A blood pressure calculation method and a blood pressure measurement device based on blood pressure pulse waves

Abstract

A blood pressure calculation method based on blood pressure pulse waves provided by the present application is based on the original pressure pulse wave data of blood pressure, fits an amplitude curve, then performs resampling and normalization processing on the amplitude curve, extracts all key feature information required for calculating blood pressure values, without collecting other data for auxiliary calculation, calculates the blood pressure values of diastolic blood pressure and systolic blood pressure based on the pressure pulse wave, not only has higher accuracy, but also is more targeted; constructs a systolic blood pressure prediction model and a diastolic blood pressure prediction model based on a machine learning algorithm, trains the two models with labeled training data, and through feature extraction, respectively transforms the normalized resampling amplitude coefficients collected based on the original pressure pulse wave into input data applicable to the diastolic blood pressure prediction model and the systolic blood pressure prediction model, while improving the accuracy of blood pressure detection, and ensuring that the blood pressure calculation method of the present application is applicable to the calculation of different types of blood pressure data.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent medical device control, and specifically relates to a blood pressure calculation method based on blood pressure pulse waves. Background Art

[0002] For an oscillometric blood pressure measurement device, such as an oscillometric sphygmomanometer, it includes: a controller, a cuff wrapped around the arm, and an inflation device, and a pressure sensor is provided on the cuff. After the cuff is inflated by the inflation device and wrapped around the arm of the subject, the blood flow through the blood vessel has a certain oscillatory wave, that is, a blood pressure pulse wave, which is received by the pressure sensor and transmitted to the controller. During a complete measurement process, the controller automatically adjusts the inflation amount of the cuff wrapped around the upper arm, changes the gas pressure in the cuff, and according to the change of the pulse wave of the subject, the pressure and fluctuation detected by the pressure sensor also change accordingly. In the controller, the moment with the largest fluctuation is selected as the mean pressure, and based on the mean pressure, a certain value is searched forward for the systolic pressure and a certain value is searched backward for the diastolic pressure. Usually, two groups of data are collected in chronological order during a complete measurement process, and after fitting, a blood pressure pulse wave curve and a sphygmomanometer pressure curve are obtained respectively, specifically as Figure 1 shown.

[0003] In the prior art, the methods for calculating the systolic pressure and diastolic pressure from the waveform of the pulse wave include: the proportional coefficient method and the eigenvalue method.

[0004] The proportional coefficient method, that is, the amplitude method, there is a certain relationship between the mean pressure of the blood, the systolic pressure, and the diastolic pressure, and these two coefficients are called Ks and Kd.

[0005] SP / MP = Ks (value range: 0.3 to 0.75), DP / MP = Kd (value range: 0.45 to 0.90)

[0006] Where SP is the systolic pressure, MP is the mean pressure, and DP is the diastolic pressure, and the corresponding diastolic pressure and systolic pressure are calculated through the proportional coefficient. The calculation process of the proportional coefficient method is simple, but in the proportional coefficient method, the two data of Ks and Kd are the results obtained from big data statistics. When used for individual blood pressure calculation, such as calculating for hypertensive patients, due to the large individual differences and inconsistencies with the statistical values, the calculation result accuracy of the proportional coefficient method is not high.

[0007] The calculation of the eigenvalue method depends on whether there are characteristic changes at the corresponding points of the systolic pressure and diastolic pressure of the pulse wave. The eigenvalue method resolves the inflection points of the curve and establishes a model to analyze the data. Although the eigenvalue method can calculate based on individual characteristics, the calculation steps are relatively complex. At the same time, in the actually collected blood pressure pulse waves, there are often no obvious inflection points, so the accuracy will also be relatively low. Summary of the Invention

[0008] To solve the problem that the accuracy of the method for calculating systolic blood pressure and diastolic blood pressure based on the waveform of the pulse wave in the prior art is relatively low, the present invention provides a blood pressure calculation method based on the blood pressure pulse wave, which can calculate the blood based on individual characteristics, and the calculation result has a higher accuracy; at the same time, the calculation speed is faster. At the same time, the present application also discloses a blood pressure measurement device based on the blood pressure pulse wave.

[0009] The technical solution of the present invention is as follows: A blood pressure calculation method based on the blood pressure pulse wave, characterized in that it includes the following steps:

[0010] S1: Obtain the original measurement data of blood pressure detection, label the diastolic blood pressure and systolic blood pressure to obtain: the original data of the training sample;

[0011] The original data of the training sample includes: the original data of the systolic blood pressure sample and the original data of the diastolic blood pressure sample;

[0012] S2: Based on the original data of the training sample, extract the original pulse wave curve corresponding to the original measurement data; after correcting the error data, find all the peak points and valley points in the original pulse wave curve, and fit to obtain the peak envelope line and the valley envelope line;

[0013] Calculate and fit the original amplitude curve based on the peak envelope line and the valley envelope line;

[0014] S3: After removing the abnormal data in the original data of the training sample, resample the original amplitude curve based on the pressure value to complete the data enhancement operation of the original data of the training sample;

[0015] Then normalize the coefficients of the amplitude curve obtained after data enhancement to obtain the resampled amplitude curve coefficients;

[0016] S4: Perform feature extraction operations and dimensionality reduction operations on the resampled amplitude curve coefficients in sequence;

[0017] The obtained low-dimensional systolic blood pressure data is the systolic blood pressure sample data; the obtained low-dimensional diastolic blood pressure data is the diastolic blood pressure sample data;

[0018] S5: Generate a systolic blood pressure prediction model and a diastolic blood pressure prediction model respectively based on the machine learning algorithm;

[0019] S6: Use the systolic blood pressure sample data to train the systolic blood pressure prediction model, and use the diastolic blood pressure sample data to train the diastolic blood pressure prediction model to obtain the trained systolic blood pressure prediction model and the diastolic blood pressure prediction model respectively;

[0020] S7: Obtain the original blood pressure data to be measured, and extract the corresponding original pulse wave curve of the original blood pressure data to be measured; find all peak points and trough points in the original pulse wave curve, and fit to obtain a peak envelope line and a trough envelope line; calculate and fit to obtain an original amplitude curve based on the peak envelope line and the trough envelope line;

[0021] Perform pressure value resampling on the original amplitude curve, and then normalize the coefficients of the amplitude curve obtained after resampling to obtain resampled amplitude curve coefficients;

[0022] Obtain the resampled amplitude curve coefficients corresponding to the original blood pressure data to be measured, denoted as: amplitude curve coefficients to be calculated;

[0023] S8: Perform the feature extraction operation and the dimensionality reduction operation on the amplitude curve coefficients to be calculated in sequence, and obtain the low-dimensional systolic blood pressure data and the low-dimensional diastolic blood pressure data, denoted as: systolic blood pressure input data and diastolic blood pressure input data respectively;

[0024] S9: Input the systolic blood pressure input data into the systolic blood pressure prediction model, and input the diastolic blood pressure input data into the systolic blood pressure prediction model and the diastolic blood pressure prediction model to perform blood pressure prediction; the systolic blood pressure prediction model and the diastolic blood pressure prediction model respectively output a systolic blood pressure prediction value and a diastolic blood pressure prediction value, that is, obtain the measurement value corresponding to the original blood pressure data to be measured.

