CN115089147A

CN115089147A - Blood pressure measuring device

Info

Publication number: CN115089147A
Application number: CN202210555805.7A
Authority: CN
Inventors: 王怡; 盛奕冰
Original assignee: Yanhe Intelligent Technology Hangzhou Co ltd
Current assignee: Yanhe Intelligent Technology Hangzhou Co ltd
Priority date: 2022-05-20
Filing date: 2022-05-20
Publication date: 2022-09-23

Abstract

The invention discloses a blood pressure measuring device, which comprises a pressure air bag, a pressure sensor and a data processor, wherein the pressure sensor acquires an original pressure signal and transmits the original pressure signal to the data processor while the pressure air bag compresses a blood vessel; the data processor processes the raw pressure signal and obtains a blood pressure value as follows: separating a pulse wave signal from the raw pressure signal; extracting a feature value from the pulse wave signal; and taking the characteristic value as the input of a regression model to obtain a blood pressure value. According to the invention, through establishing a proper regression model, the influence of various morphological characteristics in the pulse wave signals is considered, and the measurement result obtained by the blood pressure measurement device is ensured to be more accurate.

Description

Blood pressure measuring device

Technical Field

The invention relates to a blood pressure measuring device, in particular to a blood pressure measuring device based on a regression model.

Background

Blood pressure is an important physiological parameter of human body, and has very important value in clinical diagnosis.

Currently, most of electronic sphygmomanometers on the market measure blood pressure by an oscillometric method, and then derive corresponding diastolic pressure (DBP) and systolic pressure (SBP) by identifying average pressure (ABP) and using a coefficient formula. For different electronic sphygmomanometers, the calculation formula can be subjected to various evolutions, and finally a mathematical relationship more suitable for the current electronic sphygmomanometer is obtained.

However, in practical applications, the above mathematical relationship is not only a single linear transformation formula, but also the characteristic values extracted from the pulse wave signals are different, and the blood pressure values obtained through the transformation formula with the same logic are also greatly different, and different influences of various factors need to be considered, so that it becomes more difficult and more necessary to process the acquired original signals and obtain accurate blood pressure values from the original signals.

Disclosure of Invention

The invention aims to provide a blood pressure measuring device, which is capable of ensuring that a measuring result obtained by the blood pressure measuring device is more accurate by establishing a proper regression model and considering the influence of various morphological characteristics in a pulse wave signal.

In order to realize the purpose of the invention, the invention adopts the following technical scheme: a blood pressure measuring device comprises a pressure air bag, a pressure sensor and a data processor, wherein the pressure air bag compresses a blood vessel, and the pressure sensor collects an original pressure signal and transmits the original pressure signal to the data processor; the data processor processes the raw pressure signal and obtains a blood pressure value as follows: separating a pulse wave signal from the original pressure signal; extracting a feature value from the pulse wave signal; and taking the characteristic value as the input of a regression model to obtain a blood pressure value.

The original pressure signal is formed by superposing a direct current pressure signal and an alternating current pulse wave signal, the direct current pressure signal and the alternating current pulse wave signal are separated through filtering processing, the direct current part is used for comparing actual pressure values, and the alternating current part is used for extracting pulse wave characteristics. And after the pulse wave signals are extracted, obtaining characteristic values from the pulse wave signals, and taking the characteristic values as the input of a regression model to obtain the blood pressure value.

Preferably, the characteristic value is extracted from the pulse wave signal, mainly including a pulse wave characteristic, a pressure characteristic and a time position characteristic, specifically, but not limited to, any one or more of the following combinations:

(a) ABP: the maximum pressure value is the pressure value corresponding to the maximum oscillation amplitude value in the pulse wave signal;

(b)ABP _p : location of maximum pressure value, point in time ABP at which ABP is _t And the sampling frequency fs, i.e. ABP _p ＝ABP _t *fs；

(c)SBP ₀ : the reference pressure value of systolic pressure is the proportional amplitude corresponding pressure value of the maximum pressure value ABP in the pulse wave signal, i.e. SBP ₀ ＝ABP*S _SBP Wherein S is _SBP Is the systolic pressure proportionality coefficient;

(d)SBP _p : the position of the reference pressure value of systolic pressure, in the descending segment of the pulse wave signal after the maximum pressure value ABP, SBP ₀ At the time point SBP _t And the product of the sampling frequency fs, i.e. SBP _p ＝SBP _t *fs；

