A personalized blood pressure parameter modeling method based on hydrostatic pressure effect

Abstract

The application provides a personalized blood pressure parameter modeling method based on hydrostatic pressure effect, comprising: two simple postures are used to realize personalized blood pressure proportional coefficient modeling; based on the hydrostatic pressure effect, the change of the PPG waveform corresponding to a specific pressure change is obtained by changing the height of the measured position; by changing the relative height between the finger and the heart, the change of the PPG waveform corresponding to the pressure change is observed; the measured diastolic pressure and systolic pressure are brought into a blood pressure algorithm with the solved proportional coefficient k to solve the deviation coefficient; and then personalized blood pressure monitoring is carried out; compared with the existing blood pressure calibration method, the application can simplify the calibration process of the user, only needs one measurement, reduces the heavy calibration burden of the blood pressure estimation model, reduces the calculation complexity, realizes more general continuous blood pressure monitoring, is scientific and rigorous in design, and is suitable for popularization and application.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of non-invasive continuous arterial blood pressure monitoring, and particularly relates to a personalized blood pressure parameter modeling method based on hydrostatic pressure effect. BACKGROUND

[0002] In the prior art, the comfort, portability and continuity of traditional pressurized blood pressure devices are usually poor, so continuous and comfortable blood pressure monitoring technology using wearable devices has attracted more and more attention.

[0003] A photoplethysmography (PPG) sensor, as a commonly used sensor in wearable devices, can reflect the blood volume change at the measured position, and is therefore widely used in continuous blood pressure monitoring in cutting-edge scientific research. However, the fact that individual vascular elasticity and diameter differ, under the same systolic blood pressure (SBP) and diastolic blood pressure (DBP), the blood volume change is different, which means that the physiological signal obtained by PPG lacks a direct mapping relationship with the blood pressure reading, and it is difficult to monitor the blood pressure of different people through a general blood pressure parameter.

[0004] In order to solve the above problems, the prior art personnel tries to overcome in various ways, for example:

[0005] ① Some research works reduce the number of blood pressure measurements required for customized model training through transfer learning;

[0006] But this still needs 5-10 measurements, and each measurement needs to be at least 30 minutes apart or to be cold pressure tested or to be physically exercised to ensure that the blood pressure changes significantly, which is complicated and inefficient.

[0007] ② Some research works try to learn individual information such as gender, age, height, weight, etc. through deep learning models, and take a relatively general blood pressure model within a range to estimate the blood pressure of each person;

[0008] However, the differences in personal information entry cannot reflect the individual differences in blood vessels, resulting in low accuracy of these research works.

[0009] ③ The relative relationship between individual blood pressure and PPG data is modeled through hemodynamic modeling, and then a small number of blood pressure measurements can also be used to calibrate the individual blood pressure coefficient to realize continuous blood pressure monitoring.

[0010] Although this method reduces the number of measurements required for calibration compared with the deep learning method, it still needs at least 3 blood pressure measurements for calibration of 4 proportional coefficients and 2 bias coefficients, and the operation is still complicated and not easy to popularize and apply. SUMMARY

[0011] In order to solve the above technical problems, the application provides a personalized blood pressure parameter modeling method based on hydrostatic pressure effect, which reduces the calibration burden of the traditional blood pressure monitoring method from multiple measurements to one blood pressure measurement, is simple and convenient to operate, can be applied in a blood pressure monitoring algorithm or a blood pressure monitoring deep learning model based on a PPG sensor, realizes non-invasive, continuous, comfortable and personalized blood pressure monitoring, has high detection result accuracy, and is suitable for promotion.

[0012] A personalized blood pressure parameter modeling method based on hydrostatic pressure effect comprises the following specific technical design:

[0013] Step one, two simple postures are used to realize personalized blood pressure proportional coefficient modeling;

[0014] Based on the hydrostatic pressure effect, the change of the PPG waveform corresponding to a specific pressure change is obtained by changing the height of the measured position;

[0015] As an example, due to the difference in blood vessel elasticity and diameter of individuals, the change of the PPG waveform is different for different people, so that the PPG waveform is suitable for modeling as personalized blood pressure parameters.