[0025] Its further feature lies in:

[0026] It further includes the following steps:

[0027] S10: Based on the amplitude curve coefficients to be calculated, the systolic blood pressure prediction value and the diastolic blood pressure prediction value, calculate the prediction value credibility corresponding to the systolic blood pressure prediction value and the diastolic blood pressure prediction value respectively; the calculation process includes:

[0028] a1: Calculate the prediction value accuracy X:

[0029]

[0030] where DIA ai is the diastolic blood pressure prediction value, SYS ai is the systolic blood pressure prediction value, DIA amplitude is the diastolic blood pressure calculated based on the amplitude method, SYS amplitude is the systolic blood pressure calculated based on the amplitude method; abs() is used to calculate the absolute value of the parameter;

[0031] a2: Calculate the prediction value credibility:

[0032]

[0033] Among them, accuracy is the confidence level of the predicted value; X is the accuracy of the predicted value;

[0034] S11: Use the systolic blood pressure predicted value, the diastolic blood pressure predicted value, and the confidence level of the predicted value as the final output values together;

[0035] In step S2, the fitting of the peak envelope line, the trough envelope line, and the original amplitude curve includes the following steps:

[0036] b1: Perform an operation to correct error data on the original pulse wave curve;

[0037] The operation to correct error data is: Consider all pressure values in the original pulse wave curve that are greater than the pressure data of the previous one as error data and delete them together with the corresponding pulse wave curve values to obtain a pulse wave curve for calculation;

[0038] b2: Extract the pulse wave curve for calculation, find all peak points, and use linear interpolation to fit to obtain the peak envelope line;

[0039] b3: Extract the pulse wave curve for calculation, find all trough points, and use linear interpolation to fit to obtain the trough envelope line;

[0040] b4: Place the peak envelope line and the trough envelope line in the same coordinate system, with the abscissa being time and the ordinate being the pressure value, and calculate the original amplitude curve; The abscissa of the original amplitude curve is time, and the ordinate is the ordinate of the peak envelope line minus the ordinate of the trough envelope line;

[0041] In step S3, the calculation method of the resampled amplitude curve coefficient includes the following steps:

[0042] c1: Intercept and resample the pressure values of the fitted original amplitude curve at intervals of 1 mmHg, starting from 0 mmHg and ending at X max mmHg; Based on the pressure values obtained after the pressure value resampling, construct a fitted curve again to obtain: a calculated fitted curve;

[0043] where X max is the maximum range of human blood pressure;

[0044] c2: Calculate the curve amplitude coefficient of the calculated fitted curve, denoted as: calculated amplitude coefficient;

[0045] c3: Normalize the calculated amplitude coefficient, and the normalization method is:

[0046] (x - Min) / (Max - Min)

[0047] Among them, x is the amplitude coefficient; Min is the minimum value among the amplitude coefficients, and Max is the maximum value among the amplitude coefficients;

[0048] In step S4, the feature extraction operation specifically includes the following steps:

[0049] d1: Find the highest point of the resampled amplitude curve coefficient after normalization to obtain the mean pressure mp;

[0050] d2: Transform the resampled amplitude curve coefficient X_Co to generate samples X_SYS for systolic blood pressure prediction and samples

[0051] X_DIA for diastolic blood pressure prediction; The specific transformation method is:

[0052] X_SYS[i] = X_Co[i] if i > mp else 0

[0053] X_DIA[i] = X_Co[i] if i < mp else 0

[0054] Among them, mp is the mean pressure and i is the pressure value;

[0055] In step S4, the dimensionality reduction operation specifically includes the following steps:

[0056] e1: Construct a predicted value dimensionality reduction model. The input of the predicted value dimensionality reduction model is 251-dimensional and the output is 16-dimensional;

[0057] The predicted value dimensionality reduction model is a deep learning model constructed based on the encoder-decoder architecture;

[0058] The predicted value dimensionality reduction model includes: an Encoder layer and a Decoder layer connected in sequence;

[0059] The Encoder layer includes: three consecutive multi-head self-attention layers and a fully connected layer connected in sequence. Each multi-head self-attention layer includes 8 self-attention modules. The fully connected layer reduces the data dimension to 16 dimensions, and the output intermediate data is defined as: XM;

[0060] The Decoder layer includes: a fully connected layer and three consecutive multi-head self-attention layers connected in sequence. Each multi-head self-attention layer includes 8 self-attention modules; The fully connected layer receives the XM output by the Encoder layer, and after raising the dimension of XM to 251 dimensions, it sends it to the subsequent three multi-head self-attention layers, and the output final result is denoted as: XR;

[0061] The loss function LossFuction is set as:

[0062] LossFuction = crossEnt(X输入 , XR 输入 ) + crossEnt(X 原始 , XR 原始 ) + crossEnt(XM 输入 , XM 原始 )

[0063] Among them, corssEnt() is the cross - entropy calculation; X 输入 represents the systolic blood pressure prediction X_SYS or diastolic blood pressure prediction X_DIA that needs to be dimension - reduced; XR 输入 represents the output value of the prediction value dimension - reduction model corresponding to X 输入 ; XM 输入 represents the intermediate data output by the Encoder layer in the prediction value dimension - reduction model corresponding to X 输入 ; X 原始 is the resampled amplitude curve coefficient X_Co; XR 原始 and XM 原始 mean that after inputting the resampled amplitude curve coefficient X_Co into the prediction value dimension - reduction model corresponding to X 输入 the final output value and the intermediate data output by the Encoder layer are obtained;

[0064] e2: Use the prediction value dimension - reduction model to perform dimension - reduction on the systolic blood pressure prediction sample X_SYS and the diastolic blood pressure prediction sample X_DIA respectively, to obtain the low - dimensional systolic blood pressure data and the low - dimensional diastolic blood pressure data;

[0065] During the process of generating the training samples, before performing the feature extraction operation, it is also necessary to perform data augmentation operations on the systolic blood pressure prediction sample X_SYS and the diastolic blood pressure prediction sample X_DIA;

[0066] The data augmentation operations include:

[0067] Based on the original training sample data, set the resampled amplitude curve coefficients on the right side of the mean pressure mp in the original training sample data to 0 and random numbers respectively, to obtain twice the sample data, and add them to the systolic blood pressure prediction sample X_SYS;

[0068] Based on the original training sample data, set the resampled amplitude curve coefficients on the left side of the mean pressure mp in the original training sample data to 0 and random numbers respectively, to obtain twice the sample data, and add them to the diastolic blood pressure prediction sample X_DIA;

[0069] In step S1, collect the original measurement data of blood pressure detection based on the blood pressure cuff device;

[0070] The systolic blood pressure prediction model and the diastolic blood pressure prediction model are constructed based on the random forest algorithm.

