information (θ ０ ), and an individual attractor in the two-dimensional delay coordinate system in an individual time range shorter than the reference time range, and individual angle information (θ ｉ ) and the difference (θ ｉ－０ ) is calculated sequentially while moving the individual time range, and the difference (θ ｉ－０ a time series data analysis unit for obtaining time series data of the difference (θ ｉ－０ a frequency analysis unit that performs frequency analysis on time series data of the arterial pressure-related frequency band and displays a spectral diagram on a double logarithmic graph; an approximation curve calculation unit that sets two types of approximation curves, a quartic function and a quintic function, for the arterial pressure-related frequency band in the spectral diagram; a best approximation curve selection unit that selects the best approximation curve to be used for analysis from the two types of approximation curves, the quartic function and the quintic function; an inflection point / intersection extraction unit that extracts inflection points and tangent intersections for the best approximation curve; and an approximation gradient curve setting unit that sets two types of approximation curves, a cubic function and a quartic function, as approximate gradient curves for each frequency band within the arterial pressure-related frequency band, which corresponds to all combinations of the frequencies of the inflection points, the frequencies of the tangent intersections, and the two frequencies selected from the lower and upper limits of the arterial pressure-related frequency band, in each of the operating bands of the long-term blood pressure regulation mechanism and the short-term blood pressure regulation mechanism.

2. The blood pressure estimation device of claim 1, further comprising: an optimized approximate gradient curve selection unit that selects an optimized approximate gradient curve to be used for analysis from two types of approximate gradient curves, a cubic function and a quartic function, for each of the operating bands of the long-term blood pressure regulation mechanism and the short-term blood pressure regulation mechanism; a circulatory dynamics index determination unit that sets the value of the frequency gradient of the optimized approximate gradient curve for the operating band of the long-term blood pressure regulation mechanism as the first circulatory dynamics index and the value of the frequency gradient of the optimized approximate gradient curve for the operating band of the short-term blood pressure regulation mechanism as the second circulatory dynamics index; and a three-dimensional response surface model creation unit that creates the three-dimensional response surface model using the first circulatory dynamics index and the second circulatory dynamics index.

3. The blood pressure estimation device according to claim 2, wherein the estimation index calculation unit analyzes time series data of the instantaneous heart rate of the subject whose blood pressure is to be estimated using the time series data analysis unit, the frequency analysis unit, the approximation curve calculation unit, the best approximation curve selection unit, the inflection point / intersection extraction unit, the approximation gradient curve setting unit, the optimized approximation gradient curve selection unit and the circulatory dynamics index determination unit to obtain a first circulatory dynamics index and a second circulatory dynamics index of the subject whose blood pressure is to be estimated, and sets these as the first circulatory dynamics index for estimation and the second circulatory dynamics index for estimation, respectively.

4. The blood pressure estimation device of claim 2, wherein the circulatory dynamics index determination unit determines the frequency gradient as the gradient of the tangent at the inflection point when the optimization approximate curve is an approximate gradient curve represented by the cubic function, and determines the frequency gradient as the gradient of the bitangent when the optimization approximate curve is an approximate gradient curve represented by the quartic function.

5. The blood pressure estimation device of claim 2, wherein the best approximation curve selection unit selects the best approximation curve using at least one of the following evaluation indices regarding the residuals between each of the two types of approximation curves, the quartic function and the quintic function, and the spectrogram: standard deviation of the residuals; average value of the absolute values ​​of the residuals; best approximation error, which is the difference between the standard deviation of the residuals and the average value of the absolute values ​​of the residuals; and distribution shape of the residuals.

6. The blood pressure estimation device of claim 2, wherein the optimized approximate gradient curve selection unit selects the optimized approximate gradient curve using at least one of the following evaluation indices regarding the residuals between each of the two types of approximate curves, the cubic function and the quartic function, and the spectrogram: standard deviation of the residuals; average value of the absolute values ​​of the residuals; best approximation error, which is the difference between the standard deviation of the residuals and the average value of the absolute values ​​of the residuals; distribution shape of the residuals; and correlation between the standard deviation of the residuals and the best approximation error.

7. The blood pressure estimation device of claim 2, wherein the optimized approximate gradient curve selection unit determines whether an overlap band (OLF) modified by both the long-term blood pressure regulation mechanism (LTM) and the short-term blood pressure regulation mechanism (STM) exists at a predetermined position in the selected optimized approximate gradient curve, and if it does not exist, recalculates the curve by excluding outliers included in the long-term blood pressure regulation mechanism (LTM) in the spectrogram.

