A blood pressure measurement method and device

Abstract

The present invention discloses a blood pressure measurement method and device, comprising the following steps: S1: obtaining the pressure and oscillatory waves within a cuff through a pressure sensing device, and obtaining Korotkoff sounds through a Korotkoff sound sensing device; S2: performing feature extraction on the oscillatory waves within the cuff to obtain oscillatory wave features; performing feature extraction on the Korotkoff sounds to obtain Korotkoff sound features; S3: fusing the oscillatory wave features, the Korotkoff sound features, and the pressure within the cuff to obtain comprehensive features; and S4: inputting the comprehensive features into a prediction model to obtain systolic pressure, diastolic pressure, and pulse rate. The present invention utilizes data from two modalities, Korotkoff sounds and oscillatory waves, to enrich the extracted features. The output of the prediction model includes three parts: systolic pressure, diastolic pressure, and pulse rate. The constructed loss function is a multi-objective loss function. Multi-objective learning is used to improve the generalization ability of the model, combining the advantages of the oscillometric method and the auscultatory method, and improving the accuracy and anti-interference ability of blood pressure measurement.

Description

Technical Field

[0001] The present invention relates to the technical field of blood pressure measurement, and in particular to a blood pressure measurement method and device. Background Art

[0002] In the existing technology, there are two main methods for non-invasive blood pressure measurement: auscultation and oscillometrics. Auscultation, also known as the Korotkoff sound method, is divided into manual Korotkoff sound measurement and electronic Korotkoff sound measurement. The manual Korotkoff sound method is the method commonly seen by doctors and nurses using a pressure gauge and stethoscope to measure blood pressure. The electronic Korotkoff sound method is an electronic blood pressure measurement method. The basic principle is to use electronic technology to perform the manual Korotkoff sound method. Specifically, an air pump is used to inflate and deflate the cuff, and an electronic pickup is used to listen to the Korotkoff sounds. The judgment method is almost the same as the manual method, except that a computer replaces human judgment. The oscillometric method, also known as the oscillation wave method, is an electronic measurement method. Its principle is to automatically pressurize the cuff to block the brachial artery blood flow, and then slowly decompress it. During this period, a pressure oscillation wave will be generated in the cuff. This oscillation wave is monitored by an air pressure sensor. As the cuff pressure gradually decreases, the amplitude of the pressure oscillation wave gradually increases. The amplitude is maximum when the cuff pressure reaches the mean arterial blood pressure, and then the pressure oscillation wave gradually decreases. During the entire decompression process, the amplitude of the pressure oscillation wave forms a bell-shaped envelope. The systolic and diastolic blood pressures are determined by the oscillation wave through a software algorithm, and the entire process is completed automatically.

[0003] Both measurement methods have some disadvantages: Disadvantages of the manual Korotkoff sound method: 1. Users must undergo a period of learning and training; 2. The user's experience and hearing affect the measurement results. Disadvantages of the electronic Korotkoff sound method: Existing electronic Korotkoff sound methods use algorithms to identify Korotkoff sounds. First, the features of the Korotkoff sounds are extracted, and then a classifier is used to identify the Korotkoff sounds. Commonly used classifier models include logistic regression, support vector machines, decision trees, and random forests. Current Korotkoff sound recognition algorithms do not consider the relationship between Korotkoff sounds at different stages, resulting in poor anti-interference ability and recognition accuracy. Disadvantages of the oscillometric method: The oscillometric method uses two algorithms to measure blood pressure: the amplitude coefficient method and the waveform feature method. For an individual, neither the waveform feature method nor the amplitude coefficient method can clearly identify a point where the external pressure equals the blood pressure. Therefore, blood pressure measurements using the oscillometric method are based on statistical inferences, which can result in large errors for some individuals, making accurate blood pressure measurement difficult.