[0016] In a large number of scientific and rigorous experiments, it is observed that two simple postures, i.e. placing a finger at a position level with the heart and a position lower than the heart, can produce better modeling performance;

[0017] As an example, the position lower than the heart is preferably a position level with the navel.

[0018] According to the two simple postures, candidate features related to the proportional coefficient k are extracted for personalized blood pressure proportional coefficient modeling; wherein:

[0019] represents the hydrostatic pressure difference between two specific heights and .

[0020] and represent the direct current components of the PPG signals collected at the positions and .

[0021] represents the direct current component difference between the positions and .

[0022] and represent the diastolic time of the PPG signal segments collected at the positions and .

[0023] and represent the amplitudes of the PPG signal segments collected at the and ;

[0024] A multi-layer perceptron (MLP) model is used to learn the relationship between the scaling factor k and the candidate feature set :

[0025]

[0026] Table 1: Scaling factor modeling features for 2 typical blood pressure features

[0027]

[0028] For each scaling factor k, the input features are carefully selected from the candidate features according to their intrinsic relationship or through feature engineering methods, as shown in Table 1:

[0029] Blood pressure feature 's scaling factor represents the proportional relationship between the direct current component change and the blood pressure change. When modeling this scaling factor, features and are used to capture these relationships;

[0030] For the scaling factor that describes the complex relationship between blood pressure and diastolic time , in addition to the directly related features: , , and , other potential related features are determined through feature engineering methods: , and ;

[0031] As an example, the feature selection process for other parameter modeling is similar to the scaling factor k construction features of the 2 typical blood pressure features in Table 1 (same or consistent).

[0032] During the training of the multi-layer perceptron model, multiple blood pressure measurements are input into the blood pressure algorithm (Formula 2) to obtain the scaling factor k of each feature as the true value for supervised training;

[0033]

[0034] where PP is the pulse pressure, systolic pressure can be calculated by diastolic pressure plus pulse pressure, and are the scaling factors of personalized blood pressure features DC and T, respectively;

[0035] After the multi-layer perceptron model is trained, only the features of the new user need to be input into the trained multi-layer perceptron model to obtain the personalized proportion coefficient k.

[0036] Step two, one blood pressure measurement to realize bias coefficient calibration;

[0037] By changing the relative height between the finger and the heart, the PPG waveform change corresponding to the pressure change is observed; however, to accurately estimate the absolute value of blood pressure, one blood pressure measurement must be performed to determine the bias coefficient of diastolic pressure and systolic pressure;

[0038] Specifically, the measured diastolic pressure and systolic pressure are brought into the blood pressure algorithm with the solved proportion coefficient k to solve the bias coefficient;

[0039] Step three, personalized blood pressure monitoring;

[0040] The personalized proportion coefficient and bias coefficient obtained in steps one and two are brought into the blood pressure algorithm, so that the diastolic pressure and pulse pressure can be calculated according to the features DC and T in the PPG signal segment of the user, to realize personalized blood pressure monitoring.

[0041] As an example, an intelligent wearable device based on the hydrostatic pressure effect personalized blood pressure parameter modeling method, comprising: a wearable device and a PPG sensor, the PPG sensor is arranged in the wearable device, the PPG sensor is used to collect PPG data of two simple postures, to extract features related to blood vessel elasticity and diameter;

[0042] As an example, the intelligent wearable device is, for example, a smart ring, a smart watch or a smart bracelet.

