JP6436663B2 - Blood pressure estimation device - Google Patents

Description

  The present invention relates to a blood pressure estimation device.

  There has been proposed a method of estimating a subject's blood pressure from a pulse wave feature value using a neural network learned based on a cuff sphygmomanometer measurement value and a pulse wave feature value (see Patent Document 1). .

JP 2000-126142 A

  However, the technique described in Patent Document 1 cannot accurately estimate the blood pressure of the subject. The present invention has been made in view of these problems, and an object thereof is to provide a blood pressure estimation device that can accurately estimate the blood pressure of a subject.

  The blood pressure estimation device of the present invention includes a pulse wave acquisition unit that acquires a pulse wave, a feature amount extraction unit that extracts a feature amount from the pulse wave, a blood pressure estimation unit that estimates a blood pressure from the feature amount using a model, A model creating / updating unit that creates or updates a model based on teacher data including personal information or feature quantity, and the model creating / updating unit is configured to create or update a model according to personal information or feature quantity. It is characterized by changing a learning threshold. The blood pressure estimation apparatus of the present invention can accurately estimate a subject's blood pressure.

1 is a block diagram illustrating a configuration of a blood pressure estimation device 1. FIG. It is a flowchart showing the blood-pressure estimation process which the blood-pressure estimation apparatus 1 performs. 3A, 3B, and 3C are explanatory diagrams illustrating examples of feature amounts. FIG. 4A is a graph showing the correlation between the estimated blood pressure obtained using only the first model and the cuff measurement blood pressure, and FIG. 4B shows the first model and the second model according to the age of the subject. It is a graph showing the correlation of the estimated blood pressure acquired by using properly and the cuff measurement blood pressure. It is a flowchart showing the blood-pressure estimation process which the blood-pressure estimation apparatus 1 performs. It is explanatory drawing showing an example of teacher data.

Embodiments to which the present invention is applied will be described below with reference to the drawings.
<First Embodiment>
1. Configuration of Blood Pressure Estimation Device 1 The configuration of the blood pressure estimation device 1 will be described with reference to FIG. The blood pressure estimation device 1 includes an interface 3, a storage unit 5, and a control unit 7.

The interface 3 transmits and receives information to and from the external terminal 101. The interface 3 is an example of a pulse wave acquisition unit and an age acquisition unit.
The storage unit 5 stores, stores, and reads information. The storage unit 5 can be configured by, for example, an HDD (hard disk drive).

  The control unit 7 is a known computer including a CPU, a RAM, a ROM, and the like. The control unit 7 functionally includes a feature amount extraction unit 9, a blood pressure estimation unit 11, and a model creation / update unit 13. The control part 7 performs the process mentioned later by each of these units.

  The terminal 101 is connected to a sphygmomanometer 103, a caffres sphygmomanometer 105, and an input unit 107. The terminal 101 relays transmission / reception of information between these and the blood pressure estimation apparatus 1. The sphygmomanometer 101 measures the accurate blood pressure of the subject using a cuff. Hereinafter, the blood pressure measured by the sphygmomanometer 101 is referred to as cuff measurement blood pressure. The cuff measurement blood pressure is transmitted to the blood pressure estimation device 1 via the terminal 101.

  The cufflink sphygmomanometer 105 estimates the blood pressure of the subject using the blood pressure estimation device 1. That is, the caffres sphygmomanometer 105 acquires the pulse wave of the subject and transmits the pulse wave to the blood pressure estimation device 1 via the terminal 101. A known pulse wave sensor can be used for acquiring the pulse wave. Examples of the pulse wave sensor include the following. The pulse wave sensor includes a light emitting diode that emits green light (light including a wavelength of 5000 to 8000 mm), and irradiates the fingertip or the like of the subject with the light. In addition, the pulse wave sensor receives reflected light of the irradiated light reflected in the capillary of the patient's finger with a photodiode, and outputs an electrical signal (pulse wave) corresponding to the received light.