[0004] Therefore, in response to the above problems, a blood pressure measurement method and device are proposed, which combines the advantages of the oscillometric method and the auscultatory method to improve the accuracy and anti-interference ability of blood pressure measurement. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to overcome the existing defects and provide a blood pressure measurement method and equipment, which utilizes data from two modes, Korotkoff sounds and oscillation waves, to enrich the extracted features. The output of the prediction model includes three parts: systolic pressure, diastolic pressure, and pulse rate. The constructed loss function is a multi-objective loss function. Multi-objective learning is used to improve the generalization ability of the model. The advantages of the oscillometric method and the auscultation method are combined to improve the accuracy and anti-interference ability of blood pressure measurement, which can effectively solve the problems in the background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a blood pressure measurement method comprising the following steps:

[0007] S1 obtains the pressure and shock wave in the cuff through the pressure sensor device and obtains Korotkoff sound through the Korotkoff sound sensor device;

[0008] S2 performs feature extraction on the shock wave in the cuff to obtain shock wave features; performs feature extraction on the Korotkoff sound to obtain Korotkoff sound features;

[0009] S3 combines shock wave characteristics, Korotkoff sound characteristics and pressure in the cuff to obtain comprehensive characteristics;

[0010] S4 inputs the comprehensive features into the prediction model to obtain systolic blood pressure, diastolic blood pressure and pulse rate.

[0011] Furthermore, in S1 , it is also necessary to perform preprocessing on the acquired pressure, shock waves in the cuff, and Korotkoff sounds.

[0012] Furthermore, the pretreatment method includes:

[0013] (1) Use a noise reduction algorithm to reduce the pressure, shock waves, and Korotkoff sounds in the cuff;

[0014] (2) Normalize the pressure inside the cuff, the shock wave inside the cuff, and the Korotkoff sound.

[0015] Furthermore, in S2, the method for extracting shock wave features includes the following steps:

[0016] (1) Divide the preprocessed shock wave into multiple time segments, calculate the recursive graph for each time segment to obtain the recursive graph sequence. The recursive graph calculation formula is R i，j (ε)=H(ε-||y i -y j ||, where ε is the distance threshold, H(·) is the Heaviside function, and y i 、y jThe calculation formula is y(θ, δ) = {x(θ), x(θ+δ), x(θ+2δ), …, x(θ+(d-1)δ)}, where x(θ) is the oscillation wave, δ is the phase delay parameter, and d is the dimension parameter;

[0017] (2) The recursive graph sequence is sequentially passed through the convolutional neural network to obtain the shock wave feature sequence. The convolutional neural network is structurally divided into convolutional layer, ReLU layer, Max-pooling layer, Drop-out layer, convolutional layer, ReLU layer, Max-pooling layer and Drop-out layer. The parameters in each layer can be customized, and the number of convolutional layers can also be customized.

[0018] Furthermore, in S2, the method for extracting Korotkoff sound features includes the following steps:

[0019] (1) Divide the preprocessed Korotkoff sounds into multiple time segments, calculate the recursive graph for each time segment to obtain the recursive graph sequence. The recursive graph calculation formula is R i，j (ε)=H(ε-||y i -y j ||, where ε is the distance threshold, H(·) is the Heaviside function, and y i 、y j The calculation formula is y(θ, δ) = {x(θ), x(θ+δ), x(θ+2δ), …, x(θ+(d-1)δ)}, where x(θ) is the Korotkoff sound, δ is the phase delay parameter, and d is the dimension parameter;

[0020] (2) The recursive graph sequence is sequentially passed through a convolutional neural network to obtain a Korotkoff sound feature sequence. The convolutional neural network is structurally divided into a convolutional layer, a ReLU layer, a Max-pooling layer, a Drop-out layer, a convolutional layer, a ReLU layer, a Max-pooling layer, and a Drop-out layer. The parameters in each layer can be customized, and the number of convolutional layers can also be customized.

[0021] Furthermore, in S3, the shock wave feature sequence, the Korotkoff sound feature sequence, and the pressure in the cuff are fused in time order to obtain a comprehensive feature sequence, wherein the shock wave feature sequence is a time series, the Korotkoff sound feature sequence is a time series, and the pressure in the cuff is a time series.

[0022] Furthermore, in S4, the prediction model is composed of an LSTM neural network followed by a fully connected layer. The LSTM neural network consists of multiple LSTM units. The objective function of training the prediction model is

[0023] Among them, L tot (X,Y 1:K) is the overall objective function, which is the sum of three sub-objective functions. K is 3. The sub-objective functions are the systolic blood pressure objective function L1(X, Y1), the diastolic blood pressure objective function L2(X, Y2) and the pulse rate objective function L3(X, Y3). X is the comprehensive feature of the input prediction model, Y1 is the true value of systolic blood pressure, Y2 is the true value of diastolic blood pressure, Y3 is the true value of pulse rate, and w i (t) is the weight that changes over time,

[0024] , N is 3, L n is the sub-objective function, r n is the ratio of the objective function, and T is an adjustable parameter.