8. A computer program that causes a computer to function as a blood pressure estimation device, comprising the steps of: analyzing time-series data of instantaneous heart rates of a predetermined number of subjects, determining a first circulatory dynamics index related to the long-term blood pressure regulation mechanism (LTM) and a second circulatory dynamics index related to the short-term blood pressure regulation mechanism (STM) in the arterial pressure-related frequency band, and constructing a database of a three-dimensional response surface model using the first circulatory dynamics index and the second circulatory dynamics index as explanatory variables and the measured data of the systolic blood pressure of each of the subjects as a target variable; analyzing time-series data of instantaneous heart rates of subjects whose blood pressure is to be estimated, determining the first circulatory dynamics index and the second circulatory dynamics index of each subject whose blood pressure is to be estimated, and setting them as a first circulatory dynamics index for estimation and a second circulatory dynamics index for estimation, respectively; and applying the first circulatory dynamics index and the second circulatory dynamics index for estimation to the three-dimensional response surface model, thereby estimating the systolic blood pressure of each subject whose blood pressure is to be estimated.

9. In the procedure for constructing the database, a two-dimensional delay coordinate system is reconstructed by applying a Lorenz plot to the time series data of the instantaneous heart rate of the predetermined number of subjects, an attractor in a reference time range is created, and reference angle information (θ ０ ), and an individual attractor in the two-dimensional delay coordinate system in an individual time range shorter than the reference time range, and individual angle information (θ ｉ ) and the difference (θ ｉ－０ ) is calculated sequentially while moving the individual time range, and the difference (θ ｉ－０ ) time series data, and the difference (θ ｉ－０ ) and displaying a spectral diagram on a double logarithmic graph; setting two types of approximation curves, a quartic function and a quintic function, for the arterial pressure-related frequency band in the spectral diagram; selecting the best approximation curve to be used for analysis from the two types of approximation curves, the quartic function and the quintic function; extracting inflection points and tangent intersections for the best approximation curve; setting two types of approximation curves, a cubic function and a quartic function, as approximation gradient curves for each frequency band within the arterial pressure-related frequency band, corresponding to all combinations of the frequencies of the inflection points, the frequencies of the tangent intersections, and two frequencies selected from the lower and upper limits of the arterial pressure-related frequency band, in each of the operating bands of the long-term blood pressure regulation mechanism and the short-term blood pressure regulation mechanism; selecting an optimized approximation gradient curve to be used for analysis from the two types of approximation gradient curves, the cubic function and the quartic function, in each of the operating bands of the long-term blood pressure regulation mechanism and the short-term blood pressure regulation mechanism.

9. The computer program according to claim 8, further comprising the steps of: setting the value of the frequency gradient of the optimized approximate gradient curve for the operating band of the long-term blood pressure regulation mechanism as the first circulatory dynamics index; and setting the value of the frequency gradient of the optimized approximate gradient curve for the operating band of the short-term blood pressure regulation mechanism as the second circulatory dynamics index; and creating the three-dimensional response surface model using the first circulatory dynamics index and the second circulatory dynamics index.

10. The computer program of claim 9, wherein the step of determining the estimated first circulatory dynamics index and estimated second circulatory dynamics index of the subject whose blood pressure is to be estimated comprises the steps of: determining time series data of the instantaneous heart rate of the subject whose blood pressure is to be estimated; performing the frequency analysis and displaying a spectrogram; setting the approximation curve; selecting the best approximation curve; extracting the inflection points and tangent intersections; setting the approximation gradient curve; and selecting the optimized approximation gradient curve; determining the first circulatory dynamics index and second circulatory dynamics index of the subject whose blood pressure is to be estimated, and setting them as the estimated first circulatory dynamics index and estimated second circulatory dynamics index, respectively.

11. A computer program as described in claim 9, wherein in the procedure for determining the first circulatory dynamics index and the second circulatory dynamics index, if the optimization approximate curve is an approximate gradient curve represented by the cubic function, the gradient of the tangent at the inflection point is taken as the frequency gradient, and if the optimization approximate curve is an approximate gradient curve represented by the quartic function, the gradient of the bitangent is taken as the frequency gradient.

12. The computer program according to claim 9, wherein the step of selecting the best approximation curve uses at least one of the following evaluation indices regarding the residuals between each of the two types of approximation curves, the quartic function and the quintic function, and the spectrogram: standard deviation of the residuals; average value of the absolute values ​​of the residuals; best approximation error, which is the difference between the standard deviation of the residuals and the average value of the absolute values ​​of the residuals; distribution shape of the residuals; and correlation between the standard deviation of the residuals and the best approximation error.

13. The computer program according to claim 9, wherein the step of selecting the optimized approximate gradient curve uses at least one of the following evaluation indices regarding the residuals between each of the two types of approximate curves, the cubic function and the quartic function, and the spectrogram: standard deviation of the residuals; average value of the absolute values ​​of the residuals; best approximation error, which is the difference between the standard deviation of the residuals and the average value of the absolute values ​​of the residuals; and distribution shape of the residuals.

14. A computer program as described in claim 9, wherein the step of selecting the optimized approximate gradient curve comprises determining whether or not an overlap band (OLF) modified by both the long-term blood pressure regulation mechanism (LTM) and the short-term blood pressure regulation mechanism (STM) exists at a predetermined position in the selected optimized approximate gradient curve, and if not, recalculating the spectrum diagram by excluding outliers included in the long-term blood pressure regulation mechanism (LTM).

15. A computer-readable recording medium on which the computer program according to claim 8 is recorded.

Images