  The blood pressure estimation device 1 estimates the blood pressure of the subject based on the pulse wave transmitted from the cufflink sphygmomanometer 105, as will be described later. Hereinafter, the blood pressure estimated by the blood pressure estimation device 1 is referred to as an estimated blood pressure. The blood pressure estimation device 1 transmits the estimated blood pressure to the caffres sphygmomanometer 105 via the terminal 101. As described above, the caffres sphygmomanometer 105 can obtain the estimated blood pressure of the subject from the pulse wave of the subject.

  The input unit 107 has a configuration in which a user (subject or another person) can input information, and the information input to the input unit 107 is transmitted to the blood pressure estimation device 1 via the terminal 101. Examples of information input to the input unit 107 include personal information (height, weight, age, sex, etc.) of the subject.

The blood pressure estimation device 1 can be realized by a cloud computer, for example.
2. Blood pressure estimation process executed by the blood pressure estimation apparatus 1 The blood pressure estimation process executed by the blood pressure estimation apparatus 1 (particularly the control unit 7) will be described with reference to FIGS. This process is performed when an instruction to execute the blood pressure estimation process is given from the caffres sphygmomanometer 105.

  In step 1 of FIG. 2, the personal information (height, weight, age, sex, etc.) of the subject is acquired from the terminal 101 using the interface 3. The personal information is input to the input unit 107 and transmitted to the blood pressure estimation apparatus 1 via the terminal 101.

  In step 2, the subject's pulse wave is acquired from the terminal 101 using the interface 3. This pulse wave is acquired by the cufflink sphygmomanometer 105 from the subject and transmitted to the blood pressure estimation device 1 via the terminal 101.

  In step 3, the feature quantity extraction unit 9 extracts a feature quantity (pulse wave feature quantity) from the pulse wave acquired in step 2. Examples of the feature amount include P0, P1, P2, and ΔT extracted from the pulse wave as shown in FIG. 3A. Further, as the feature amount, for example, as shown in FIG. 3B, there are a1 to f1 extracted from the first derivative (velocity pulse wave) of the pulse wave. Further, as the feature amount, for example, as shown in FIG. 3C, there are a to f extracted from the second-order differential (acceleration pulse wave) of the pulse wave. Examples of other pulse waves include PTT, HR, and volume AI.

  Returning to FIG. 2, in step 4, it is determined whether the age included in the personal information acquired in step 1 is less than 60 years old. If it is less than 60 years old, the process proceeds to step 5; In addition, 60 years old is an example of a threshold value.

  In step 5, the blood pressure estimation unit 11 estimates the blood pressure (estimated blood pressure) of the subject from the feature amount extracted in step 3 using the first model. In the first model, when a feature amount is input, an estimated blood pressure corresponding to the feature amount is output. The first model will be described later.

  In step 6, the blood pressure estimation unit 11 estimates the blood pressure (estimated blood pressure) of the subject using the second model from the feature amount extracted in step 3. In the second model, when a feature amount is input, an estimated blood pressure corresponding to the feature amount is output. The second model will be described later.

In step 7, the estimated blood pressure of the subject estimated in step 5 or 6 is transmitted to the caffres sphygmomanometer 105 via the terminal 101.
3. Model creation processing executed by blood pressure estimation device 1 (1) A method in which blood pressure estimation device 1 (particularly model creation / update unit 13) creates a first model will be described. First, teacher data is stored in the storage unit 5. The teacher data is composed of a plurality of unit data. Each unit data includes a subject's cuff measurement blood pressure, a plurality of feature amounts obtained simultaneously from the same subject, and personal information (height, weight, age, sex) of the same subject. Here, it is assumed that the teacher data is composed of n unit data. n is a sufficiently large natural number. The n unit data are distinguished by attaching identification numbers of 1, 2, 3,... i, i + 1,.

