WO2020135725A1

WO2020135725A1 - Apparatus and electronic device for calculating blood pressure

Info

Publication number: WO2020135725A1
Application number: PCT/CN2019/129235
Authority: WO
Inventors: 李霞; 李梦亭; 方真; 周秦武; 支周; 卢忱
Original assignee: 中兴通讯股份有限公司; 西安交通大学
Priority date: 2018-12-29
Filing date: 2019-12-27
Publication date: 2020-07-02
Also published as: CN111374652A

Abstract

An apparatus and electronic device for calculating blood pressure, the apparatus (100) comprising: an extraction module (110), which is used for extracting feature parameters from a pulse wave, the feature parameters comprising time domain feature parameters, wavelet domain feature parameters, Fourier transformation domain feature parameters and Hilbert transformation domain feature parameters; a generation module (120), which is connected to the extraction module (110) and which is used to adjust the weights of training data and a weak predictor according to an error of training data in the feature parameters to form a strong predictor; and an output module (130), which is connected to the generation module (120) and which is used for inputting feature parameters into the strong predictor so as to output blood pressure.

Description

Device and electronic equipment for calculating blood pressure

cross reference

The present invention requires the priority of a Chinese patent application filed on December 29, 2018 in the Chinese Patent Office with the application number 201811642572.4 and the invention titled "A device and electronic device for calculating blood pressure". The entire content of the application is cited by reference Incorporated in the invention.

Technical field

This application relates to the technical field of medical devices, in particular to a device and electronic equipment for calculating blood pressure.

Background technique

Common blood pressure measurement methods are divided into direct method and indirect method. The direct method of puncturing transfers the pressure in the artery to the external pressure sensor through the liquid in the catheter to measure the blood pressure, but the operation is complicated, invasive, and easy to cause infection.

Indirect methods include: Korotkoff sound method, oscillometric method, constant volume method, etc. The Korotkoff sound method uses the overflow sound during the blood flow obstruction and the corresponding pressure points to determine the systolic and diastolic pressures; the oscillometric method detects the oscillating waves originating from the blood vessel wall during the vascular obstruction, according to the envelope of the oscillating waves and The relationship between the pressures determines the systolic and diastolic pressures; the constant volume method regulates the applied pressure through the servo pressure control system to keep the arterial volume constant, and measures the applied pressure to obtain continuous arterial blood pressure. Ultrasound uses the Doppler principle to detect the Doppler frequency shift caused by the relative motion of blood flow and the vessel wall. The change in Doppler frequency caused by the applied pressure determines the systolic and diastolic pressure. These blood pressure detection methods need to continuously inflate and deflate using cuffs, or require a more complicated control system to control the amount of applied pressure, or require a more complicated device to simultaneously measure multiple physiological signals, and cannot achieve continuous long-term measurement of blood pressure .

The information disclosed in this Background section is only intended to increase the understanding of the overall background of the invention, and should not be taken as an acknowledgement or in any way suggesting that this information constitutes prior art that is well known to those of ordinary skill in the art.

Summary of the invention

In a first aspect, an embodiment of the present application provides an apparatus for calculating blood pressure, an extraction module for extracting characteristic parameters from a pulse wave, the characteristic parameters including time-domain characteristic parameters, wavelet-domain characteristic parameters, and Fourier transform domain Feature parameters and Hilbert transform domain feature parameters; a generation module, connected to the extraction module, used to adjust the weights of the training data and the weak predictor to form strong based on the error of the training data in the feature parameters Predictor; output module, connected to the generation module, for inputting the characteristic parameter into the strong predictor to output blood pressure.

In a second aspect, an embodiment of the present application provides an electronic device, including: a memory, a processor, and computer-executable instructions stored on the memory and executable on the processor, the computer-executable instructions being When the processor is executed, the implementation step is: extracting feature parameters from the pulse wave, the feature parameters including time domain feature parameters, wavelet domain feature parameters, Fourier transform domain feature parameters and Hilbert transform domain feature parameters; The error of the training data in the feature parameter adjusts the weight of the training data and the weak predictor to form a strong predictor; the feature parameter is input into the strong predictor to output blood pressure.

In a third aspect, an embodiment of the present application provides a computer-readable storage medium for storing computer-executable instructions. When the computer-executable instructions are executed by a processor, a step is implemented: from the pulse wave Feature parameters are extracted from the feature parameters, including time domain feature parameters, wavelet domain feature parameters, Fourier transform domain feature parameters and Hilbert transform domain feature parameters; according to the error of the training data in the feature parameters, the The training data and the weight of the weak predictor are adjusted to form a strong predictor; the feature parameters are input to the strong predictor to output blood pressure.

According to a fourth aspect, an embodiment of the present application provides a computer program product, the computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program includes program instructions, and when the program instructions When executed by a computer, the computer is caused to perform the methods described in the above aspects.

BRIEF DESCRIPTION

In order to more clearly explain the embodiments of the present application or the technical solutions in the prior art, the following will briefly introduce the drawings used in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only These are some of the embodiments described in this application. For those of ordinary skill in the art, without paying any creative labor, other drawings can also be obtained based on these drawings.