[0071] A blood pressure measurement device based on a blood pressure pulse wave, comprising: a controller, a display module, and a data acquisition module. The display module and the data acquisition module are respectively electrically connected to the controller, and are characterized in that:

[0072] The controller includes: a calculation module and a calculation model; the calculation model includes: a trained systolic blood pressure prediction model, a diastolic blood pressure prediction model, and a predicted value dimensionality reduction model;

[0073] The data acquisition module is electrically connected to an external blood pressure cuff device for blood pressure detection to collect measurement data; after the data acquisition module collects the original blood pressure data to be measured, it sends the data to the calculation module;

[0074] In the calculation module, the resampled amplitude curve coefficient corresponding to the original blood pressure data to be measured is calculated, denoted as: the amplitude curve coefficient to be calculated; after performing feature extraction operations on the amplitude curve coefficient to be calculated, the systolic blood pressure prediction sample and the diastolic blood pressure prediction sample extracted are input into the predicted value dimensionality reduction model for dimensionality reduction to obtain systolic blood pressure input data and diastolic blood pressure input data;

[0075] The calculation module inputs the systolic blood pressure input data and the diastolic blood pressure input data into the corresponding systolic blood pressure prediction model and diastolic blood pressure prediction model for blood pressure prediction;

[0076] The systolic blood pressure prediction model and the diastolic blood pressure prediction model send the output systolic blood pressure prediction value and diastolic blood pressure prediction value to the calculation module. The calculation module calculates the prediction value credibility corresponding to the systolic blood pressure prediction value and the diastolic blood pressure prediction value respectively based on the amplitude curve coefficient to be calculated, the systolic blood pressure prediction value, and the diastolic blood pressure prediction value;

[0077] The calculation module uses the systolic blood pressure prediction value, the diastolic blood pressure prediction value, and the prediction value credibility as the final output value and transmits it to the display module for display.

[0078] A blood pressure calculation method based on blood pressure pulse waves provided by the present application is based on the original pressure pulse wave data of blood pressure, fits an amplitude curve, then performs resampling and normalization processing on the amplitude curve, extracts all key feature information required for calculating blood pressure values, and does not need to collect other data for auxiliary calculation. The blood pressure values of diastolic blood pressure and systolic blood pressure are calculated based on the pressure pulse wave, which not only has higher accuracy but also is more targeted. A systolic blood pressure prediction model and a diastolic blood pressure prediction model are constructed based on machine learning algorithms, and the two models are trained with labeled training data. Through feature extraction, the normalized resampling amplitude coefficients collected based on the original pressure pulse wave are respectively transformed into input data suitable for the diastolic blood pressure prediction model and the systolic blood pressure prediction model, improving the accuracy of blood pressure detection while ensuring that the blood pressure calculation method of the present application is applicable to the calculation of different types of blood pressure data. At the same time, the present application first reduces the dimensionality of the feature values input into the systolic blood pressure prediction model and the diastolic blood pressure prediction model, and then inputs them into the systolic blood pressure prediction model and the diastolic blood pressure prediction model, ensuring that the calculation speed of this patent is faster without losing accuracy. Description of the Drawings

[0079] Figure 1 It is an example of a waveform diagram displayed by an oscillometric blood pressure monitor in the prior art;

[0080] Figure 2 It is a flowchart of the blood pressure calculation method based on blood pressure pulse waves in the present application;

[0081] Figure 3 It is an example of a pulse wave curve and a pressure curve corresponding to the original data of the training sample;

[0082] Figure 4 It is an example of a peak envelope line, a trough envelope line, and an amplitude curve;

[0083] Figure 5 It is an example of resampled amplitude curve coefficients;

[0084] Figure 6 It is an example of an amplitude curve of systolic blood pressure obtained after the feature extraction operation;

[0085] Figure 7 It is an example of an amplitude curve of diastolic blood pressure obtained after the feature extraction operation;

[0086] Figure 8 It is an example of original sample data with abnormal data;

[0087] Figure 9 For Figure 8 It is an example of sample data after removing outliers from the original sample data of;

[0088] Figure 10It is a trend chart of the confidence level of predicted values;

[0089] Figure 11 It is an embodiment of the original blood pressure data to be measured. Specific implementation manner

[0090] As Figure 2 shown, the present invention includes a blood pressure calculation method based on blood pressure pulse waves, which comprises the following steps.

[0091] S1: Obtain the original measurement data of blood pressure detection, label the diastolic blood pressure and systolic blood pressure, and obtain: the original training sample data;

[0092] The original training sample data includes: the original systolic blood pressure sample data and the original diastolic blood pressure sample data;

[0093] In this application, the original measurement data of blood pressure detection is collected based on a blood pressure cuff device; when using this method for blood pressure calculation, only the pressure pulse waves at the wrist, arm or elbow need to be collected to complete the blood pressure calculation, which is applicable to all oscillometric blood pressure monitors without the need for additional equipment. When actually constructing the original training sample data, preferably use the historical data with labels as the training sample data to train the systolic blood pressure prediction model and the diastolic blood pressure prediction model. However, when there is no suitable sample data with labels, it is necessary to use an auscultatory-assisted electronic blood pressure monitor for measurement, and during the measurement process, the doctor listens to the actual systolic blood pressure and diastolic blood pressure at the same time as the annotation result of the training data to form the original training sample data.

[0094] S2: Based on the original training sample data, extract the original pulse wave curve corresponding to the original measurement data, specifically as Figure 3 shown. Figure 3 In it, the abscissa is time and the ordinate is the pressure value of blood pressure.

[0095] After correcting the error data, find all the peak points and valley points in the original pulse wave curve, and fit to obtain the peak envelope line and the valley envelope line; calculate and fit the original amplitude curve based on the peak envelope line and the valley envelope line; as Figure 4 shown. Figure 4 In it, the abscissa is time and the ordinate is the blood pressure value.