(e)DBP ₀ : the reference pressure value of diastolic pressure is the proportional amplitude corresponding pressure value of the maximum pressure value ABP in the pulse wave signal, i.e. DBP ₀ ＝ABP*S _DBP Wherein S is _DBP Is the diastolic pressure proportionality coefficient;

(f)DBP _p : the position of the reference pressure value of diastolic pressure, the rising section of the pulse wave signal before the maximum pressure value ABP, DBP ₀ At the time point DBP _t And the product of the sampling frequency fs, i.e. DBP _p ＝DBP _t *fs；

(g) OSBP: symmetrical reference pressure values, namely pressure values corresponding to the proportional amplitude of the maximum pressure value ABP in the ascending section of the pulse wave signal before the maximum pressure value ABP;

(h)BP _p : the air bag compensates the air pressure value, the pressure value when the pressure air bag reaches the compensation point, and the position of the compensation point is determined by the pulse wave quantity and the amplitude in the pulse wave signal.

Preferably, the regression model is a multiple linear regression model. Multiple linear regression is for a case where one dependent variable is determined by multiple independent variables, i.e.

h _θ (x ₁ ,x ₂ ,...x _n )＝θ ₀ +θ ₁ x ₁ +...+θ _n x _n

In the present model, the mean square error is used as a loss function, i.e.

Wherein m is the number of samples, n is the characteristic number of the samples,

are true values.

Preferably, the regression model may also be a support vector machine regression model. Finding a hyperplane using support vector machine regression model (SVM regression model)

So that the sample points are fitted as much as possible to a linear model, i.e.

y _i ＝ω ^T x _i +b

In the present model, the error function ε is used as a loss function if y is predicted _i And true value

Difference between them

No loss is generated; otherwise the loss cost is

Preferably, the regression model is a random forest regression model. And the random forest regression model is used as a weak learner through a decision tree, and a final strong learning result is output after multiple iterations.

Assuming that the input data set D { (x1, y1), (x2, y2),. (xj, yj) }, the weak classifier is iterated multiple times, the number of times is T, and the output is the final strong classifier.

In the model, when T is 1,2, …, T, the training set is randomly sampled for the T time, and j times are collected in total to obtain a sampling set Dj containing j samples; training the jth decision tree model Gj (x) by using a sampling set Dj, selecting a part of sample characteristics from all sample characteristics on nodes when training the nodes of the decision tree model, and selecting an optimal characteristic from the randomly selected part of sample characteristics to divide left and right subtrees of the decision tree. If the model adopts classification algorithm to predict, the T weak learners put out the category or one of the categories with the most votes as the final category. And if the model is a regression algorithm, performing arithmetic mean on regression results obtained by the T weak learners to obtain a value which is the final model output. The algorithm of the random forest regression model can be selected according to actual requirements.

Preferably, the regression model adopts the following training method:

data acquisition: acquiring test data through the blood pressure measuring device, and acquiring reference data through a mercury sphygmomanometer; taking test data and corresponding reference data as combined data, and collecting a combined data S group;

wherein, the test data and the reference data in the group of combined data come from the same blood pressure measurer, the test data is an original pressure signal, and the reference data comprises a reference diastolic pressure value and a reference systolic pressure value;

data processing: separating from the test data to obtain pulse wave signals in each group of test data;

characteristic value extraction: respectively extracting single or a plurality of characteristic values from the S groups of pulse wave signals;

screening a data set: and taking the diastolic pressure value and the systolic pressure value in the reference data as label values, wherein the label values correspond to corresponding characteristic values in the combined data, and screening out a training set and a test set according to the ratio of the corresponding characteristic values in the S group and the label values.

Preferably, said training method is saidData set screeningAnd then, further comprising the following steps of: the characteristic value in each group of combined data is used as the input of a regression model to obtainPredicting a blood pressure value, the predicted blood pressure value comprising a predicted diastolic blood pressure value and a predicted systolic blood pressure value; and judging the prediction effect of the regression model according to the predicted blood pressure values and the reference blood pressure values in different groups.

Preferably, the verification of the predicted effect further comprises: obtaining S groups of blood pressure error values through the S groups of data: the difference value between the predicted blood pressure value and the reference data is a blood pressure error value, wherein the difference value between the predicted diastolic pressure and the reference diastolic pressure is a diastolic pressure error value, and the difference value between the predicted systolic pressure and the reference systolic pressure is a systolic pressure error value; and judging the prediction effect of the regression model according to the average value and/or the standard deviation of the error values.