FIG. 1 shows a schematic structural diagram of an apparatus for calculating blood pressure provided by an embodiment of the present application;

2 is a schematic structural diagram of a blood pressure calculation device according to another embodiment of the present application;

Figure 3 shows a schematic diagram of the network structure of a restricted Boltzmann machine;

4 is a schematic structural diagram of a blood pressure calculation device according to another embodiment of the present application;

FIG. 5 shows a schematic structural diagram of a blood pressure calculation device according to another embodiment of the present application;

FIG. 6 is a schematic diagram of a hardware structure of an electronic device provided by an embodiment of the present application.

detailed description

In order to enable those skilled in the art to better understand the technical solutions in the present application, the technical solutions in the embodiments of the present application will be described clearly and completely in conjunction with the drawings in the embodiments of the present application. Obviously, the described The embodiments are only a part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the scope of protection of this application.

Non-invasive continuous blood pressure testing has developed rapidly, especially continuous blood pressure testing based on Pulse Wave Transit Time (PTT) has made great progress. Pulse wave transmission time PTT is the time required for the pulse wave to pass from the proximal artery to the distal artery. Studies have shown that PTT is inversely related to blood pressure. In the traditional method of measuring PTT, it is necessary to use ECG leads to obtain ECG, which increases the difficulty and cost of measurement. Based on this, the first PTT measurement device proposed has placed two PPG sensors on the skin of the brachial artery and the middle finger artery, of which the reflective photoelectric sensor is used on the brachial artery and the transmissive sensor is used on the finger artery. The correlation between the PTT obtained by using the interval between the characteristic points of the pulse wave of the brachial artery and the finger artery and the PTT obtained by the combined technique of ECG and PPG reached 0.7. In addition, the correlation between PTTp (the time interval between RPG and PPG peaks in the same cardiac cycle) and PTTf (the time interval between RPG and PPG troughs) and systolic and diastolic blood pressure were found. The correlation between PTTf and SBP and PBP (the difference between systolic and diastolic blood pressure) is better than that of PTTp. The second device has been proposed to use variables such as pulse wave conduction time, systolic wave height, stroke volume, K value, etc., to filter variables through stepwise regression analysis and establish regression between systolic pressure and average pressure and these variables In the equation, the error of the blood pressure calculation of 75% of the cardiac cycle of each participant is below 5%. The third proposed device uses the pulse wave of the ulnar artery and radial artery to obtain the pulse wave transmission speed, and uses the hydrostatic pressure method to register the physical model P=k1ln(c2)+k2. The fourth proposed device uses the heart sound signal as a reference point to calculate PTT, and uses pulse wave parameters such as pulse transit time, stroke output, waveform coefficient, average ascending slope, and pulse rate to perform multiple linear regression analysis. The average error of systolic blood pressure and diastolic blood pressure is 1.62mmHg and 1.12mmHg, respectively, which has high measurement accuracy. The fifth proposed device proposes a characteristic parameter k, where k is the reciprocal of the first peak of the first-order differential of PPG and represents the transmission time per unit amplitude. The sixth proposed device has proposed a new indicator photoplethysmogram intensity ratio (PIR) that can track the low-frequency component of blood pressure changes, and realized the blood pressure estimation using PIR and PTT, and finally the systolic, diastolic and The average value and standard deviation of the average pressure estimation error are -0.37±5.21, -0.08±4.06, and -0.18±4.13 mmHg, respectively. The seventh device has been proposed to extract the characteristics of ECG and pulse wave signals, including heart rate, arterial stiffness index and other physiological parameters and the time and morphological characteristics of the signal, use principal component analysis to reduce the dimension of the characteristic parameters, and use linear Regression, decision tree regression and Adaboost's method are used to establish the blood pressure model. Finally, the comparison found that the average absolute error of the Adaboost method in calculating the blood pressure result is the smallest. The above method uses multiple physiological signals such as pulse wave and electrocardiogram to obtain blood pressure-related features. Multiple physiological signals need to be measured, the measurement device is complicated, and the measurement accuracy needs to be further improved.

Based on this, embodiments of the present application provide a device for calculating blood pressure, which can be used in various portable blood pressure measuring devices and instruments, and can also be combined with wearable devices such as smart bracelets.

FIG. 1 shows a schematic structural diagram of an apparatus for calculating blood pressure provided by an embodiment of the present application. The apparatus 100 includes an extraction module 110, a generation module 120, and an output module 130.

The extraction module 110 is configured to extract feature parameters from the pulse wave, and the feature parameters include time domain feature parameters, wavelet domain feature parameters, Fourier transform domain feature parameters, and Hilbert transform domain feature parameters.

Regarding pulse wave feature parameter extraction technology, it specifically includes: Time domain feature parameters are mainly divided into four categories, including amplitude features, area features, energy features, and physiological parameter features, and some time domain feature parameters are shown in Table 1. The time domain mainly extracts the amplitude characteristic, area characteristic, energy characteristic and physiological parameter characteristic from the pulse wave signal and its first-order differential signal. The characteristics of Fourier transform domain, wavelet domain and Hilbert transform domain are the features extracted from each transformed signal by performing Fourier transform, wavelet transform and Hilbert transform on the original pulse wave signal respectively.