[0096] Among them, the fitting process of the peak envelope line, the valley envelope line and the original amplitude curve specifically includes the following steps:

[0097] b1: When the blood pressure monitor starts to measure, the cuff will gradually deflate and the corresponding pressure value will gradually decrease. However, due to reasons such as sensor accuracy, the collected pressure curve will occasionally oscillate and increase. Therefore, it is necessary to correct the error data of the original pulse wave curve;

[0098] All pressure values in the original pulse wave curve that are greater than the pressure of the previous data are regarded as error data and deleted together with the corresponding pulse wave curve values to obtain the pulse wave curve for calculation;

[0099] b2: Extract the pulse wave curve for calculation, find all peak points, and use linear interpolation to fit to obtain the peak envelope curve;

[0100] b3: Extract the pulse wave curve for calculation, find all valley points, and use linear interpolation to fit to obtain the valley envelope curve;

[0101] b4: Place the peak envelope curve and the valley envelope curve in the same coordinate system, with the abscissa being time and the ordinate being the pressure value, and calculate the original amplitude curve (marked in the figure as: amplitude curve); the abscissa of the original amplitude curve is time, and the ordinate is the ordinate of the peak envelope curve minus the ordinate of the valley envelope curve;

[0102] In this application, the amplitude curve is used as the main data source for subsequent blood pressure calculation because the amplitude curve contains the main characteristic information of the blood pressure pulse wave, that is, the information of the peak envelope and the valley envelope, ensuring that this application only needs to use the pressure pulse wave data based on the original blood pressure as the most basic data and can obtain accurate blood pressure calculation results without using other auxiliary data. And through actual data verification, based on other curves, such as: the original peak envelope curve or the valley envelope curve, etc., the accuracy of the blood pressure value calculation results is lower than that of the amplitude curve.

[0103] S3: After removing the abnormal data in the original data of the training samples, resample the original amplitude curve based on the pressure value to complete the data augmentation operation on the original data of the training samples;

[0104] Then normalize the coefficients of the amplitude curve obtained after data augmentation to obtain the resampled amplitude curve coefficients, as specifically Figure 5 shown, where the horizontal axis is the pressure curve and the vertical axis is the normalization result.

[0105] In this method, an unsupervised learning algorithm Isolation Tree based on clustering is constructed to remove the abnormal data in the original data of the training samples. As Figure 8 shown in the original data of the training samples, after removing the abnormal data through the Isolation Tree model, as Figure 9 shown.

[0106] In step S3, the calculation method of the resampled amplitude curve coefficients includes the following steps:

[0107] c1: Starting from 0 mmHg at intervals of 1 mmHg, X maxThe original amplitude curve obtained by fitting is intercepted with mmHg as the end point, and the pressure values are resampled; based on the pressure values obtained after resampling the pressure values, a fitting curve is constructed again to obtain: the fitting curve for calculation;

[0108] where X max is the maximum range of human blood pressure. In this method, X max = 250.

[0109] c2: Calculate the curve amplitude coefficient of the fitting curve for calculation, denoted as: the amplitude coefficient for calculation;

[0110] c3: Normalize the amplitude coefficient for calculation. The normalization method is:

[0111] (x - Min) / (Max - Min)

[0112] where x is the amplitude coefficient; Min is the minimum value in the amplitude coefficient, and Max is the maximum value in the amplitude coefficient.

[0113] In this method, the amplitude curve is selected as the main data source for calculation and analysis. The pressure values of the amplitude curve are resampled, and the obtained resampling coefficients contain both the corresponding pressure information and pulse wave information. The resampling coefficients adopted carry complete characteristic information, ensuring the accuracy of the subsequent blood pressure calculation results.

[0114] S4: Perform feature extraction operations and dimensionality reduction operations on the resampled amplitude curve coefficients in sequence;

[0115] The obtained low - dimensional systolic blood pressure data is the systolic blood pressure sample data; the obtained low - dimensional diastolic blood pressure data is the diastolic blood pressure sample data.

[0116] In step S4, the feature extraction operation specifically includes the following steps:

[0117] d1: Find the highest point of the normalized resampled amplitude curve coefficients to obtain the mean pressure mp;

[0118] The systolic blood pressure is on the left side of the mean pressure and is only related to the left - hand part of the data. When the pressure in the cuff reaches the mean pressure, the systolic blood pressure has been measured, so the right - hand part of the data has no effect on the systolic blood pressure; similarly, the left - hand part of the data has no effect on the diastolic blood pressure.

[0119] d2: Transform the resampled amplitude curve coefficients X_Co to generate the sample X_SYS for systolic blood pressure prediction and the sample

[0120] X_DIA for diastolic blood pressure prediction; the specific transformation method is:

[0121] X_SYS[i] = X_Co[i] if i > mp else 0

[0122] X_DIA[i] = X_Co[i] if i < mp else 0

[0123] Among them, mp is the mean pressure, and i is the pressure value.

[0124] The systolic blood pressure curve obtained after the feature extraction operation is as Figure 6 shown, and the diastolic blood pressure is as Figure 7 shown. Figure 6 and Figure 7 In, the units of the horizontal axis and the vertical axis are the same as those in Figure 5 . The horizontal axis is the pressure curve, and the vertical axis is the normalization result. In this application, the information of the peaks, valleys, and pressure in the pressure pulse wave is extracted through the feature extraction operation, and the feature information that can most accurately express the pressure pulse wave is extracted, ensuring that the data used in subsequent calculations includes all the data information required for prediction, and ensuring that this method can obtain high-accuracy calculation results without collecting other additional auxiliary data.

[0125] The dimensions of X_DIA and X_SYS are 251-dimensional and contain a large number of 0s. Therefore, in this method, a deep learning model based on the encoder-decoder architecture is constructed to reduce the dimensions of X_DIA and X_SYS.

[0126] In step S4, the dimension reduction operation specifically includes the following steps:

[0127] e1: Construct a prediction value dimension reduction model;

[0128] In this application, the blood pressure value range is [0, 205]. Therefore, the input of the prediction value dimension reduction model is 251-dimensional, and the output is 16-dimensional;

[0129] The prediction value dimension reduction model is a deep learning model constructed based on the encoder-decoder architecture;

[0130] The prediction value dimension reduction model includes: an Encoder layer and a Decoder layer connected in sequence;

[0131] The Encoder layer includes: three consecutive multi-head self-attention layers and a fully connected layer connected in sequence. Each multi-head self-attention layer includes 8 self-attention modules. The fully connected layer reduces the data dimension to 16-dimensional, and the output intermediate data is defined as: XM;

[0132] The Decoder layer includes: a fully connected layer and three consecutive multi-head self-attention layers connected in sequence. Each multi-head self-attention layer includes 8 self-attention modules; the fully connected layer receives the XM output by the Encoder layer, raises the dimension of XM to 251-dimensional and then sends it to the subsequent three multi-head self-attention layers, and the output final result is denoted as: XR.