[0025] A blood pressure measurement device includes a pressure sensor, a Korotkoff sound sensor, a processor, a drive circuit for an air pump valve, and a power supply circuit. The processor is respectively connected to the pressure sensor, the Korotkoff sound sensor, and the drive circuit for the air pump valve. The power supply circuit supplies power to the pressure sensor, the Korotkoff sound sensor, and the drive circuit for the air pump valve. The pressure sensor monitors the pressure and shock waves within the cuff, the Korotkoff sound sensor monitors the Korotkoff sounds, the drive circuit for the air pump valve drives the air pump valve to inflate and deflate the cuff, and the processor executes an analysis algorithm.

[0026] Furthermore, it also includes a shell, a wrist cuff and a through tube connecting the shell and the wrist cuff, an air bag is provided in the wrist cuff, a pressure sensor device, a Korotkoff sound sensor device, a processor, a driving circuit of the air pump valve and a power supply circuit are provided in the shell, the through tube is connected to the connector on the side of the shell, and the connector is connected to the driving air pump.

[0027] Furthermore, a display screen, an adjustment button and a switch button are provided on the surface of the shell. The display screen and the adjustment button are electrically connected to the processor respectively, and the switch button is electrically connected to the power circuit.

[0028] Compared with the existing technology, this blood pressure measurement method and device has the following advantages:

[0029] 1. The present invention obtains the pressure and shock waves in the cuff through a pressure sensing device, obtains Korotkoff sounds through a Korotkoff sound sensing device, and then extracts shock wave features and Korotkoff sound features respectively. The use of data from both Korotkoff sounds and shock waves makes the extracted features richer.

[0030] 2. The shock wave features, Korotkoff sound features, and the pressure inside the cuff are integrated to obtain comprehensive features. Finally, the comprehensive features are input into the prediction model to obtain systolic pressure, diastolic pressure, and pulse rate. The output of the prediction model includes three parts: systolic pressure, diastolic pressure, and pulse rate. The constructed loss function is a multi-objective loss function. The use of multi-objective learning improves the generalization ability of the model, combines the advantages of the oscillometric method and the auscultation method, and improves the accuracy and anti-interference ability of blood pressure measurement. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 This is a flow chart of a blood pressure measurement method of the present invention;

[0032] Figure 2 A graphical flow chart of a blood pressure measurement method of the present invention;

[0033] Figure 3 The waveform diagram of shock wave, Korotkoff sound and pressure in the present invention;

[0034] Figure 4 A recursive diagram for calculating shock waves and Korotkoff sounds in the present invention;

[0035] Figure 5 Extracting characteristic sequences from recursive graph sequences of shock waves and Korotkoff sounds in the present invention;

[0036] Figure 6 It is a fusion diagram of shock wave characteristics, Korotkoff sound characteristics and pressure characteristics in the present invention;

[0037] Figure 7 In the present invention, the comprehensive features are input into the prediction model to obtain systolic blood pressure, diastolic blood pressure and pulse rate;

[0038] Figure 8 This is a connection block diagram of a blood pressure measurement device in the present invention. DETAILED DESCRIPTION

[0039] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0040] The embodiments of this application acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to achieve optimal results.

[0041] Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0042] See also Figure 1-8 , this embodiment provides a technical solution:

[0043] like Figure 1 As shown, a blood pressure measurement method includes the following steps:

[0044] S1 obtains the pressure and shock wave in the cuff through the pressure sensor device and obtains Korotkoff sound through the Korotkoff sound sensor device;

[0045] S2 performs feature extraction on the shock wave in the cuff to obtain shock wave features; performs feature extraction on the Korotkoff sound to obtain Korotkoff sound features;

[0046] S3 combines shock wave characteristics, Korotkoff sound characteristics and pressure in the cuff to obtain comprehensive characteristics;

[0047] S4 inputs the comprehensive features into the prediction model to obtain systolic blood pressure, diastolic blood pressure and pulse rate.