Table 1. Partial time domain characteristic parameter table

Almost all the energy of human pulse wave is distributed between 0～10Hz. The spectrum energy within 10 Hz of a healthy person accounts for more than 99% of the total energy, which shows that the spectrum energy ratio contains certain physiological information. In the calculation of the characteristic parameters of the Fourier transform domain, the pulse wave is taken as the center, and the 10 left and right pulse waves as a whole are analyzed as a whole in the Fourier transform domain. The characteristics of the Fourier transform domain mainly include fundamental frequency, second harmonic frequency, fundamental energy, second harmonic energy, total energy, energy of certain characteristic frequency bands, energy ratio, and energy of each frequency band and corresponding amplitude Ratio etc.

In wavelet domain feature extraction, the'db6' wavelet is used as the mother wavelet, and the pulse wave signal is decomposed into 6 layers to obtain the low-frequency approximation signal A _j and the high-frequency detail signal D _j . According to the wavelet coefficient spectrum of the pulse wave, the single-layer reconstructed detail signals of layers 1 and 2 contain a lot of high-frequency noise, so the energy of the detail signals of each layer is calculated and normalized using the detail signals of layers 3-6 To obtain the energy percentage of the detail signal, construct the wavelet coefficient energy feature vector E=[E _d3 E _d4 E _d5 E _d6 ]; at the same time, use the single reconstructed detail signal d ₁ -d ₆ and the approximate signal a ₆ to calculate each The layer energy is normalized to p ₁ -p ₇ , and the wavelet entropy can be obtained by using Formula 2-1. The wavelet entropy can represent the change of the pulse wave complexity in the time domain, and can also characterize many frequency domain characteristics of the pulse wave. Has good time-frequency localization ability.

W _E =-∑p _j logp _j

Perform EEMD decomposition on the pulse wave to obtain N IMF components C _i (t), i=1, 2, ..., N. Use Equation 2-2 to obtain the IMF energy moment. The IMF energy moment can not only reflect the energy of the IMF component of the pulse wave signal, but also reflect the distribution of its parameters with time.

In addition, performing the Hilbert transform on the obtained IMF component C _i (t) according to formula 2-3 can obtain the time-frequency distribution of the pulse wave signal energy. Find the Hilbert marginal spectrum h(w) from formula 2-4, and then find the area enclosed by h ² (w) and the frequency axis in the natural frequency interval w ₁ ～ w ₂ from formula 2-5, that is, the characteristic energy of the marginal spectrum .

The generating module 120 is connected to the extracting module 110, and is used to adjust the weights of the training data and the weak predictor according to the error of the training data in the feature parameters to form a strong predictor.

Specifically, you can use the same data to train multiple neural networks, use multiple neural networks as weak predictors, adjust the weights of each sample according to the prediction error of the training data, and determine the weights of the weak predictors to form strong Predictor. In wavelet domain feature extraction, the'db6' wavelet is used as the mother wavelet, and the pulse wave signal is decomposed into 6 layers to obtain the low-frequency approximation signal A _j and the high-frequency detail signal D _j . According to the wavelet coefficient spectrum of the pulse wave, the single-layer reconstructed detail signals of layers 1 and 2 contain a lot of high-frequency noise, so the energy of the detail signals of each layer is calculated and normalized using the detail signals of layers 3-6 To obtain the energy percentage of the detail signal, construct the wavelet coefficient energy feature vector E=[E _d3 E _d4 E _d5 E _d6 ]; at the same time, use the single reconstructed detail signal d ₁ -d ₆ and the approximate signal a ₆ to calculate each The layer energy is normalized to p ₁ -p ₇ , and the wavelet entropy can be obtained by using Formula 2-1. The wavelet entropy can represent the change of the pulse wave complexity in the time domain, and can also characterize many frequency domain characteristics of the pulse wave. Has good time-frequency localization ability.

W _E =-∑p _j logp _j

Specifically, you can use the same data to train multiple neural networks, use multiple neural networks as weak predictors, adjust the weight of each sample according to the prediction error of the training data, and determine the weight of the weak predictor to form a strong predictor. Predictor.

The output module 130 is connected to the generation module 120, and is used to input the characteristic parameter into the strong predictor to output blood pressure.

Therefore, an apparatus for calculating blood pressure provided by an embodiment of the present application extracts characteristic parameters from a pulse wave through an extraction module, and the characteristic parameters include time-domain characteristic parameters, wavelet-domain characteristic parameters, Fourier transform-domain characteristic parameters, and Greek Feature parameters in the Albert transform domain; the generation module adjusts the weights of the training data and weak predictors according to the error of the training data in the feature parameters to form a strong predictor; the output module inputs the feature parameters into the The strong predictor can output blood pressure, and can measure blood pressure without measuring other physiological signals, and realize continuous blood pressure calculation. By only acquiring a pulse wave signal, the systolic blood pressure of the subject can be continuously detected in real time. And diastolic blood pressure, has the advantages of low cost, low environmental requirements, comfort, convenience and higher pertinence. The device for calculating blood pressure provided by the embodiment of the present application only uses the pulse wave as a physiological signal as an intermediate result, but fully excavates from the time domain, Fourier transform domain, wavelet domain and Hilbert transform domain The physiological information contained in the pulse wave signal can improve the accuracy of blood pressure calculation.