[0133] The Encoder layer contains 3 layers of multi-head self-attention layers, with 8 self-attention modules in each layer. The self-attention modules are used for feature extraction to extract the main information in the original data. Finally, a fully connected layer is used to reduce the data dimension to 16 dimensions, and the output intermediate data is defined as XM. The Decoder layer is the opposite of the Encoder layer. First, a fully connected layer is used to expand XM to 251 dimensions, and then 3 layers of multi-head self-attention layers are used. The self-attention layers in the Encoder layer are used for feature restoration to restore the main information features to the original data. Each multi-head self-attention layer has 8 self-attention modules, and the output result is XR.

[0134] The loss function LossFuction of the predicted value dimensionality reduction model is set as:

[0135] LossFuction = crossEnt(X 输入 , XR 输入 ) + crossEnt(X 原始 , XR 原始 ) + crossEnt(XM 输入 , XM 原始 )

[0136] Among them, crossEnt() is the cross-entropy calculation to ensure the consistency of input and output. X 输入 represents the systolic blood pressure prediction X_SYS or diastolic blood pressure prediction X_DIA that needs to be dimensionally reduced; XR 输入 represents the output value of the predicted value dimensionality reduction model corresponding to X 输入 ; XM 输入 represents the intermediate data output by the Encoder layer in the predicted value dimensionality reduction model corresponding to X 输入 ; X 原始 is the resampled amplitude curve coefficient X_Co; XR 原始 and XM 原始 mean that after inputting the resampled amplitude curve coefficient X_Co into the predicted value dimensionality reduction model corresponding to X 输入 , the final output value and the intermediate data output by the Encoder layer are obtained.

[0137] The loss function LossFuction of the predicted value dimensionality reduction model ensures the consistency of input and output through crossEnt(). crossEnt(X 输入 , XR 输入 ) ensures that it can be inversely deduced back to X 输入 from XM 输入 , and crossEnt(X 原始 , XR 原始 ) ensures that it can be inversely deduced back to X 原始 from XM 原始That is, ensure that the intermediate result contains the main information of the original data, so the intermediate result can be used as the data after dimensionality reduction for subsequent calculations.

[0138] By crossEnt(XM 输入 ,XM 原始 ), the consistency between the original data X 原始 and the intermediate result generated from the data X 输入 after feature extraction is ensured. This is because the information that X 原始 and X 输入 have during diastolic or systolic blood pressure calculation is the same, and the results will be similar, so they should remain consistent after dimensionality reduction. The final result XR 输入 is the input value for the subsequent random forest, with a dimension of 16. In this application, the 251-dimensional data is reduced to 16 dimensions, reducing the computational amount without changing the accuracy.

[0139] e2: Use the prediction value dimensionality reduction model to perform dimensionality reduction on the systolic blood pressure prediction sample X_SYS and the diastolic blood pressure prediction sample X_DIA respectively, to obtain low-dimensional systolic blood pressure data and low-dimensional diastolic blood pressure data.

[0140] If feature extraction is performed on the sample data during the process of generating training samples, data augmentation operations also need to be performed on the systolic blood pressure prediction sample X_SYS and the diastolic blood pressure prediction sample X_DIA before performing the feature extraction operation, to ensure that the training sample data has sufficient quantity and data coverage.

[0141] The data augmentation operations include:

[0142] Based on the original training sample data, set the resampling amplitude curve coefficients on the right side of the mean pressure mp in the original training sample data to 0 and random numbers respectively, to obtain twice the sample data, and add them to the systolic blood pressure prediction sample X_SYS; then the data volume of the systolic blood pressure prediction sample X_SYS includes: the systolic blood pressure data corresponding to the original training sample data, the systolic blood pressure data corresponding to setting the right side to 0, and the systolic blood pressure data corresponding to setting the right side to random numbers, that is, the data augmentation of the systolic blood pressure prediction sample X_SYS is realized.

[0143] Similarly, based on the original training sample data, set the resampling amplitude curve coefficients on the left side of the mean pressure mp in the original training sample data to 0 and random numbers respectively, to obtain twice the sample data, and add them to the diastolic blood pressure prediction sample X_DIA, to complete the data augmentation of the diastolic blood pressure prediction sample X_DIA.

[0144] The left-side mean pressure data contains complete systolic blood pressure feature information, and the right-side mean pressure data contains complete systolic blood pressure feature information. The accuracy of the calculated diastolic and systolic blood pressure results of the original data and the data after cropping, filling with random numbers, and filling with 0s is approximately the same. Therefore, in this method, by cropping the data, filling with random numbers, and filling with 0s, the purpose of data augmentation is achieved, and the training sample data is tripled, further improving the accuracy of model prediction.

[0145] S5: Generate a systolic blood pressure prediction model and a diastolic blood pressure prediction model respectively based on machine learning algorithms.

[0146] Common self-supervised machine learning algorithms, such as: boosting tree models, convolutional neural networks, or random forests can all be used to construct systolic blood pressure prediction models and diastolic blood pressure prediction models.

[0147] The systolic blood pressure prediction model and the diastolic blood pressure prediction model in this embodiment are constructed based on the random forest algorithm. The input parameters are the normalized resampled amplitude curve coefficients after feature extraction and dimensionality reduction, and the output parameters are diastolic blood pressure and systolic blood pressure respectively.

[0148] S6: Use the systolic blood pressure sample data to train the systolic blood pressure prediction model, and use the diastolic blood pressure sample data to train the diastolic blood pressure prediction model, respectively obtaining the trained systolic blood pressure prediction model and diastolic blood pressure prediction model. Through experiments, it is known that the result accuracy is the highest when the random forest parameter n_estimators is set to 100.

[0149] S7: Obtain the original blood pressure data to be measured, and extract the corresponding original pulse wave curve of the original blood pressure data to be measured; find all peak points and trough points in the original pulse wave curve, and fit to obtain the peak envelope line and the trough envelope line; calculate and fit the original amplitude curve based on the peak envelope line and the trough envelope line;

[0150] Perform pressure value resampling on the original amplitude curve, and then normalize the coefficients of the amplitude curve obtained after resampling to obtain the resampled amplitude curve coefficients;

[0151] Obtain the resampled amplitude curve coefficients corresponding to the original blood pressure data to be measured, denoted as: the amplitude curve coefficients to be calculated.

[0152] S8: Perform feature extraction operations and dimensionality reduction operations on the amplitude curve coefficients to be calculated in sequence, and obtain low-dimensional systolic blood pressure data and low-dimensional diastolic blood pressure data, denoted as: systolic blood pressure input data and diastolic blood pressure input data respectively.