The teacher data for the same subject is obtained by using the sphygmomanometer 103 for the cuff measurement blood pressure of the subject, and by using the cufflink sphygmomanometer 105 and the blood pressure estimation device 1 (step 2, 3), and can be obtained by inputting the personal information of the subject to the input unit 107.
FIG. 6 shows an example of teacher data. Each unit data constituting the teacher data stores a blood pressure value measured by an electronic sphygmomanometer, personal information (height, weight, age, sex, etc.) and a feature amount. There are a plurality of unit data in the teacher data, which are obtained for various subjects.

  The model creation / update unit 13 creates a first model by support vector regression analysis based on the teacher data. The first model can be expressed as (Equation 1) below. Here, y is the estimated blood pressure of the subject, and x is the feature amount of the subject. W and b are model coefficients for representing y with respect to x.

上記（式１）は線形回帰式であって、推定式はε−インセンシティブ損失関数Ｌを用いて（式２）のように表される。式（４）〜（６）が成立する条件の下で、式（３）が最小となるように最適化問題を解くことで、（式１）におけるｗ、ｂが求められ、第１のモデルが得られる。

The above (Equation 1) is a linear regression equation, and the estimation equation is expressed as (Equation 2) using the ε-insensitive loss function L. By solving the optimization problem so that the expression (3) is minimized under the condition that the expressions (4) to (6) are satisfied, w and b in the expression (1) are obtained, and the first model is obtained. Is obtained.

なお、（式３）から（式６）において、ξとその複素共役はスラック変数であり、誤分類の度合いを表す変数である。また、（式３）の第１項はモデルの複雑さを表し、これが大きいほどモデルは複雑である。また、εは誤差許容度を表すパラメータであり、Ｃは適合度を表すパラメータである。εが小さいほど、第１のモデルの曖昧さが大きくなり、Ｃが大きいほど、第１のモデルの曖昧さが大きくなる。

In (Expression 3) to (Expression 6), ξ and its complex conjugate are slack variables, which are variables representing the degree of misclassification. In addition, the first term of (Expression 3) represents the complexity of the model, and the larger the value, the more complicated the model. Further, ε is a parameter representing the error tolerance, and C is a parameter representing the fitness. The smaller the ε, the greater the ambiguity of the first model, and the larger the C, the greater the ambiguity of the first model.

  Here, large ambiguity means learning weakly the relationship between cuff measured blood pressure and estimated blood pressure in model creation, and small ambiguity means relationship between cuff measured blood pressure and estimated blood pressure. Means to learn strongly.

In (Expression 3) to (Expression 6), i represents an identification number assigned to the unit data, and is any one of 1 to n.
(2) The blood pressure estimation device 1 creates a second model basically as in the case of the first model. However, in the support vector regression analysis, the parameter ε representing the error tolerance is smaller than that of the first model, and the parameter C representing the fitness is larger than that of the first model. Create a second model. As a result, the second model is more ambiguous than the first model.

  In the present embodiment, the above-described parameter ε representing the error tolerance and the parameter C representing the fitness are examples of learning determination thresholds. In this way, the ambiguity of the model can be adjusted by using the parameter ε representing the error tolerance and the parameter C representing the fitness as learning determination thresholds in model creation.

  (3) When the first model already exists, the blood pressure estimation device 1 newly creates the first model by the above-described method, and replaces the conventional first model with the new first model. (Update).

  The teacher data used at the time of update can be obtained by adding newly obtained unit data to the teacher data used when creating the conventional first model. Alternatively, the teacher data used at the time of updating may consist only of unit data obtained after creating the conventional first model.

  In addition, when the second model already exists, the blood pressure estimation device 1 newly creates the second model by the above-described method, and replaces the conventional second model with the new second model ( Update).

  The teacher data used at the time of updating can be obtained by adding newly obtained unit data to the teacher data used when creating the conventional second model. Alternatively, the teacher data used at the time of updating may consist only of unit data obtained after creating the conventional second model.

  The timing for updating the first model and the second model can be set as appropriate. For example, at the timing when a predetermined time has elapsed since the first model or the second model was created or updated last time, or when a predetermined amount of teacher data is accumulated, the first model, the second model, etc. Can be updated. Further, the first model and the second model can be updated in accordance with a user instruction.