Drawings

FIG. 1 is a flow chart of blood pressure measurement according to the present invention.

Fig. 2 is an exploded view of the raw pressure signal.

FIG. 3 is a graph of the training effect of the multiple linear regression model.

FIG. 4 is a graph of the predicted effect of a multiple linear regression model.

FIG. 5 is a diagram of the training effect of the regression model of the support vector machine.

FIG. 6 is a graph of the predicted effect of the regression model of the support vector machine.

FIG. 7 is a graph of the training effect of the random forest regression model.

FIG. 8 is a diagram of the predicted effect of the random forest regression model.

Detailed Description

The invention is further described below with reference to the accompanying drawings.

A first part: data processing in blood pressure measurement

The blood pressure measuring device shown in fig. 1 comprises a pressure air bag, a pressure sensor and a data processor, wherein the pressure air bag compresses a blood vessel, and the pressure sensor collects an original pressure signal and transmits the original pressure signal to the data processor; the data processor processes the raw pressure signal and obtains a blood pressure value as follows: the original pressure signal is formed by superposing a direct current pressure signal and an alternating current pulse wave signal, the direct current pressure signal and the alternating current pulse wave signal are separated through filtering processing, as shown in fig. 2, a direct current part is used for comparing actual pressure values, and an alternating current part is used for extracting pulse wave characteristics.

The method comprises the steps of obtaining characteristic values from pulse wave signals, using the characteristic values as the input of a regression model, and mainly using extracted characteristics to comprise one or any combination of the following eight characteristic values. The characteristic values are as follows:

(c)SBP ₀ : the reference pressure value of systolic pressure is the proportional amplitude corresponding pressure value of the maximum pressure value ABP in the pulse wave signal, i.e. SBP ₀ ＝ABP*S _SBP Wherein S is _SBP Is a systolic pressure proportionality coefficient;

(e)DBP ₀ : the reference pressure value of diastolic pressure is the proportional amplitude corresponding pressure value of the maximum pressure value ABP in the pulse wave signal, namely DBP ₀ ＝ABP*S _DBP Wherein S is _DBP Is the diastolic pressure proportionality coefficient;

(h)BP _p : the air bag compensates the air pressure value, when the pressure air bag reaches the pressure value of the compensation point, the position of the compensation point passes through the pulse wave signalThe number and amplitude of the pulse waves in (1) are determined.

And inputting the characteristic values into a regression model, wherein the regression model can process the characteristic values in a mode of adopting one model or a combination of a plurality of models of a multiple linear regression model, a support vector machine regression model and a random forest regression model to obtain a corresponding calculation result.

The processing formula or basic processing principle of the above three different regression models on the feature values is as follows:

multiple linear regression model:

wherein, P _xx Weight coefficients (e.g., P) corresponding to each feature _ABP Weight coefficient corresponding to ABP), and C is a constant term.

Support vector machine regression model:

(1) searching optimal parameters including a penalty function c and an RBF nuclear parameter g for the model;

(2) and training an SVM regression model through the training set and the optimal parameters, and storing the model for prediction. The model comprises important parameters of the SVM, such as support vectors and coefficients thereof, decision functions, kernel function types, iteration times and the like;

(3) and using the prediction set data and the model parameters in a prediction function, and finally outputting a prediction result.

Random forest regression model:

(1) training a random forest model by using training set data, and customizing the number ntree of trees and the number mtry of sample predictors at each split node;

(2) after the model is trained, the random forest model can evaluate the importance of the features to obtain an importance factor so that the model can better evaluate sample data;

(3) and finally, putting the test set data into a model for prediction to obtain a final regression result.

The above is a method for processing and calculating the original pressure signal in the actual blood pressure detection process of the blood pressure measuring device in the present embodiment.

A second part: training method of regression model

The training method of the regression model in this embodiment is described below.

The training method of the regression model mainly comprises data acquisition, data processing, characteristic value extraction, data set screening and prediction effect verification.

Data acquisition: acquiring test data through a blood pressure measuring device, and acquiring reference data through a mercury sphygmomanometer; taking test data and corresponding reference data as combined data, and collecting a combined data S group; wherein, the test data and the reference data in the group of combined data come from the same blood pressure measurer, the test data is an original pressure signal, and the reference data comprises a reference diastolic pressure value and a reference systolic pressure value.