[0048] like Figure 2As shown, a graphical flow chart of a blood pressure measurement method is shown. During the cuff inflation or deflation process, the pressure and oscillation waves in the cuff are obtained by a pressure sensing device, and Korotkoff sounds are obtained by a Korotkoff sound sensing device. The pressure, oscillation waves, and Korotkoff sounds in the cuff are preprocessed respectively. The preprocessing step includes two steps: first, using a noise reduction algorithm to reduce the noise of the pressure, oscillation waves, and Korotkoff sounds in the cuff. Second, the pressure, oscillation wave, and Korotkoff sounds in the cuff are normalized. The purpose of noise reduction is to improve the signal-to-noise ratio, and normalization is for the needs of the prediction model. The preprocessed oscillation wave is then divided into multiple time segments, and a recurrence plot (RP) is calculated for each time segment to obtain a recurrence plot sequence. The recurrence plot sequence is sequentially passed through a convolutional neural network to obtain an oscillation wave feature sequence. The preprocessed Korotkoff sounds are divided into multiple time segments, and a recurrence plot (RP) is calculated for each time segment to obtain a recurrence plot sequence. The recurrence plot sequence is sequentially passed through a convolutional neural network to obtain a Korotkoff sound feature sequence. The oscillation wave feature sequence, the Korotkoff sound feature sequence, and the pressure in the cuff are then fused in chronological order to obtain a comprehensive feature sequence. The comprehensive feature sequence is a multidimensional time series; it is then fed into a prediction model to obtain systolic blood pressure, diastolic blood pressure, and pulse rate. The prediction model consists of an LSTM neural network followed by a fully connected layer. The LSTM neural network consists of multiple LSTM units. Long short-term memory (LSTM) neural networks are a special type of recurrent neural network (RNN). During training, RNNs are prone to exploding or vanishing gradients as training time increases and the number of network layers increases. This makes it impossible to process long sequences of data and, consequently, to obtain information from long-distance data. LSTM is an excellent variant of RNN, inheriting most of the characteristics of RNN models while addressing the vanishing gradient problem caused by the gradual reduction of gradient backpropagation.

[0049] like Figure 3As shown in the figure, the waveforms of oscillation waves, Korotkoff sounds, and pressure are obtained. During cuff inflation or deflation, the pressure and oscillation waves within the cuff are acquired through a pressure sensor, and the Korotkoff sounds are acquired through a Korotkoff sound sensor. The acquired pressure, oscillation waves, and Korotkoff sounds are all one-dimensional time series. The length of the time series is determined by the inflation and deflation time and the sampling rate. The inflation and deflation time is determined by the cuff capacity and the inflation and deflation algorithm. The sampling rate is the rate at which the processor samples the signal, which is determined by the frequency of the Korotkoff sounds and oscillation waves. According to the Nyquist sampling theorem, a sampling rate greater than twice the sampled signal is required to fully preserve the signal information. If the processor speed allows, the sampling rate can be higher, generally at least four times the sampled signal. During cuff inflation or deflation, the pressure within the cuff continuously increases or decreases. The amplitude of the oscillation waves within the cuff increases and then decreases, and the intensity of the Korotkoff sounds increases and then decreases. The Korotkoff sounds can be divided into five phases (also called time periods).

[0050] like Figure 4 As shown in the figure, the recurrence plot of the oscillator and Korotkoff sound is calculated. The pre-processed oscillator is divided into N time segments of equal length. The recurrence plot (RP) is calculated for each time segment to obtain N recurrence plot sequences. The recurrence plot (RP) is an important method for analyzing the periodicity, chaos and non-stationarity of time series, and can reveal the internal structure of time series. The recurrence plot calculation formula is R i，j (ε)=H(ε-||y i -y j ||), where ε is the distance threshold, H(·) is the Heaviside function, and y i 、y j The calculation formula is y(θ, δ) = {x(θ), x(θ+δ), x(θ+2δ), ..., x(θ+(d-1)δ)}, where x(θ) is an oscillation wave or a Korotkoff sound, δ is a phase delay parameter, and d is a dimension parameter.

[0051] like Figure 5As shown, feature sequences are extracted from the recurrence graph sequences of oscillators and Korotkoff sounds. The oscillator recurrence graph sequence is sequentially passed through a convolutional neural network to obtain the oscillator feature sequence, and the Korotkoff sound recurrence graph sequence is sequentially passed through a convolutional neural network to obtain the Korotkoff sound feature sequence. The CNN backbone network is a neural network based on a convolutional neural network. It is used for feature extraction and generates feature maps. The Flatten layer "flattens" the feature map, converting multi-dimensional data into a single dimension. The CNN backbone network is a pre-trained deep convolutional neural network. Pre-trained deep convolutional neural networks are neural networks trained on large datasets using either supervised or self-supervised methods. Deep neural network training requires a large amount of data, and collecting and annotating this data consumes considerable resources. Pre-trained deep convolutional neural networks are a method for reducing development costs and enabling reuse.