4). Effects exhibited by the blood pressure estimation apparatus 1 According to the blood pressure estimation apparatus 1 described in detail above, the following effects can be obtained.
The blood pressure estimation device 1 can accurately estimate the blood pressure of the subject. This was confirmed by the following test. At the same time, the estimated blood pressure was acquired using the blood pressure estimation device 1 and the cuff measurement blood pressure was acquired using a cuff sphygmomanometer for the same subject. As a result, a set of the estimated blood pressure and the corresponding cuff measurement blood pressure can be obtained. A plurality of sets were acquired.

  Then, as shown in FIG. 4B, the plurality of sets are plotted in a graph in which the horizontal axis is the cuff measurement blood pressure and the vertical axis is the estimated blood pressure. When the error standard deviation of each plot was calculated, it was 9.2 mmHg, which was a small value. That is, the estimated blood pressure was small in variation and highly correlated with the cuff measurement blood pressure. Note that the feature values extracted from the pulse wave by the blood pressure estimation device 1 in this test were P0 to P2, a1 to f1, and a to f shown in FIGS. 3A to 3C.

  As a reference example, the estimated blood pressure was simultaneously acquired for the same subject using the blood pressure estimation device 1 and the cuff measured blood pressure was acquired using a cuff sphygmomanometer. However, here, the estimated blood pressure is acquired not by the above-described blood pressure estimation process but by a method that always uses the first model regardless of the age of the subject. Then, as shown in FIG. 4A, a plurality of sets of estimated blood pressures always obtained using the first model and corresponding cuff measured blood pressures are plotted on the graph. When the error standard deviation of each plot was calculated, it was 10.0 mmHg, which was a relatively large value. That is, the estimated blood pressure obtained by the method using the first model always has a larger variation in the estimated blood pressure than the method using the first model and the second model depending on the age of the subject. Correlation with cuff blood pressure was low.

The reason why the blood pressure estimation apparatus 1 can accurately estimate the blood pressure of the subject can be estimated as follows. An elderly subject has a large individual difference, and the relationship between blood pressure and pulse wave tends to collapse, so a model that allows ambiguity is desirable. Since the blood pressure estimation device 1 uses the second model having greater ambiguity than the first model for the elderly, an appropriate model according to the age of the subject can be used. As a result, it is considered that the estimated blood pressure can be estimated with high accuracy.
<Second Embodiment>
1. Configuration of Blood Pressure Estimation Device 1 The blood pressure estimation device 1 of the present embodiment has a configuration similar to that of the first embodiment.

2. The blood pressure estimation process performed by the blood pressure estimation apparatus 1 The blood pressure estimation process performed by the blood pressure estimation apparatus 1 (particularly the control unit 7) will be described with reference to FIG. This process is performed when an instruction to execute the process is given from the caffres sphygmomanometer 105.

  In step 11 of FIG. 5, the pulse wave of the subject is acquired from the terminal 101 using the interface 3. This pulse wave is acquired by the cufflink sphygmomanometer 105 from the subject and transmitted to the blood pressure estimation device 1 via the terminal 101.

In step 12, the feature quantity extraction unit 9 extracts a feature quantity (pulse wave feature quantity) from the pulse wave acquired in step 11. The feature amount is the same as that in the first embodiment.
In step 13, the blood pressure estimation unit 11 estimates the blood pressure (estimated blood pressure) of the subject from the feature amount extracted in step 12 using a model. In this model, when a feature amount is input, an estimated blood pressure corresponding to the feature amount is output. The model will be described later.

In step 14, the estimated blood pressure of the subject estimated in step 13 is transmitted to the caffres sphygmomanometer 105 via the terminal 101.
3. Model creation processing executed by blood pressure estimation device 1 (1) The blood pressure estimation device 1 of the present embodiment is basically the same as the method of creating the first model and the second model in the first embodiment. And create a model. However, in the present embodiment, as shown in (Expression 7) is used as ε which is a parameter representing the error tolerance, and as shown in (Expression 8) is used as C which is a parameter representing the fitness.