FIG. 2 shows a schematic structural diagram of an apparatus for calculating blood pressure according to another embodiment of the present application. The apparatus 100 includes an extraction module 110, a generation module 120, and an output module 130.

For the pulse wave feature parameter extraction technology, please refer to the detailed description of the embodiment of FIG. 1 for details, and details are not described herein again.

In a possible implementation manner, the generating module 120 is used to divide a single subject data into multiple groups by using a 10-fold cross-validation method, one of the data is used as the test data, and the other group of data is used as the training Data; using back propagation BP neural network 121, support vector machine 122 and deep belief network 123 for model training and prediction, the back propagation (English: Backpropagation, abbreviation: BP) neural network, support vector machine and depth The belief network method is used as 3 weak predictors; according to the error of the training data in the feature parameters on the 3 weak predictors, the weights of the training data and the 3 weak predictors are adjusted to form a strong predictor As a result, the blood pressure calculation accuracy of the strong predictor is higher than the three weak predictors of BP neural network, support vector machine and deep belief network, and finally the multiple characteristic parameters (such as 78) of the subject are used as Input, you can get the exact blood pressure value of the subject.

Specifically, the Adaboost iterative algorithm can be used to train different classifiers against the same training set, such as the above three weak classifiers, and then combine these weak classifiers to form a stronger final classifier, such as the above strong classifier Device.

Specifically, the BP neural network has 78 nodes in the input layer, 5 nodes in the hidden layer, and 1 node in the output layer. In a possible implementation, the generation module 120 may be used to adopt particle swarm Optimize the algorithm to determine the network weights and network thresholds of the network.

The input of the BP neural network is the original feature parameter value, and the output is the calculated blood pressure value; the support vector machine first normalizes 78 feature parameters and uses a radial basis function (English: Radial Basis Function, abbreviation: RBF), At the same time, the penalty parameter c and the kernel function parameter g are in the range of -8-8, and the best model parameters are selected by cross-validation method to achieve a more accurate calculation of systolic blood pressure; deep belief network (abbreviation: DBN ) Method can generate very good parameter initialization values, avoiding the shortcomings of random initialization that causes the network to fall into global optimization and long training time. For deep belief networks, first use the contrast divergence algorithm to perform unsupervised training on two restricted Boltzmann machines (abbreviations: RBM), and use the output of the first RBM network as the input of the second RBM network; two RBMs The structure of the network is 78×20 and 20×5 respectively, and then the parameters of the trained RBM network are used to initialize the weights of the neural network. The structure of the neural network is 78×20×5×1. At this time, the neural network has only the last layer of weights randomly initialized, and then the training data is combined with the error back propagation algorithm to fine-tune the weights, and finally the trained DBN network is obtained.

Figure 3 shows a schematic diagram of the network structure of a restricted Boltzmann machine. In RBM, a weight W between any two connected neurons represents the strength of its connection, and each neuron has its own bias coefficient b (For explicit layer neurons) and c (for hidden layer neurons) to express their own bias values, obtained through learning, v is the input vector, h is the output vector.

For a piece of sample data x, the contrast divergence algorithm is used to train it. The specific steps of training are as follows:

1. Assign x to the apparent layer v ₁ , and calculate the probability P(h ₁ |v ₁ ) of each neuron in the hidden layer using formula (3-1);

P(h _j |v)=σ(b _j +Σ _i W _ij x _i )

2. Take a sample from Gibbs sampling from the calculated probability distribution: h ₁ ～P(h ₁ |v ₁ );

3. Reconstruct the display layer with h ₁ , that is, invert the display layer through the hidden layer, and calculate the probability of activation of each neuron in the display layer using the formula (3-2): P(v ₂ |h ₁ );

P(v _i |h)=σ(c _i +Σ _j W _ij h _j )

4. Take a sample from Gibbs sampling from the obtained probability distribution: v ₂ ～P(v ₂ |h ₁ );

5. Calculate the probability of activation of each neuron in the hidden layer again by v ₂ to obtain the probability distribution P(h ₂ |v ₂ );

6. Update the weight according to formula (3-3):

W←W+λ(P(h ₁ |v ₁ )v ₁ -P(h ₂ |v ₂ )v ₂ )

b←b+λ(v ₁ -v ₂ )

c←c+λ(h ₁ -h ₂ )

When the usual network is created, the weights and thresholds of the neural network are generally initialized to random numbers in the interval [-1,1], but this also results in a certain degree of difference in the training results of the same sample each time, and it is also easy to fall into The local optimal solution generally needs multiple tests to select the one with the smallest error as the trained network, and it is difficult to guarantee the global optimal solution. Based on this, the particle swarm optimization algorithm is used in this step to find the optimal network weights and thresholds, so that the calculation results of the same sample training remain stable and the global optimal solution, reducing the repetitive process of multi-test training.

The output module 130 is connected to the generation module 120 and used to input the characteristic parameter into the strong predictor to output blood pressure.