[0153] S9: Input the systolic blood pressure input data into the systolic blood pressure prediction model, and input the diastolic blood pressure input data into the systolic blood pressure prediction model and the diastolic blood pressure prediction model for blood pressure prediction; the systolic blood pressure prediction model and the diastolic blood pressure prediction model respectively output the systolic blood pressure prediction value and the diastolic blood pressure prediction value, that is, the measured values corresponding to the original blood pressure data to be measured are obtained.

[0154] S10: Based on the amplitude curve coefficient to be calculated, the systolic blood pressure prediction value and the diastolic blood pressure prediction value, calculate the prediction value credibility corresponding to the systolic blood pressure prediction value and the diastolic blood pressure prediction value respectively. The specific calculation process is as follows.

[0155] a1: Calculate the prediction value accuracy X:

[0156]

[0157] where, DIA ai is the diastolic blood pressure prediction value, SYS ai is the systolic blood pressure prediction value, DIA amplitude is the diastolic blood pressure calculated based on the amplitude method, SYS amplitude is the systolic blood pressure calculated based on the amplitude method; abs() is used to calculate the absolute value of the parameter.

[0158] Based on the most commonly used amplitude method, evaluate the calculation results of this method. Since the systolic blood pressure must be greater than the diastolic blood pressure, the difference between the diastolic blood pressures of the two algorithms of the amplitude method and this method should be weighted and added to the difference in systolic blood pressure. Therefore, the differences in systolic and diastolic blood pressures calculated by the two methods are weighted to solve the problem of uneven numerical distribution. Since the calculation results of this method are compared with those of the amplitude method, a reference value of the algorithm credibility is given. If the model results are similar to or consistent with the calculation results of the amplitude method, it can be determined that the accuracy of the model calculation results is high; but vice versa, it is not necessarily the case, because there is also a great possibility that the low accuracy of the amplitude method leads to inconsistency with the calculation results of this method.

[0159] Therefore, this patent constructs the function accuracy for calculating credibility.

[0160] a2: Calculate the prediction value credibility:

[0161]

[0162] where, accuracy is the prediction value credibility; X is the accuracy of the prediction value.

[0163] The prediction value accuracy X is the distance between the calculation results of this method and the amplitude method. Through the function accuracy, X is mapped to the interval [0, 1], indicating the credibility of the calculation results of this method. For example Figure 10As shown in the figure, the abscissa is the calculation result of the function accuracy, and the ordinate is the value of X. When the value of X is small and less than 5, the return result is stable and close to 1, that is, the model result and the amplitude result are similar or consistent, and it is determined that the accuracy of the model calculation result is high. When the value of X increases and is greater than 5, accuracy drops rapidly. That is, in practical applications, the situation where accuracy is too small can be ignored.

[0164] S11: Use the systolic blood pressure prediction value, diastolic blood pressure prediction value, and prediction value credibility as the final output values together to ensure that users can clearly judge the reliability of the diastolic and systolic blood pressure results calculated based on this method, thereby ensuring the greater practicality of this method.

[0165] A blood pressure measurement device implemented based on the above blood pressure calculation method for blood pressure pulse waves includes: a controller, a display module, and a data acquisition module. The display module and the data acquisition module are respectively electrically connected to the controller. The controller includes: a calculation module and a calculation model; the calculation model includes: a trained systolic blood pressure prediction model, a diastolic blood pressure prediction model, and a prediction value dimensionality reduction model. In actual application, the controller is implemented based on a single-chip microcomputer with calculation functions, and the data acquisition module is implemented based on the module for collecting data in existing blood pressure measurement devices.

[0166] The data acquisition module is electrically connected to an external blood pressure cuff device for blood pressure detection to collect measurement data; after the data acquisition module collects the original blood pressure data to be measured, it sends it to the calculation module; the calculation module calculates the resampled amplitude curve coefficient corresponding to the original blood pressure data to be measured, denoted as: the amplitude curve coefficient to be calculated; after performing feature extraction operations on the amplitude curve coefficient to be calculated, the extracted systolic blood pressure prediction samples and diastolic blood pressure prediction samples are input into the prediction value dimensionality reduction model for dimensionality reduction to obtain systolic blood pressure input data and diastolic blood pressure input data;

[0167] The calculation module inputs the systolic blood pressure input data and the diastolic blood pressure input data into the corresponding systolic blood pressure prediction model and diastolic blood pressure prediction model for blood pressure prediction;

[0168] The systolic blood pressure prediction model and the diastolic blood pressure prediction model send the output systolic blood pressure prediction value and diastolic blood pressure prediction value to the calculation module at the same time. The calculation module calculates the prediction value credibility corresponding to the systolic blood pressure prediction value and the diastolic blood pressure prediction value respectively based on the amplitude curve coefficient to be calculated, the systolic blood pressure prediction value, and the diastolic blood pressure prediction value;

[0169] The calculation module uses the systolic blood pressure prediction value, the diastolic blood pressure prediction value, and the prediction value credibility as the final output values and transmits them to the display module for display.

[0170] Next, take Figure 11 the original blood pressure data to be measured as an example to illustrate the calculation process of using the trained prediction model. ForFigure 11 After the feature extraction of the shown systolic and diastolic pressure curves, dimensionality reduction and normalization are performed. The highest point of the resampled amplitude curve coefficient after normalization, that is, the mean pressure is 70.

[0171] Normalized amplitude coefficient for diastolic pressure calculation after dimensionality reduction: [0.0810434 0.07978018 0.11057794 0.04914938 0.13818289 0.06857166 0.00107338 0.00782576 0.13650589 0.06877731 0.12506482 0.04323351 0.00172375 0.06200045 0.01784967 0.00864001]

[0175] Normalized amplitude coefficient for systolic pressure calculation after dimensionality reduction: [0.17834567 0.0743443 0.16016477 0.00881322 0.06127112 0.02101125 0.10085562 0.09179793 0.00225599 0.02353307 0.00511314 0.02181476 0.01593208 0.11093742 0.07417119 0.04963847]

[0179] The normalized amplitude coefficient for diastolic pressure calculation after dimensionality reduction is fed into the trained diastolic pressure prediction model, and the normalized amplitude coefficient for systolic pressure calculation after dimensionality reduction is fed into the trained systolic pressure prediction model:

[0180] The predicted diastolic pressure value DIA output by the diastolic pressure prediction model ai is: 60.66,

[0181] The predicted systolic pressure value SYS output by the systolic pressure prediction model ai is: 100.085.

[0182] Using the same raw blood pressure data to be measured, it is calculated by the amplitude method. In this embodiment, the mean pressure is 70. When calculating the diastolic and systolic pressures of the raw blood pressure data to be measured based on the amplitude method, the pressure corresponding to 0.45 times the height on the left side of the mean pressure is selected as the systolic pressure, and the pressure corresponding to 0.68 times the height on the right side is selected as the diastolic pressure. The calculated result DIA of the diastolic pressure by the amplitude method amplitude is: 60, and the calculated result of the systolic pressure by the amplitude method is SYS amplitude: 97.