ここで、ε_０、Ｃ_０はそれぞれ定数であり、ageは教師データにおける被験者の年齢である。よって、本実施形態では、サポートベクター回帰分析において、教師データにおける年齢が高いほど、誤差許容度を表すパラメータεが小さく、適合度を表すパラメータＣが大きい、という条件の下でモデルを作成する。すなわち、本実施形態では、サポートベクター回帰分析において、教師データにおける年齢が高いほど、曖昧さを大きくする条件でモデルを作成する。

Here, ε ₀ and C ₀ are constants, and age is the age of the subject in the teacher data. Therefore, in the present embodiment, in the support vector regression analysis, a model is created under the condition that as the age in the teacher data is higher, the parameter ε representing the error tolerance is smaller and the parameter C representing the fitness is larger. That is, in this embodiment, in the support vector regression analysis, a model is created under a condition that the ambiguity increases as the age in the teacher data increases.

  In the present embodiment, the above-described parameter ε representing the error tolerance and the parameter C representing the fitness are examples of learning determination thresholds. These parameters vary depending on the age of the subject in the teacher data.

(2) When a model already exists, the blood pressure estimation apparatus 1 can newly create a model by the above-described method, and replace (update) the conventional model with the new model.
The teacher data used at the time of updating can be obtained by adding newly obtained unit data to the teacher data used when creating a conventional model. Or the teacher data used at the time of update may consist only of unit data obtained after creating a conventional model.

The timing for updating the model can be set as appropriate, and can be the same as in the first embodiment, for example.
4). Effects exhibited by the blood pressure estimation apparatus 1 According to the blood pressure estimation apparatus 1 described in detail above, the following effects can be obtained. That is, the blood pressure estimation apparatus 1 of the present embodiment can accurately estimate the blood pressure of the subject.
<Other embodiments>
As mentioned above, although embodiment of this invention was described, this invention can take a various form, without being limited to the said embodiment.

  (1) In step 4 in the first embodiment, the threshold age is not limited to 60 years, and can be set as appropriate. For example, it can be 55 years old, 65 years old, 70 years old, 75 years old, etc.

  (2) In the first embodiment, the parameter ε representing the error tolerance is the same as that of the first model, and the parameter C representing the fitness is larger than that of the first model. A second model may be created under Even in this case, substantially the same effects as those of the first embodiment can be obtained.

  In the first embodiment, the parameter ε representing the error tolerance is smaller than that of the first model, and the parameter C representing the fitness is the same as that of the first model. A second model may be created. Even in this case, substantially the same effects as those of the first embodiment can be obtained.

  (3) In the second embodiment, the parameter ε representing the error tolerance is constant regardless of the age in the teacher data, and the parameter C representing the fitness is larger as the age in the teacher data is higher. You may create a model with Even in this case, substantially the same effects as those of the second embodiment can be obtained.

  The model may be created under the condition that the parameter ε representing the error tolerance is smaller as the age in the teacher data is higher and the parameter C representing the fitness is constant regardless of the age in the teacher data. . Even in this case, substantially the same effects as those of the second embodiment can be obtained.

(4) In the first and second embodiments, the blood pressure estimation device 1 may perform only one of model creation and update.
(5) In the first and second embodiments, the blood pressure estimation device 1 may create or update a model based on teacher data by a method other than support vector regression analysis.

  (6) In the first and second embodiments, the learning determination threshold that is changed by the blood pressure estimation device 1 may be other than the parameter ε representing the error tolerance and the parameter C representing the fitness.

  (7) In the first and second embodiments, the blood pressure estimation device 1 determines a learning determination threshold (for example, a parameter ε representing an error tolerance or the like) according to the feature amount, not the age included in the teacher data. The parameter C) representing the degree of fitness may be changed.

  (8) In the first embodiment, when the first model is created or updated, the model can be created or updated in the same manner as in the second embodiment. That is, in support vector regression analysis, the first model is created or updated under the condition that the parameter ε representing the error tolerance is small and the parameter C representing the fitness is large as the age in the teacher data is high. Can do.