[0052] like Figure 6 As shown in the figure, the fusion diagram of the shock wave feature, Korotkoff sound feature, and pressure feature is fused in chronological order to obtain a comprehensive feature sequence. Among them, the shock wave feature sequence is a time series, the Korotkoff sound feature sequence is a time series, and the pressure inside the cuff is a time series.

[0053] like Figure 7 As shown in the figure, the comprehensive feature sequence is input into the prediction model to obtain systolic blood pressure, diastolic blood pressure, and pulse rate. The prediction model is composed of an LSTM neural network followed by a fully connected layer. The LSTM neural network consists of multiple LSTM units. The objective function of the training prediction model is

[0054] Among them, L tot (X,Y 1:K ) is the overall objective function, which is the sum of three sub-objective functions. K is 3, and the sub-objective functions are the systolic blood pressure objective function L1(X, Y1), the diastolic blood pressure objective function L2(X, Y2), and the pulse rate objective function L3(X, Y3). X is the comprehensive feature of the input prediction model, Y1 is the true value of systolic blood pressure, Y2 is the true value of diastolic blood pressure, Y3 is the true value of pulse rate, and w i (t) is the weight that changes over time,

[0055] , N is 3, L n is the sub-objective function, r nis the ratio of the objective function, and T is an adjustable parameter. Multi-task learning is to train multiple related tasks together, so that different tasks can complement and promote each other, thereby achieving better results on a single task (accuracy, robustness, etc.). However, multi-task learning is not as simple as piling up all tasks. How to balance the training of each task so that each task can achieve beneficial improvements as much as possible is still a topic worth studying. From the perspective of loss function, multi-task learning is to have multiple loss functions L1, L2, ..., L n , generally they have a large number of shared parameters and a small number of independent parameters, and our goal is to make each loss function as small as possible.

[0056] like Figure 8 As shown, a blood pressure measurement device includes a pressure sensor device, a Korotkoff sound sensor device, a processor, a driving circuit for an air pump valve, and a power supply circuit. The processor is connected to the pressure sensor device, the Korotkoff sound sensor device, and the driving circuit for the air pump valve. The power supply circuit supplies power to the pressure sensor device, the Korotkoff sound sensor device, and the driving circuit for the air pump valve. The pressure sensor device monitors the pressure and shock waves in the cuff. The Korotkoff sound sensor device monitors the Korotkoff sounds. The driving circuit for the air pump valve drives the air pump valve to inflate and deflate the cuff. The processor executes an analysis algorithm.

[0057] As an embodiment, a blood pressure measuring device further includes a shell, a wrist cuff and a through tube connecting the shell and the wrist cuff, an air bag is provided in the wrist cuff, a pressure sensor device, a Korotkoff sound sensor device, a processor, a driving circuit of an air pump valve and a power supply circuit are provided in the shell, the through tube is connected to a connector on the side of the shell, the connector is connected to the driving air pump, a display screen, an adjustment button and a switch button are provided on the surface of the shell, the display screen and the adjustment button are electrically connected to the processor respectively, and the switch button is electrically connected to the power circuit.

[0058] The above are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A blood pressure measurement method, characterized in that: The following steps are involved: S1 obtains the pressure and shock wave in the cuff through the pressure sensor device and obtains Korotkoff sound through the Korotkoff sound sensor device; S2 performs feature extraction on the shock wave in the cuff to obtain shock wave features; Perform feature extraction on Korotkoff sounds to obtain Korotkoff sound features; The method for extracting shock wave characteristics includes the following steps: (1) Divide the preprocessed shock wave into multiple time segments, calculate the recursive graph for each time segment to obtain the recursive graph sequence. The recursive graph calculation formula is R i，j (ε)=H(ε-||y i -y j ||), where ε is the distance threshold, H(·) is the Heaviside function, yi, y j The calculation formula is y(θ,δ)={x(θ),x(θ+δ),x(θ+2δ),…,x(θ+(d-1)δ)}, where x(θ) is the oscillation wave, δ is the phase delay parameter, and d is the dimension parameter; (2) The recursive graph sequence is sequentially passed through the convolutional neural network to obtain the shock wave feature sequence, wherein the convolutional neural network is structurally divided into a convolutional layer, a ReLU layer, a Max-pooling layer, a Drop-out layer, a convolutional layer, a ReLU layer, a Max-pooling layer, and a Drop-out layer. The parameters in each layer can be customized, and the number of convolutional layers can also be customized. The method for extracting Korotkoff sound features comprises the following steps: (1) Divide the preprocessed Korotkoff sounds into multiple time segments, calculate the recursive graph for each time segment to obtain the recursive graph sequence. The recursive graph calculation formula is R i，j (ε)=H(ε-||y i -y j ||), where ε is the distance threshold, H(·) is the Heaviside function, and y i 、y j The calculation formula is y(θ, δ) = {x(θ), x(θ+δ), x(θ+2δ), …, x(θ+(d-1)δ)}, where x(θ) is the Korotkoff sound, δ is the phase delay parameter, and d is the dimension parameter; (2) The recursive graph sequence is sequentially passed through a convolutional neural network to obtain a Korotkoff sound feature sequence, wherein the convolutional neural network is structurally divided into a convolutional layer, a ReLU layer, a Max-pooling layer, a Drop-out layer, a convolutional layer, a ReLU layer, a Max-pooling layer, and a Drop-out layer. The parameters in each layer can be customized, and the number of convolutional layers can also be customized. S3 combines shock wave characteristics, Korotkoff sound characteristics and pressure in the cuff to obtain comprehensive characteristics; S4 inputs the comprehensive features into the prediction model to obtain systolic blood pressure, diastolic blood pressure and pulse rate.