[0183] Substitute DIA ai , SYS ai , DIA amplitude and SYS amplitude into the calculation formula of the prediction value accuracy X:

[0184]

[0185] The calculated result obtained is: X = 1.5808657886040647.

[0186] Substitute the value of X into the formula and the result obtained is: 1.0.

[0187] Then the final output item of this blood pressure calculation is:

[0188] The predicted value of diastolic blood pressure is: 60.66; the predicted value of systolic blood pressure is: 100.085; the confidence level of the predicted value is 1.0.

[0189] After using the technical solution of the present invention, a systolic blood pressure prediction model and a diastolic blood pressure prediction model are constructed by using a machine learning algorithm. The blood pressure pulse wave is obtained by calculating the original measurement data, the feature with the highest accuracy in the blood pressure pulse wave is extracted, and the pressure is resampled. The generated resampling amplitude coefficient contains all the data information used for prediction, and accurate prediction can be achieved without additional data. After extracting and dimension-reducing the key feature information in the blood pressure pulse wave, it is used as the input feature for calculating the blood pressure value, ensuring that the calculation speed is faster without loss of accuracy. Moreover, based on the construction of a systolic blood pressure prediction model and a diastolic blood pressure prediction model by using a machine learning algorithm, after training with the training samples after data augmentation, it can cover all user groups and is more practical. Through experiments, using the same original data, comparing the calculation results of this method with those of the amplitude method, the calculation results of this method are more accurate. Among them, the accuracy of systolic blood pressure is 12% higher, and the accuracy of diastolic blood pressure is 8.6% higher.

Claims

1. A blood pressure calculation method based on blood pressure pulse waves, characterized in that, It includes the following steps: S1: Obtain the original measurement data of blood pressure detection, label the diastolic blood pressure and systolic blood pressure to obtain: the original training sample data; The original training sample data includes: the original systolic blood pressure sample data and the original diastolic blood pressure sample data; S2: Based on the original training sample data, extract the original pulse wave curve corresponding to the original measurement data; after correcting the error data, find all the peak points and valley points in the original pulse wave curve, and fit to obtain the peak envelope line and the valley envelope line; Calculate and fit the original amplitude curve based on the peak envelope line and the valley envelope line; S3: After removing the abnormal data in the original training sample data, resample the original amplitude curve based on the pressure value to complete the data augmentation operation on the original training sample data; Then normalize the coefficients of the amplitude curve obtained after data augmentation to obtain the resampled amplitude curve coefficients; S4: Perform feature extraction operations and dimensionality reduction operations on the resampled amplitude curve coefficients in sequence; The obtained low-dimensional systolic blood pressure data is the systolic blood pressure sample data; the obtained low-dimensional diastolic blood pressure data is the diastolic blood pressure sample data; S5: Generate a systolic blood pressure prediction model and a diastolic blood pressure prediction model respectively based on the machine learning algorithm; S6: Use the systolic blood pressure sample data to train the systolic blood pressure prediction model, and use the diastolic blood pressure sample data to train the diastolic blood pressure prediction model to obtain the trained systolic blood pressure prediction model and the diastolic blood pressure prediction model respectively; S7: Obtain the original blood pressure data to be measured, extract the original pulse wave curve corresponding to the original blood pressure data to be measured; find all the peak points and valley points in the original pulse wave curve, and fit to obtain the peak envelope line and the valley envelope line; calculate and fit the original amplitude curve based on the peak envelope line and the valley envelope line; Resample the original amplitude curve based on the pressure value, and then normalize the coefficients of the amplitude curve obtained after resampling to obtain the resampled amplitude curve coefficients; Obtain the resampled amplitude curve coefficients corresponding to the original blood pressure data to be measured, denoted as: the amplitude curve coefficients to be calculated; S8: Perform the feature extraction operation and the dimensionality reduction operation on the amplitude curve coefficients to be calculated in sequence to obtain the low-dimensional systolic blood pressure data and the low-dimensional diastolic blood pressure data, denoted as: the systolic blood pressure input data and the diastolic blood pressure input data respectively; S9: Input the systolic blood pressure input data into the systolic blood pressure prediction model, and input the diastolic blood pressure input data into the systolic blood pressure prediction model and the diastolic blood pressure prediction model for blood pressure prediction; the systolic blood pressure prediction model and the diastolic blood pressure prediction model respectively output the systolic blood pressure prediction value and the diastolic blood pressure prediction value, that is, obtain the measurement value corresponding to the original blood pressure data to be measured.

2. The blood pressure calculation method based on blood pressure pulse wave according to claim 1, wherein: It further includes the following steps: S10: Based on the amplitude curve coefficients to be calculated, the systolic blood pressure prediction value and the diastolic blood pressure prediction value, calculate the prediction value credibility corresponding to the systolic blood pressure prediction value and the diastolic blood pressure prediction value respectively; the calculation process includes: a1: Calculate the prediction value accuracy X: Among them, DIA ai is the predicted value of diastolic blood pressure, SYS ai is the predicted value of systolic blood pressure, DIA amplitude is the diastolic blood pressure calculated based on the amplitude method, SYS amplitude is the systolic blood pressure calculated based on the amplitude method; abs() is used to calculate the absolute value of the parameter; a2: Calculate the prediction value credibility: Among them, accuracy is the confidence level of the predicted value; X is the accuracy of the predicted value; S11: Use the systolic blood pressure predicted value, the diastolic blood pressure predicted value, and the confidence level of the predicted value as the final output value together.

3. The blood pressure calculation method based on blood pressure pulse wave according to claim 1, characterized in that: In step S2, the fitting of the peak envelope line, the trough envelope line, and the original amplitude curve includes the following steps: b1: Perform an operation to correct error data on the original pulse wave curve; The operation of correcting error data is: Consider all pressure values in the original pulse wave curve that are greater than the pressure of the previous data as error data and delete them together with the corresponding pulse wave curve values to obtain a pulse wave curve for calculation; b2: Extract the pulse wave curve for calculation, find all peak points, and use linear interpolation to fit to obtain the peak envelope line; b3: Extract the pulse wave curve for calculation, find all trough points, and use linear interpolation to fit to obtain the trough envelope line; b4: Place the peak envelope line and the trough envelope line in the same coordinate system, with the abscissa being time and the ordinate being the pressure value, and calculate the original amplitude curve; the abscissa of the original amplitude curve is time, and the ordinate is the ordinate of the peak envelope line minus the ordinate of the trough envelope line.