In this embodiment, the blood pressure values of 88 blood pressure measuring persons participate in the training of the model, and three sets of data are acquired for the same blood pressure measuring person during the measurement, that is, each blood pressure measuring person needs to perform three measurements with the mercury sphygmomanometer and three measurements with the blood pressure measuring device of this embodiment, and finally data 264 sets are acquired.

Data processing: and (4) carrying out filtering and other processing on the test data in each group of combined data, namely the original pressure data, so as to obtain the pulse wave signals.

Characteristic value extraction: the 8 characteristic values are extracted from the pulse wave signals.

And (3) screening a data set: one group of data includes 8 feature values and corresponding tag values, the tag values are reference data in the combined data, 264 groups of data of the embodiment are screened by using a Kennard-stone algorithm, and meanwhile, the data are combined with the unprocessed 264 groups of combined data in the data acquisition process, and the ratio of 2: 1 the training set and the test set are selected, so 60% are training sets and 40% are test sets to train the regression model.

And (3) verifying the prediction effect: in order to ensure the validity of the prediction value of the regression model and the correlation of the features, the prediction result needs to be verified. Taking the characteristic value in each group of data in the data set as the input of a regression model to obtain a predicted blood pressure value, wherein the predicted blood pressure value comprises a predicted diastolic pressure value and a predicted systolic pressure value; and judging the prediction effect of the regression model according to the predicted blood pressure value and the reference blood pressure value in different groups.

Wherein, the verification of the prediction effect further comprises: obtaining S groups of blood pressure error values through the S groups of data: the difference value between the predicted blood pressure value and the reference data is a blood pressure error value, wherein the difference value between the predicted diastolic pressure and the reference diastolic pressure is a diastolic pressure error value, and the difference value between the predicted systolic pressure and the reference systolic pressure is a systolic pressure error value; and judging the prediction effect of the regression model according to the average value and/or standard deviation of the error values.

Systolic blood pressure was SBP and diastolic blood pressure was DBP.

The following describes in detail the verification of the prediction effect for different regression models.

In a multiple linear regression model, fitting was performed using the above data set, and the results of SBP validation in the multiple linear regression model were as follows:

R ² ＝0.5999；

F＝27.9283；

P＝3.788152001481342e-26

wherein R is ² Representing the goodness of fit, and expressing the learning rate of the regression function to the sample reference value, wherein the closer the value is to 1, the better the fitting effect is, the better the degree is, the more 60% in the regression model; f represents a significance test, and the larger the value of the significance test is, the more significant the regression is; the P value represents the error probability after receiving the regression equation, and the P value is extremely small in the regression model, which indicates that the fitting effect is good.

The results of the validation of DBP in the multiple linear regression model are as follows:

R ² ＝0.5381；

F＝21.6964；

P＝1.230098291834975e-21

these three parameters are as described above, in this regression model, R ² About 54%, the learning rate is moderate; the P value is extremely small, and the fitting is good.

For the blood pressure values in the reference data in the ith set of data,

the predicted blood pressure value of the ith group of data is the blood pressure error value x of the ith group of data _i Is composed of

By average of error values

And the standard deviation is used as a verification standard of the prediction effect of the regression model.

Wherein, the average value calculation formula is as follows:

the standard deviation is calculated as follows:

according to the formula, the prediction effect of the multiple linear regression model is calculated as follows:

mean_sbp_train:-1.7359e-14；std_sbp_train:8.48

mean_dbp_train:-2.6983e-15；std_dbp_train:7.44

mean_sbp_test:1.63；std_sbp_test:7.39

mean_dbp_test:1.35；std_dbp_test:6.53

wherein mean _ SBP _ train is the mean value of the errors of the training set SBP; std _ SBP _ train is the training set SBP error standard deviation; mean _ DBP _ train is the training set DBP error mean; std _ DBP _ train is the standard deviation of DBP error of the training set; mean _ SBP _ test is the average value of the test set SBP errors; std _ SBP _ test is the standard deviation of the test set SBP error; mean _ DBP _ test is the DBP error average value of the test set; std _ DBP _ test is the standard deviation of the DBP error for the test set.

FIG. 3 is a graph of the learning effect of the multiple linear regression model in the training set data, and FIG. 4 is a graph of the predicted effect of the multiple linear regression model in the test set data. From the above results, it can be seen that the multiple linear regression model has a small average error and a good prediction effect on blood pressure values.