Therefore, an apparatus for calculating blood pressure provided by an embodiment of the present application extracts characteristic parameters from a pulse wave through an extraction module, and the characteristic parameters include time-domain characteristic parameters, wavelet-domain characteristic parameters, Fourier transform-domain characteristic parameters, and Greek Feature parameters in the Albert transform domain; the generation module adjusts the weights of the training data and weak predictors according to the error of the training data in the feature parameters to form a strong predictor; the output module inputs the feature parameters into the The strong predictor can output blood pressure, and can measure blood pressure without measuring other physiological signals, and realize continuous blood pressure calculation. By only acquiring a pulse wave signal, the systolic blood pressure of the subject can be continuously detected in real time. And diastolic blood pressure.

Therefore, a blood pressure calculation device provided by an embodiment of the present application divides a single subject's data into multiple groups by using a 10-fold cross-validation method through the generation module, and uses one group of data as test data and other groups The data is used as the training data; the methods of back propagation BP neural network, support vector machine and deep belief network are used for model training and prediction, and the method of back propagation BP neural network, support vector machine and deep belief network are used. As a weak predictor; adjust the weight of the training data and the weak predictor according to the error of the training data in the feature parameters to form a strong predictor, which can make the blood pressure calculation accuracy of the strong predictor higher than the BP neural network , Support vector machine and deep belief network three weak predictors to improve the accuracy of blood pressure calculation.

Therefore, an apparatus for calculating blood pressure provided by an embodiment of the present application is used by a generation module to adopt a particle swarm optimization algorithm to determine the network weights and network thresholds of the network. The optimal network weights and thresholds can be set so that The calculation results of the same sample training remain stable and the global optimal solution, reducing the repetitive process of multi-test training.

FIG. 4 shows a schematic structural diagram of an apparatus for calculating blood pressure according to another embodiment of the present application. The apparatus 100 includes: an extraction module 110, a generation module 120, an output module 130, and a preprocessing module 140.

Specifically, you can use the same data to train multiple neural networks, use multiple neural networks as weak predictors, adjust the weights of each sample according to the calculation error of the training data, and determine the weights of the weak predictors to form strong Predictor.

In a possible implementation, the generation module 120 is used to divide the individual test data into 10 groups by using the 10-fold cross-validation method, and one of the data is used as the test data, and the other 9 groups of data are used as the Training data; using back propagation BP neural network 121, support vector machine 122 and deep belief network 123 for model training and calculation, the back propagation (English: Backpropagation, abbreviation: BP) neural network, support vector machine and The method of deep belief network is used as 3 weak predictors; according to the error of the training data in the feature parameters on the 3 weak predictors, the weights of the training data and the 3 weak predictors are adjusted to form a strong prediction Therefore, the blood pressure prediction accuracy of the strong predictor is higher than the three weak predictors of BP neural network, support vector machine and deep belief network, and finally achieves multiple feature parameters of the subject (for example, 78) As input, the exact blood pressure value of the subject can be obtained.

The input of the BP neural network is the original feature parameter value, and the output is the calculated blood pressure value; the support vector machine first normalizes 78 feature parameters and uses a radial basis function (English: Radial Basis Function, abbreviation: RBF), At the same time, the penalty parameter c and the kernel function parameter g are in the range of -8-8, and the best model parameters are selected by cross-validation method to achieve a more accurate calculation of systolic blood pressure; deep belief network (abbreviation: DBN ) Method can generate very good parameter initialization values, avoiding the shortcomings of random initialization that causes the network to fall into global optimization and long training time. For deep belief networks, first use the contrast divergence algorithm to perform unsupervised training on two restricted Boltzmann machines (abbreviations: RBM), and use the output of the first RBM network as the input of the second RBM network; two RBMs The structure of the network is 78×20 and 20×5 respectively, and then the parameters of the trained RBM network are used to initialize the weights of the neural network. The structure of the neural network is 78×20×5×1. At this time, the neural network has only the last layer of weights randomly initialized, and then the training data is combined with the error back propagation algorithm to fine-tune the weights, and finally the trained DBN network is obtained. For the network structure of RBM, please refer to the description in FIG. 3, which will not be repeated here.

The preprocessing module 140 is connected to the extraction module 110, and is used for preprocessing the training data to remove abnormal feature parameters.

In a possible implementation, the preprocessing module 140 is used to abnormally mark the feature parameters in the training data that are greater than the first predetermined multiple of the average or smaller than the second predetermined multiple of the average; when the When the number of abnormal markers exceeds a predetermined abnormal marker threshold, the training data is deleted.

In a possible implementation manner, the preprocessing module 140 is configured to delete the current calculated blood pressure when the amount of change between this calculated blood pressure and the adjacent previous calculated blood pressure exceeds a predetermined change threshold. Due to the existence of abnormal data, it will affect the training accuracy of the model and affect the accuracy of blood pressure calculation. This step can eliminate abnormal data, avoid the impact of abnormal data on training accuracy, and improve the accuracy of blood pressure calculation.

Specifically, it may include: marking the corresponding position of the feature parameter mean value greater than 1.7 times or the feature parameter mean value less than 0.3 times in the training data as 1, and using the systolic blood pressure as a feature, and changing the adjacent two systolic blood pressures by more than 5 The corresponding position is marked as 1. Count the number of abnormal feature parameters in each set of training data. If the amount of abnormal data exceeds a certain threshold, the set of data will be removed as abnormal data.

The output module 130 is connected to the generation module 120 and used to input the characteristic parameter into the strong predictor to output the calculated blood pressure.