[0043] The beneficial effects of the present application are:

[0044] Compared with the existing blood pressure calibration method (i.e. usually more than 5 times of blood pressure measurement is needed, including cold pressure test, measurement after physical exercise or at least interval of 30 minutes of rest measurement), our method can simplify the calibration process of the user, only one blood pressure measurement is needed, which reduces the problem of heavy burden of blood pressure estimation model calibration, reduces the calculation complexity, and realizes more general continuous blood pressure monitoring. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 The flow architecture principle diagram of the present application based on the hydrostatic pressure effect personalized blood pressure parameter modeling method. DETAILED DESCRIPTION

[0046] Next, reference will be made to Figure 1As shown, a personalized blood pressure parameter modeling method based on hydrostatic pressure effect includes the following specific technical design:

[0047] A personalized blood pressure parameter modeling method based on hydrostatic pressure effect includes the following specific technical design:

[0048] Step one, two simple gestures to achieve personalized blood pressure proportional coefficient modeling;

[0049] Based on the hydrostatic pressure effect, by changing the different heights of the measured position, the changes of PPG waveform corresponding to specific pressure changes are obtained;

[0050] As an example, due to the differences in individual vascular elasticity and diameter, the changes of PPG waveform are different for different people, so that the PPG waveform is suitable for modeling as personalized blood pressure parameters.

[0051] In a large number of scientific and rigorous experiments, it is observed that two simple gestures, i.e. placing fingers at the level of the heart and below the heart, can produce better modeling performance;

[0052] As an example, the position below the heart is preferably the position at the level of the navel.

[0053] According to the two simple gestures, candidate features related to the proportional coefficient k are extracted for personalized blood pressure proportional coefficient modeling; wherein:

[0054] represents the hydrostatic pressure difference between two specific heights and .

[0055] and represent the direct current components of the PPG signals collected at the and .

[0056] represents the direct current component difference between the and .

[0057] and represent the diastolic time of the PPG signal segments collected at the and .

[0058] and represent the amplitudes of the PPG signal segments collected at the and .

[0059] A multi-layer perceptron (MLP) model is used to learn the relationship between the scaling factor k and the selected feature set

[0060]

[0061] Table 1: Scaling factor modeling features for 2 typical blood pressure features

[0062]

[0063] For each scaling factor k, the input features are carefully selected from the candidate features according to their intrinsic relationship or through feature engineering methods, as shown in Table 1:

[0064] Blood pressure feature Scaling factor: represents the proportional relationship between the direct current component change and the blood pressure change. When modeling the scaling factor, features and are used to capture these relationships;

[0065] For the scaling factor that describes the complex relationship between blood pressure and diastolic time , in addition to the directly related features: , , and , other potential related features are determined through feature engineering methods: , and ;

[0066] As an example, the feature selection process for other parameter modeling is similar to the scaling factor k construction features of the two typical blood pressure features in Table 1.

[0067] During the training of the multi-layer perceptron model, multiple blood pressure measurements are input into the blood pressure algorithm (Formula 2) to obtain the scaling factor k of each feature as the true value of supervised training;

[0068]

[0069] where PP is the pulse pressure, systolic pressure can be calculated from diastolic pressure plus pulse pressure, and are the scaling factors of personalized blood pressure features DC and T, respectively;

[0070] After the multi-layer perceptron model is trained, only the features of the new user need to be input into the trained multi-layer perceptron model to obtain the personalized scaling factor k;

[0071] Step 2: One blood pressure measurement to realize bias coefficient calibration ​

[0072] By changing the relative height between the finger and the heart, the PPG waveform changes corresponding to the pressure changes are observed; however, to accurately estimate the absolute value of blood pressure, a blood pressure measurement must be taken to determine the bias coefficient of diastolic and systolic pressure;

[0073] Specifically, the measured diastolic and systolic pressures and the solved proportional coefficient k are brought into the blood pressure algorithm to solve the bias coefficient;

[0074] Step three, personalized blood pressure monitoring;

[0075] The personalized proportional coefficient and bias coefficient obtained in steps one and two are brought into the blood pressure algorithm, so that the diastolic pressure and pulse pressure can be calculated according to the blood pressure characteristics DC and T in the PPG signal segment of the user, and personalized blood pressure monitoring is realized.