In a support vector machine regression model (SVM regression model), the data needs to be normalized first, and then the optimal parameters (c and g) need to be further screened. Where c represents a penalty parameter and g represents a kernel function parameter. The default penalty parameter c is in the range [2^ (-8),2^8], and the default RBF core parameter g is in the range [2^ (-8),2^8 ].

In the SVM regression model of SBP, the optimal parameter c is 0.5, and g is 0.25; in the SVM regression model of DBP, the optimum parameter c is 8 and g is 0.25.

And after the optimal parameters are determined, performing model training by using training set data, and then using the model for verification on a test set.

Results of SBP model training: MSE 0.0167 (regression); r is ² ＝0.5970(regression)；

Results of DBP model training: MSE 0.0159 (regression); r is ² ＝0.6426(regression)；

Results of SBP model prediction: MSE is 0.0131 (regression); r is ² ＝0.5452(regression)；

The result of the DBP model prediction: MSE ═ 0.0160 (regression); r is ² ＝0.4304(regression)。

Where MSE is Mean squared error (Mean squared error), r ² For the Squared correlation coefficient (Squared correlation coefficient), the correlation formula is as follows:

where n is the sample size, f (x) _i ) Is the predicted value of the ith sample, y _i Is the reference value of the ith sample.

From the above data, the SVM regression model has higher correlation in the training set data, and the square correlation coefficient is about 0.6 or even above, and the specific result is shown in fig. 5. While the correlation in the test set data is slightly lower. But the mean square errors are all smaller, which shows that the model fitting effect is better, and the specific result is shown in fig. 6.

Meanwhile, as in the multiple linear regression equation, the average value and the standard deviation are used as the analysis criteria, and the results are as follows:

mean_sbp_train:-0.31；std_sbp_train:8.56

mean_dbp_train:-0.64；std_dbp_train:6.55

mean_sbp_test:1.42；std_sbp_test:7.46

mean_dbp_test:0.80；std_dbp_test:6.57

for the training process of the training set, the average error is smaller, the SBP standard error is 8.6, and the DBP standard error is 6.6; the verification results in the test set show that the average error is smaller, and the standard error is within 7.5, so that a better prediction result is obtained.

In a random forest regression model (RF regression model), the number of optimal decision trees is 50 and the number of variables of the binary tree is 2 for SBP, and the importance of the 8 features used are:

18.81；20.58；18.79；19.67；19.62；18.43；18.01；18.94

the corresponding characteristic sequence is as follows:

ABP，SBP ₀ ，DBP ₀ ，OSBP，BP _p ，SBP _p ，ABP _p ，DBP _p (the same applies below).

In the regression model of DBP, through parameter optimization, the number of the optimal decision trees is 100, and the number of variables in the binary tree is 2, and the importance of the 8 features used are respectively:

16.82；20.06；19.98；21.00；19.46；17.21；17.85；19.62

it can be seen that, with the RF regression model, the importance coefficients of each feature are relatively averaged, which indicates that in the model, the importance of the features is equivalent and all the features are better regression features. Therefore, the regression model shows a good learning effect on the training set, as shown in fig. 7.

The large bias in the test set occurs, which is a major disadvantage of the RF regression model, i.e. the overfitting phenomenon may occur, which may be due to the influence of the co-linearity of the features and the small overall data volume. Even after the data normalization process, this phenomenon still exists, as shown in fig. 8, and the regression results do not perform as well in the test set as the training set. The mean error and standard error are shown below:

mean_sbp_test:1.64；std_sbp_test:9.65

mean_dbp_test:1.68；std_dbp_test:8.11

in the process of actually establishing the regression model, the prediction effects of different regression models can be checked by adopting the method. For the use of different feature values, verification may be performed by means of the mean and/or standard deviation of the respective error values to select the most preferred regression model for predicting the blood pressure value.

In the above calculation results of this embodiment, it can be seen that the multiple linear regression model and the SVM regression model both show better prediction results, and the RF regression model is prone to overfitting. Meanwhile, in the multiple linear regression model and the SVM regression model, the correlation between the characteristic value and the label value is basically similar, and each characteristic also shows very similar characteristic importance when the RF regression model is established. Therefore, it can be seen from the verification process of the regression model that the above 8 feature values play an important role in blood pressure calculation, and have a certain correlation with blood pressure, and the correlation with SBP is higher than that with DBP.