As an implementation, the device for calculating blood pressure provided by the embodiments of the present application can be combined with medical equipment on the one hand, and can be combined with wearable devices such as smart bracelets on the one hand, and the method can be used on the other hand. Combined with smart terminal products such as smartphones, blood pressure can be measured conveniently, non-invasively, in real time, and continuously.

Therefore, an apparatus for calculating blood pressure provided by an embodiment of the present application extracts characteristic parameters from a pulse wave through an extraction module, and the characteristic parameters include time-domain characteristic parameters, wavelet-domain characteristic parameters, Fourier transform-domain characteristic parameters, and Greek Feature parameters in the Albert transform domain; the generation module adjusts the weights of the training data and weak predictors according to the error of the training data in the feature parameters to form a strong predictor; the output module inputs the feature parameters into the The strong predictor uses the output to calculate blood pressure. It can achieve continuous blood pressure calculation without measuring other physiological signals under the premise of measuring the pulse wave signal. Only by acquiring a section of pulse wave signal can the subject's contraction be continuously detected in real time Pressure and diastolic pressure.

Therefore, an apparatus for calculating blood pressure provided by an embodiment of the present application preprocesses the training data through a preprocessing module to remove abnormal feature parameters, can remove abnormal data, and avoid the impact of abnormal data on training accuracy. Improve the accuracy of blood pressure calculation.

FIG. 5 shows a schematic structural diagram of an apparatus for calculating blood pressure according to another embodiment of the present application. The apparatus 100 includes an extraction module 110, a generation module 120, an output module 130, a preprocessing module 140, and a feature parameter optimization module 150.

In a possible implementation, the generation module 120 is used to divide the individual test data into 10 groups by using the 10-fold cross-validation method, and one of the data is used as the test data, and the other 9 groups of data are used as the Training data; using back propagation BP neural network 121, support vector machine 122 and deep belief network 123 for model training and calculation, the back propagation (English: Backpropagation, abbreviation: BP) neural network, support vector machine and The method of deep belief network is used as 3 weak predictors; according to the error of the training data in the feature parameters on the 3 weak predictors, the weights of the training data and the 3 weak predictors are adjusted to form a strong prediction Therefore, the blood pressure calculation accuracy of the strong predictor is higher than the three weak predictors of BP neural network, support vector machine and deep belief network, and finally achieves multiple feature parameters of the subject (for example, 78) As input, the exact blood pressure value of the subject can be obtained.

The feature parameter optimization module 150 is connected to the generation module 120, and a specific feature parameter is selected from the feature parameters through an average influence value method.

Specifically, it may include: optimizing the pulse wave characteristic parameters by using the method of MIV average influence value. Because there are certain differences between each individual, these differences lead to the 78 wave frequency sequence characteristics of the pulse wave sequence on the calculation of blood pressure in each individual or each group of individuals is very different. Therefore, the method of MIV average influence value is used to select the most influential feature parameters for each individual to calculate the blood pressure, while ensuring the same high accuracy as 78 feature parameter calculations, while reducing the excessive input parameters of the neural network and too much information The complexity leads to the instability of the calculation accuracy and further improves the calculation accuracy.

Therefore, the device for calculating blood pressure provided by the embodiment of the present application, through the feature parameter optimization module, selects specific feature parameters from the feature parameters through the average impact value method, and can filter out the most influential feature parameters for each individual Performing blood pressure calculation further improves the accuracy of blood pressure calculation.

FIG. 6 shows a schematic diagram of a hardware structure of an electronic device for implementing an embodiment of the present application. Referring to the figure, at the hardware level, the electronic device includes a processor, and optionally includes an internal bus, a network interface, and a memory. The memory may include a memory, such as a high-speed random access memory (Random-Access Memory, RAM), or may also include a non-volatile memory (non-volatile memory), such as at least one disk memory. Of course, the electronic device may also include hardware required for other services.

The processor, network interface, and memory can be connected to each other via an internal bus, which can be an industry standard architecture (ISA) bus, a peripheral component interconnect standard (Peripheral Component Interconnect, PCI) bus, or an extended industry standard Structure (Extended Industry Standard Architecture, EISA) bus, etc. The bus can be divided into an address bus, a data bus, and a control bus. For ease of representation, this figure is only indicated by a bidirectional arrow, but it does not mean that there is only one bus or one type of bus.

Memory for storing programs. Specifically, the program may include program code, and the program code includes a computer operation instruction. The memory may include memory and non-volatile memory, and provide instructions and data to the processor.

The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it, forming a device to locate the target user at a logical level. The processor executes the program stored in the memory and is specifically used to execute: extracting characteristic parameters from the pulse wave, the characteristic parameters including time domain characteristic parameters, wavelet domain characteristic parameters, Fourier transform domain characteristic parameters and Hilbert Transform domain feature parameters; adjust the weights of the training data and weak predictors according to the error of the training data in the feature parameters to form a strong predictor; input the feature parameters into the strong predictor to output the calculated blood pressure . The method disclosed in the embodiment shown in FIG. 1 of the present application may be applied to a processor, or implemented by a processor. In other words, the processor can implement the steps performed by the modules in the figure and obtain the same or similar effects.