2. A blood pressure measurement method according to claim 1, characterized in that: In S1, preprocessing is also required for the acquired pressure, shock waves in the cuff, and Korotkoff sounds.

3. A blood pressure measurement method according to claim 2, characterized in that: The pretreatment method comprises the following steps: (1) Use a noise reduction algorithm to reduce the pressure, shock waves, and Korotkoff sounds in the cuff; (2) Normalize the pressure inside the cuff, the shock wave inside the cuff, and the Korotkoff sound.

4. A blood pressure measurement method according to claim 1, characterized in that: In S3, the shock wave feature sequence, the Korotkoff sound feature sequence, and the pressure in the cuff are fused in time order to obtain a comprehensive feature sequence, where the shock wave feature sequence is a time series, the Korotkoff sound feature sequence is a time series, and the pressure in the cuff is a time series.

5. A blood pressure measurement method according to claim 1, characterized in that: In S4, the prediction model is composed of an LSTM neural network followed by a fully connected layer. The LSTM neural network consists of multiple LSTM units. The objective function of training the prediction model is Among them, L tot (X,Y 1:K ) is the overall objective function, which is the sum of three sub-objective functions. K is 3. The sub-objective functions are the systolic blood pressure objective function L1(X, Y1), the diastolic blood pressure objective function L2(X, Y2) and the pulse rate objective function L3(X, Y3). X is the comprehensive feature of the input prediction model, Y1 is the true value of systolic blood pressure, Y2 is the true value of diastolic blood pressure, Y3 is the true value of pulse rate, and w i (t) is the weight that changes over time, N is 3, L n is the sub-objective function, r n is the ratio of the objective function, and T is an adjustable parameter.

6. A blood pressure measuring device, based on a blood pressure measuring method according to any one of claims 1 to 5, characterized in that: The cuff comprises a pressure sensing device, a Korotkoff sound sensing device, a processor, a driving circuit for an air pump valve, and a power supply circuit. The processor is respectively connected to the pressure sensing device, the Korotkoff sound sensing device, and the driving circuit for the air pump valve. The power supply circuit supplies power to the pressure sensing device, the Korotkoff sound sensing device, and the driving circuit for the air pump valve. The pressure sensing device monitors the pressure in the cuff and the shock wave in the cuff. The Korotkoff sound sensing device monitors the Korotkoff sounds. The driving circuit for the air pump valve drives the air pump valve to inflate and deflate the cuff. The processor executes an analysis algorithm.

7. A blood pressure measuring device according to claim 6, characterized in that: It also includes a shell, a wrist cuff and a through tube connecting the shell and the wrist cuff. An air bag is provided in the wrist cuff. A pressure sensor, a Korotkoff sound sensor, a processor, a driving circuit for an air pump valve and a power supply circuit are provided in the shell. The through tube is connected to a connector on the side of the shell, and the connector is connected to a driving air pump.

8. The blood pressure measuring device according to claim 7, characterized in that: The shell surface is provided with a display screen, an adjustment button and a switch button. The display screen and the adjustment button are electrically connected to the processor respectively, and the switch button is electrically connected to the power circuit.

Images