And a third part: blood pressure measurement results

In this embodiment, the multiple linear regression model is applied to an actual test, 3 sets of combined data are collected by the same person in the test process according to the collection standard during the training of the early regression model, the test data is obtained by the blood pressure measuring device, and the reference data is obtained by the mercury sphygmomanometer. A total of 255 complete data sets were measured for 85 people.

The final test results using the foregoing formula are:

average error of SBP: 1.88; standard error of SBP: 8.76; DBP mean error: 2.02; standard error of DBP: 7.36.

the test result is basically consistent with the model prediction result established in the early stage, and the selected characteristic value has certain correlation with the reference blood pressure value. Therefore, in the embodiment of the present application, a more accurate blood pressure value can be predicted by using the selected 8 features.

The foregoing is a preferred embodiment of the present invention, and it will be apparent to those skilled in the art that various changes and modifications may be made without departing from the spirit of the invention, and these should be considered to be within the scope of the invention.

Claims

1. A blood pressure measuring device characterized in that: the device comprises a pressure air bag, a pressure sensor and a data processor, wherein the pressure sensor acquires an original pressure signal while the pressure air bag compresses a blood vessel and transmits the original pressure signal to the data processor; the data processor processes the raw pressure signal and obtains a blood pressure value as follows:

separating a pulse wave signal from the original pressure signal;

extracting a characteristic value from the pulse wave signal;

and taking the characteristic value as the input of a regression model to obtain a blood pressure value.

2. A blood pressure measuring device according to claim 1, characterized in that: the characteristic value comprises any one or more of the following combinations:

(b)ABP _p : location of maximum pressure value, time point ABP at which ABP is located _t And samplingProducts of frequency fs, i.e. ABP _p ＝ABP _t *fs；

(c)SBP ₀ : the reference pressure value of systolic pressure is proportional amplitude corresponding pressure value of maximum pressure value ABP in pulse wave signal, i.e. SBP ₀ ＝ABP*S _SBP Wherein S is _SBP Is a systolic pressure proportionality coefficient;

3. A blood pressure measuring device according to claim 2, wherein: the regression model adopts a multiple linear regression model.

4. A blood pressure measuring device according to claim 2, wherein: the regression model adopts a support vector machine regression model.

5. A blood pressure measuring device according to claim 2, wherein: the regression model adopts a random forest regression model.

6. A blood pressure measuring device according to any one of claims 1 to 5, wherein: the regression model adopts the following training method:

data acquisition: acquiring test data through the blood pressure measuring device, and acquiring reference data through a mercury sphygmomanometer; taking test data and corresponding reference data as combined data, and collecting N groups of combined data;

the test data and the reference data in the group of combined data come from the same blood pressure measurer, the test data is an original pressure signal, and the reference data comprises a reference diastolic pressure value and a reference systolic pressure value;

characteristic value extraction: respectively extracting single or a plurality of characteristic values from N groups of pulse wave signals;

and (3) screening a data set: and taking the diastolic pressure value and the systolic pressure value in the reference data as label values, wherein the label values correspond to corresponding characteristic values in the combined data, and screening out a training set and a test set according to the proportion of the corresponding characteristic values in the N groups and the label values.

7. A blood pressure measuring device according to claim 6, wherein: the said training methodData set sifter Sorting machineAnd then, further comprising the following steps of:

taking the characteristic value in each group of combined data as the input of a regression model to obtain a predicted blood pressure value, wherein the predicted blood pressure value comprises a predicted diastolic blood pressure value and a predicted systolic blood pressure value;

obtaining N groups of predicted blood pressure values through N groups of data, wherein the predicted blood pressure values correspond to the reference blood pressure values;

and judging the prediction effect of the regression model according to the predicted blood pressure values and the reference blood pressure values in different groups.

8. A blood pressure measuring device according to claim 7, wherein: the predictive effect verification further comprises:

obtaining N groups of blood pressure error values through N groups of data: the difference value between the predicted blood pressure value and the reference data is a blood pressure error value, wherein the difference value between the predicted diastolic pressure and the reference diastolic pressure is a diastolic pressure error value, and the difference value between the predicted systolic pressure and the reference systolic pressure is a systolic pressure error value;

and judging the prediction effect of the regression model according to the average value and/or standard deviation of the error value.