In a possible implementation, according to the error of the training data in the feature parameters, adjusting the weights of the training data and the weak predictor to form a strong predictor includes performing: using a 10-fold cross-validation method, the The individual test data is divided into 10 groups, one of which is used as the test data, and the other 9 groups of data are used as the training data; the models are back-propagation BP neural network, support vector machine, and deep belief network respectively. Training and calculation, using the methods of back propagation BP neural network, support vector machine and deep belief network as weak predictors; according to the error of the training data in the feature parameters, the training data and weak predictor The weights are adjusted to form a strong predictor.

In a possible implementation, the methods of back propagation BP neural network, support vector machine and deep belief network are used for model training and calculation, including execution: using particle swarm optimization algorithm to determine the network weight and Network threshold. The above method disclosed in the embodiment shown in FIG. 2 of the present application may be applied to a processor, or implemented by a processor. In other words, the processor can implement the steps performed by the modules in the figure and obtain the same or similar effects.

Before adjusting the weights of the training data and the weak predictor to form a strong predictor according to the error of the training data in the feature parameters, the method further includes performing: preprocessing the training data to remove abnormal feature parameters .

In a possible implementation, preprocessing the training data and removing abnormal feature parameters includes performing: performing abnormalities on the feature parameters in the training data that are greater than the first predetermined multiple of the mean or less than the second predetermined multiple of the mean Mark; when the number of abnormal marks in the training data exceeds a predetermined abnormal mark threshold, delete the training data.

In a possible implementation manner, after inputting the characteristic parameter into the strong predictor to output the calculated blood pressure, the method further includes: when the amount of change between this calculated blood pressure and the adjacent previous calculated blood pressure exceeds When the change threshold is predetermined, the blood pressure calculated this time is deleted. The above method disclosed in the embodiment shown in FIG. 4 of the present application may be applied to a processor, or implemented by a processor. In other words, the processor can implement the steps performed by the modules in the figure and obtain the same or similar effects.

In a possible implementation manner, before inputting the characteristic parameter into the strong predictor to output the calculated blood pressure, it further includes performing: filtering a specific characteristic parameter from the characteristic parameters by an average influence value method. The above method disclosed in the embodiment shown in FIG. 5 of the present application may be applied to a processor, or implemented by a processor. In other words, the processor can implement the steps performed by the modules in the figure and obtain the same or similar effects.

The processor may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method may be completed by an integrated logic circuit of hardware in the processor or instructions in the form of software. The aforementioned processor may be a general-purpose processor, including a central processor (Central Processing Unit, CPU), a network processor (Network Processor, NP), etc.; it may also be a digital signal processor (Digital Signal Processor, DSP), dedicated integration Circuit (Application Specific Integrated Circuit, ASIC), field programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps, and logical block diagrams disclosed in the embodiments of the present application may be implemented or executed. The general-purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The steps of the method disclosed in conjunction with the embodiments of the present application may be directly embodied and executed by a hardware decoding processor, or may be executed and completed by a combination of hardware and software modules in the decoding processor. The software module may be located in a mature storage medium in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, or electrically erasable programmable memory, and registers. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

The electronic device may also execute the methods described in the foregoing method embodiments, and implement the functions and beneficial effects of the methods described in the foregoing method embodiments, which will not be repeated here.

Of course, in addition to the software implementation, the electronic device of the present application does not exclude other implementations, such as a logic device or a combination of software and hardware, etc., that is to say, the execution body of the following processing flow is not limited to each logical unit, It can also be a hardware or logic device.

An embodiment of the present application also provides a computer-readable storage medium that stores one or more programs. When the one or more programs are executed by an electronic device including multiple application programs, the The electronic device performs the following operations: extracting feature parameters from the pulse wave, the feature parameters including time domain feature parameters, wavelet domain feature parameters, Fourier transform domain feature parameters and Hilbert transform domain feature parameters; based on the features The error of the training data in the parameters adjusts the weights of the training data and the weak predictor to form a strong predictor; the characteristic parameters are input to the strong predictor to output the calculated blood pressure.

Wherein, the computer-readable storage medium includes read-only memory (Read-Only Memory, ROM for short), random access memory (Random Access Memory, RAM for short), magnetic disk or optical disk, etc.

Further, an embodiment of the present application further provides a computer program product, the computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program includes program instructions, and when the program instructions When executed by a computer, the following process is realized: extracting feature parameters from the pulse wave, the feature parameters include time domain feature parameters, wavelet domain feature parameters, Fourier transform domain feature parameters, and Hilbert transform domain feature parameters; The error of the training data in the feature parameters adjusts the weights of the training data and the weak predictor to form a strong predictor; the feature parameters are input to the strong predictor to output the calculated blood pressure.

In short, the above is only the preferred embodiment of the present application and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of this application shall be included in the scope of protection of this application.

The system, device, module or unit explained in the above embodiments may be specifically implemented by a computer chip or entity, or implemented by a product having a certain function. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or A combination of any of these devices.

Computer readable media, including permanent and non-permanent, removable and non-removable media, can store information by any method or technology. The information may be computer readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, read-only compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, Magnetic tape cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media can be used to store information that can be accessed by computing devices. As defined in this article, computer-readable media does not include temporary computer-readable media (transitory media), such as modulated data signals and carrier waves.