[0076] As an example, an intelligent wearable device based on a personalized blood pressure parameter modeling method based on hydrostatic pressure effect, comprising: a wearable device and a PPG sensor, the PPG sensor is arranged in the wearable device, the PPG sensor is used to collect PPG data of two simple postures, and features related to blood vessel elasticity and diameter are extracted;

[0077] As an example, the intelligent wearable device is, for example, a smart ring, a smart watch or a smart bracelet.

[0078] In order to better illustrate the design principle of the present application, the accompanying drawings of the specification are now combined Figure 1 Brief introduction, again comb the principle characteristics of the present application:

[0079] Example 1:

[0080] Firstly, the user will be required to take a blood pressure measurement after wearing the wearable device (such as a smart ring), and two actions will be taken according to two simple postures, namely lifting and lowering the measured part to two different heights;

[0081] Secondly, the PPG sensor in the ring collects PPG data of two simple postures, which is used to extract features related to blood vessel elasticity and diameter;

[0082] Then, these features will be input into the trained MLP for modeling personalized proportional coefficient k;

[0083] Subsequently, these modeled personalized proportional coefficients k and the blood pressure measurement value will be brought into the blood pressure algorithm to obtain the personalized bias coefficient;

[0084] Finally, the personalized scale factor k and the personalized bias factor are brought into the blood pressure algorithm (formula 2), that is, the diastolic pressure and the pulse pressure can be calculated according to the features DC and T in the PPG signal segment of the user, and the accurate monitoring of the personalized blood pressure is realized.

[0085] Embodiment 2:

[0086] The international standard AAMI for evaluating electronic sphygmomanometers issued by the American Medical Device Promotion Association is adopted, and a comparative analysis report of experimental results of the technical scheme of the application and the prior art scheme is obtained:

[0087] Table 2: Comparison of experimental results

[0088]

[0089] Wherein: AAMI is the international standard for evaluating electronic sphygmomanometers issued by the American Medical Device Promotion Association; ME is the average error, and STD is the standard deviation; the blood pressure algorithm adopts the formula 2, and four personal information (age, gender, height, and weight) and candidate features are simultaneously input into the MLP of the application for modeling the personalized scale factor and the prior art scheme:

[0090]

[0091]

[0092] In the AAMI standard, the of formula 3 and formula 4 is: is the true measured blood pressure, is the sample number.

[0093] As can be seen from Table 2, the blood pressure algorithm using formula 2 can effectively meet the AAMI standard requirements in terms of ME and STD;

[0094] The method of using personal information in the traditional prior art is difficult to model the personalized scale factor, and the ME and STD thereof do not meet the AAMI standard requirements.

[0095] In order to better illustrate the architecture scheme of the application, the following is briefly described through knowledge introduction:

[0096] ① Hydrostatic pressure effect: Pascal's law is applicable to a uniform fluid medium such as the blood vessel system, according to which the pressure applied to the fluid is uniformly transmitted to any point of the fluid; this principle is also applicable to blood pressure; however, the flow of blood is affected by gravity, thereby generating a hydrostatic pressure effect, that is, the actual pressure of the artery is affected by the relative height from the heart. The pressure change caused by the hydrostatic pressure effect can be expressed as: wherein For the height offset of the measurement site relative to the heart, ρ represents the density of blood, and g represents the acceleration of gravity; generally, a height offset of 1 centimeter will result in a blood pressure change of about 0.775 mmHg.

[0097] ②Personalized blood pressure parameters: the input of the blood pressure monitoring algorithm or model contains a plurality of blood pressure related features, such as diastolic time , DC component of PPG, etc.; however, due to individual differences, the blood pressure features of each individual have different proportional coefficients and different bias coefficients; wherein the proportional coefficient describes the proportional or inverse proportional relationship of a blood pressure feature of the individual to the blood pressure, and the bias coefficient describes the baseline value of the blood pressure of the individual.