  In the first embodiment, when the second model is created or updated, the model can be created or updated by the same method as in the second embodiment. That is, in the support vector regression analysis, the second model is created or updated under the condition that the parameter ε representing the error tolerance is small and the parameter C representing the fitness is large as the age in the teacher data is high. Can do.

  (9) The functions of one component in the first and second embodiments may be distributed as a plurality of components, or the functions of a plurality of components may be integrated into one component. . For example, the blood pressure estimation device 1 may include a caffres sphygmomanometer 105. Further, the storage unit 5 may be in a device different from the control unit 7 and may be connected to the control unit 7 online. In addition, the blood pressure estimation device 1 may include one or more selected from the terminal 101, the sphygmomanometer 103, and the input unit 107.

  (10) Further, at least a part of the configuration of the first and second embodiments may be replaced with a known configuration having a similar function. Moreover, you may abbreviate | omit a part of structure of the said 1st, 2nd embodiment as long as a subject can be solved. Further, at least a part of the configuration of the first and second embodiments may be added to or replaced with the configuration of the other embodiments. In addition, all the aspects included in the technical idea specified only by the wording described in the claim are embodiment of this invention.

  (11) In addition to the blood pressure estimation device 1 described above, a system including the blood pressure estimation device 1 as a constituent element, a program for causing a computer to function as the blood pressure estimation device 1, a medium recording this program, a blood pressure estimation method, etc. The present invention can also be realized in various forms.

DESCRIPTION OF SYMBOLS 1 ... Blood pressure estimation apparatus, 3 ... Interface, 5 ... Memory | storage part, 7 ... Control part, 9 ... Feature-value extraction unit, 11 ... Blood pressure estimation unit, 13 ... Model creation / update unit, 101 ... Terminal, 103 ... Sphygmomanometer, 105 ... Caffres blood pressure monitor, 107 ... input

Claims (4)

A pulse wave acquisition unit (3) for acquiring a pulse wave;
A feature quantity extraction unit (9) for extracting feature quantities from the pulse wave;
A blood pressure estimation unit (11) for estimating blood pressure from the feature amount using a model;
The model is based on teacher data including personal information or the feature amount, and as the age in the teacher data is higher, (1) the parameter indicating the error tolerance is smaller and / or (2) the parameter indicating the fitness level A model creation / update unit (13) to create or update under the condition that
With
The model creation / update unit changes a learning determination threshold in creation or update of the model according to the personal information or the feature amount ,
The blood pressure estimation apparatus, wherein the learning determination threshold includes (1) a parameter representing an error tolerance and / or (2) a parameter representing a fitness . A pulse wave acquisition unit (3) for acquiring a pulse wave;
A feature quantity extraction unit (9) for extracting feature quantities from the pulse wave;
A blood pressure estimation unit (11) for estimating blood pressure from the feature amount using a model;
A model creation / update unit (13) for creating or updating the model based on teacher information including personal information or the feature quantity;
An age acquisition unit (3) for acquiring the age of the subject;
With
The model creation / update unit changes a learning determination threshold in creation or update of the model according to the personal information or the feature amount,
The learning threshold value includes (1) a parameter indicating an error tolerance and / or (2) a parameter indicating a fitness.
The model includes a first model and a second model having a greater ambiguity than the first model,
The blood pressure estimation unit uses the first model when the age of the subject is less than a predetermined threshold, and uses the second model when the age of the subject is equal to or greater than the threshold. Blood pressure estimation device. The blood pressure estimation device according to claim 2 ,
In the model creation / update unit, (1) a parameter representing an error tolerance is smaller than that of the first model, and / or (2) a parameter representing a goodness of fit is the first model. A blood pressure estimation device that creates or updates the second model under a condition that the second model is larger than. The blood pressure estimation device according to any one of claims 1 to 3 ,
The blood pressure estimation apparatus, wherein the model creation / update unit creates or updates the model by support vector regression analysis.

Images