It should also be noted that the terms "include", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device that includes a series of elements not only includes those elements, but also includes Other elements not explicitly listed, or include elements inherent to such processes, methods, goods, or equipment. Without more restrictions, the element defined by the sentence "include one..." does not exclude that there are other identical elements in the process, method, commodity or equipment that includes the element.

The embodiments in this specification are described in a progressive manner. The same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant part can be referred to the description of the method embodiment.

Claims

A device for calculating blood pressure, including:

An extraction module, configured to extract feature parameters from the pulse wave, the feature parameters including time domain feature parameters, wavelet domain feature parameters, Fourier transform domain feature parameters and Hilbert transform domain feature parameters;

A generation module, connected to the extraction module, for adjusting the weight of the training data and the weak predictor according to the error of the training data in the feature parameters to form a strong predictor;

The output module is connected to the generation module and used to input the characteristic parameter into the strong predictor to output blood pressure.
The device according to claim 1, wherein the generating module is used to divide a single subject data into multiple groups by using a ten-fold cross-validation method, one of which is used as test data, and the other is used as data The training data; using back propagation BP neural network, support vector machine and deep belief network method for model training and calculation, using the back propagation BP neural network, support vector machine and deep belief network method as weak Predictor; adjust the weight of the training data and the weak predictor according to the error of the training data in the feature parameters to form a strong predictor.
The apparatus according to claim 2, wherein the generation module is used to determine a network weight and a network threshold of the network using a particle swarm optimization algorithm.
The device of claim 1, further comprising:

The preprocessing module is connected to the extraction module and is used for preprocessing the training data to remove abnormal feature parameters.
The apparatus according to claim 4, wherein the preprocessing module is configured to abnormally mark the feature parameters in the training data that are greater than the first predetermined multiple of the average or smaller than the second predetermined multiple of the average; when the training data When the number of abnormal markers exceeds a predetermined abnormal marker threshold, the training data is deleted.
The device of claim 1, further comprising:

The characteristic parameter optimization module is connected to the generation module, and selects a specific characteristic parameter from the characteristic parameters through an average influence value method.
The apparatus according to claim 4, wherein the pre-processing module is further configured to delete the current time when the amount of change between the blood pressure calculated this time and the blood pressure calculated adjacent to the previous time exceeds a predetermined change threshold Calculated blood pressure.
An electronic device, including:

Processor; and

A memory arranged to store computer-executable instructions that when executed uses the processor to perform the following operations:

Feature parameters are extracted from the pulse wave, and the feature parameters include time domain feature parameters, wavelet domain feature parameters, Fourier transform domain feature parameters, and Hilbert transform domain feature parameters;

Adjust the weight of the training data and the weak predictor according to the error of the training data in the feature parameters to form a strong predictor;

The characteristic parameter is input to the strong predictor to output blood pressure.
The electronic device according to claim 8, wherein adjusting the weights of the training data and the weak predictor to form the strong predictor according to the error of the training data in the feature parameters includes:

Using the 10-fold cross-validation method, the single test data is divided into multiple groups, one of which is used as the test data, and the other data is used as the training data;

Respectively use the methods of back propagation BP neural network, support vector machine and deep belief network for model training and calculation, and use the methods of back propagation BP neural network, support vector machine and deep belief network as weak predictors;

According to the error of the training data in the characteristic parameters, the weights of the training data and the weak predictor are adjusted to form a strong predictor.
The electronic device according to claim 9, wherein model training and calculation using back propagation BP neural network, support vector machine and deep belief network respectively include:

A particle swarm optimization algorithm is used to determine the network weights and network thresholds of the network.
The electronic device according to claim 8, wherein before adjusting the weights of the training data and the weak predictor according to the error of the training data in the feature parameter to form a strong predictor, further comprising:

Preprocessing the training data to eliminate abnormal feature parameters.
The electronic device according to claim 11, wherein the training data is preprocessed to remove abnormal feature parameters including:

Abnormally mark the feature parameters in the training data that are greater than the first predetermined multiple of the mean or less than the second predetermined multiple of the mean;

When the number of abnormal markers in the training data exceeds a predetermined abnormal marker threshold, the training data is deleted.
The electronic device according to claim 8, wherein before inputting the characteristic parameter to the strong predictor to output blood pressure, further comprising:

Through the average influence value method, specific characteristic parameters are selected from the characteristic parameters.
The electronic device according to claim 8, wherein after inputting the characteristic parameter to the strong predictor to output blood pressure, further comprising:

When the amount of change between the blood pressure calculated this time and the adjacent previously calculated blood pressure exceeds a predetermined change threshold, the blood pressure calculated this time is deleted.
A computer-readable medium that stores one or more programs that, when executed by an electronic device including multiple application programs, causes the electronic device to perform the following operations:

Feature parameters are extracted from the pulse wave, and the feature parameters include time domain feature parameters, wavelet domain feature parameters, Fourier transform domain feature parameters, and Hilbert transform domain feature parameters;

Adjust the weight of the training data and the weak predictor according to the error of the training data in the feature parameters to form a strong predictor;

The characteristic parameter is input to the strong predictor to output blood pressure.