[0098] The present application focuses on designing a personalized blood pressure parameter modeling method using hydrostatic pressure effect, thereby simplifying the calibration process of the user and reducing the heavy burden of blood pressure monitoring algorithm and model calibration.

[0099] The above only describes the preferred embodiments of the present application, and it should be understood that the above description of the embodiments is only used to help understand the method of the present application and its core idea, and is not used to limit the protection scope of the present application, and any modification, equivalent replacement, etc. within the idea and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for modeling individualized blood pressure parameters based on hydrostatic pressure effect, characterized in that, The method comprises the following steps: Step 1: Two simple postures are used to realize personalized blood pressure proportional coefficient modeling; Based on the hydrostatic pressure effect, the change of the PPG waveform corresponding to a specific pressure change is obtained by changing the height of the measured position; According to the two simple postures, candidate features related to the proportional coefficient k are extracted for personalized blood pressure proportional coefficient modeling; wherein: represents the hydrostatic pressure difference between two specific heights and ​ and representing a direct current component of a PPG signal acquired at said and place; represents the difference between the direct current components of said and between said and represent the diastolic time of the PPG signal segment acquired at said and diastolic time of the PPG signal segment acquired at said and denotes the amplitude of a PPG signal segment acquired at said and denotes the amplitude of a PPG signal segment acquired at said A multi-layer perceptron model is employed to learn a relationship between a scaling factor k and a selected set of features Formula 1: between a scaling factor k and a selected set of features ， Blood pressure features a proportionality coefficient: representing the proportional relationship between the direct current component change and the blood pressure change, in modeling the proportionality coefficient of the blood pressure features and ;​ For describing blood pressure and diastolic time When considering the proportional coefficients of complex relationships between features, in addition to directly related features: , , and In addition, other potential relevant features were identified using feature engineering methods: , and ; During the training of the multi-layer perception model, the multiple blood pressure measurement values are brought into the blood pressure algorithm formula 2 to obtain the proportional coefficient k of each feature as the true value of the supervised training; ， wherein: PP is pulse pressure, and are the proportional coefficients of the individualized blood pressure characteristics DC and T, respectively. After the training of the multi-layer perception model is completed, only the features of the new user need to be input into the trained multi-layer perception model to obtain the personalized proportional coefficient k; Step 2: One blood pressure measurement is used to calibrate the bias coefficient; By changing the relative height between the finger and the heart, the PPG waveform change corresponding to the pressure change is observed; one blood pressure measurement is performed to determine the bias coefficient of diastolic pressure and systolic pressure; the measured diastolic pressure and systolic pressure and the solved proportional coefficient k are brought into the blood pressure algorithm to solve the bias coefficient; Step 3: Personalized blood pressure monitoring; The personalized proportional coefficient and bias coefficient obtained in steps 1 and 2 are brought into the blood pressure algorithm, and the diastolic pressure and pulse pressure are calculated according to the blood pressure characteristics DC and T in the PPG signal segment of the user, to realize personalized blood pressure monitoring.

2. The method of claim 1, wherein, Due to the differences in blood vessel elasticity and diameter of individuals, the change of the PPG waveform is different for different people, so that the PPG waveform is suitable for modeling as personalized blood pressure parameters.

3. The method of claim 1, wherein the method further comprises: The two simple postures are to place the finger at the level of the heart and below the heart.

4. The method of claim 3, wherein the method further comprises: The position below the heart is the position at the level of the navel.

5. The method of claim 1, wherein, The systolic pressure can be calculated by adding the diastolic pressure and the pulse pressure.

6. An intelligent wearable device for implementing the method of any one of claims 1-5, wherein the method is based on the hydrostatic effect. It comprises: A wearable device and a PPG sensor arranged in the wearable device, the PPG sensor is used to collect PPG data of two simple postures to extract features related to blood vessel elasticity and diameter. 7.The smart wearable device based on the hydrostatic pressure effect for individualized blood pressure parameter modeling method according to claim 6, characterized in that, The smart wearable device is a smart ring, a smart watch or a smart bracelet.

Images