4. The blood pressure calculation method based on blood pressure pulse wave according to claim 1, wherein: In step S3, the calculation method of the resampled amplitude curve coefficient includes the following steps: c1: Starting from 0 mmHg and at intervals of 1 mmHg, the original amplitude curve obtained by fitting is intercepted up to X max mmHg as the end point, and the pressure values are resampled; Based on the pressure values obtained after the resampling of the pressure values, a fitting curve is constructed again to obtain: a fitting curve for calculation; where X max is the maximum range of human blood pressure; c2: Calculate the curve amplitude coefficient of the calculated fitting curve, denoted as: calculated amplitude coefficient; c3: Normalize the calculated amplitude coefficient, and the normalization method is: (x - Min) / (Max - Min) where x is the amplitude coefficient; Min is the minimum value in the amplitude coefficient, and Max is the maximum value in the amplitude coefficient.

5. The blood pressure calculation method based on blood pressure pulse wave according to claim 1, wherein: In step S4, the feature extraction operation specifically includes the following steps: d1: Find the highest point of the normalized resampled amplitude curve coefficient to obtain the mean pressure mp; d2: Transform the resampled amplitude curve coefficient X_Co to generate a systolic blood pressure prediction sample X_SYS and a diastolic blood pressure prediction sample X_DIA respectively; the specific transformation method is: X_SYS[i] = X_Co[i] if i > mp else 0 X_DIA[i] = X_Co[i] if i < mp else 0 where mp is the mean pressure and i is the pressure value.

6. The blood pressure calculation method based on blood pressure pulse wave according to claim 5, characterized in that: In step S4, the dimensionality reduction operation specifically includes the following steps: e1: Construct a predicted value dimensionality reduction model, the input of the predicted value dimensionality reduction model is 251 - dimensional, and the output is 16 - dimensional; The predicted value dimensionality reduction model is a deep learning model constructed based on the encoder - decoder architecture; The predicted value dimensionality reduction model includes: an Encoder layer and a Decoder layer connected in sequence; The Encoder layer includes: three consecutive multi - head self - attention layers and a fully - connected layer connected in sequence. Each multi - head self - attention layer includes 8 self - attention modules. The fully - connected layer reduces the data to 16 - dimensional, and the output intermediate data is defined as: XM; The Decoder layer includes: a fully connected layer and three consecutive multi-head self-attention layers connected in sequence. Each multi-head self-attention layer includes 8 self-attention modules. The fully connected layer receives XM output by the Encoder layer, expands the dimension of XM to 251 dimensions, and then sends it to the subsequent three multi-head self-attention layers. The final output result is denoted as XR. The loss function LossFuction is set as: LossFuction = crossEnt(X 输入 ,XR 输入 ) + crossEnt(X 原始 ,XR 原始 ) + crossEnt(XM 输入 ,XM 原始 ) Among them, crossEnt() is the cross-entropy calculation; X 输入 represents the systolic blood pressure prediction X_SYS or diastolic blood pressure prediction X_DIA that needs to be dimensionally reduced; XR 输入 represents X 输入 the output value of the corresponding predicted value dimensionality reduction model; XM 输入 represents X 输入 the intermediate data output by the Encoder layer in the corresponding predicted value dimensionality reduction model; X 原始 is the resampled amplitude curve coefficient X_Co; XR 原始 and XM 原始 means that after inputting the resampled amplitude curve coefficient X_Co into the 输入 corresponding predicted value dimensionality reduction model, the final output value and the intermediate data output by the Encoder layer are obtained; e2: Use the predicted value dimensionality reduction model to perform dimensionality reduction on the systolic blood pressure prediction sample X_SYS and the diastolic blood pressure prediction sample X_DIA respectively, to obtain the low-dimensional systolic blood pressure data and the low-dimensional diastolic blood pressure data.

7. The blood pressure calculation method based on blood pressure pulse wave according to claim 1, characterized in that: During the process of generating training samples, before performing the feature extraction operation, it is also necessary to perform data augmentation operations on the systolic blood pressure prediction sample X_SYS and the diastolic blood pressure prediction sample X_DIA. The data augmentation operation includes: Based on the original training sample data, set the resampled amplitude curve coefficients on the right side of the mean pressure mp in the original training sample data to 0 and random numbers respectively, to obtain twice the sample data, and add it to the systolic blood pressure prediction sample X_SYS. Based on the original training sample data, set the resampled amplitude curve coefficients on the left side of the mean pressure mp in the original training sample data to 0 and random numbers respectively, to obtain twice the sample data, and add it to the diastolic blood pressure prediction sample X_DIA.

8. The blood pressure calculation method based on blood pressure pulse wave according to claim 1, characterized in that: In step S1, the original measurement data of blood pressure detection is collected based on the blood pressure cuff device.

9. The blood pressure calculation method based on blood pressure pulse wave according to claim 1, characterized in that: The systolic blood pressure prediction model and the diastolic blood pressure prediction model are constructed based on the random forest algorithm.

10. A blood pressure measuring device implemented by the blood pressure calculation method of the blood pressure pulse wave according to any one of claims 1-9, comprising: A controller, a display module, and a data acquisition module. The display module and the data acquisition module are electrically connected to the controller respectively. It is characterized in that: The controller includes: a calculation module and a calculation model. The calculation model includes: a trained systolic blood pressure prediction model, a diastolic blood pressure prediction model, and a predicted value dimensionality reduction model. The data acquisition module is electrically connected to an external blood pressure cuff device for blood pressure detection to collect measurement data. After the data acquisition module collects the original blood pressure data to be measured, it sends it to the calculation module. In the calculation module, calculate the resampled amplitude curve coefficient corresponding to the original blood pressure data to be measured, denoted as: the amplitude curve coefficient to be calculated. After performing feature extraction operations on the amplitude curve coefficient to be calculated, input the extracted systolic blood pressure prediction sample and diastolic blood pressure prediction sample into the predicted value dimensionality reduction model for dimensionality reduction, to obtain systolic blood pressure input data and diastolic blood pressure input data. The calculation module inputs the systolic blood pressure input data and the diastolic blood pressure input data into the corresponding systolic blood pressure prediction model and diastolic blood pressure prediction model respectively for blood pressure prediction. The systolic blood pressure prediction model and the diastolic blood pressure prediction model send the output systolic blood pressure prediction value and diastolic blood pressure prediction value into the calculation module. The calculation module calculates the prediction value credibility corresponding to the systolic blood pressure prediction value and the diastolic blood pressure prediction value respectively based on the amplitude curve coefficient to be calculated, the systolic blood pressure prediction value, and the diastolic blood pressure prediction value. The computing module takes the systolic blood pressure prediction value, the diastolic blood pressure prediction value, and the prediction value credibility as the final output value and transmits it to the display module for display.

Images