阅读说明：本技术 基于鸿蒙操作系统的臂戴式血压仪 (Arm-wearing type blood pressure meter based on hongmeng operating system ) 是由 陈颖 刘鹤宁 于 2021-08-20 设计创作，主要内容包括：本发明提供一种基于鸿蒙操作系统的臂戴式血压仪,属于人工智能技术领域。包括：测量模块,包括袖带以及设置于袖带上的气囊、充气单元、放气单元、压力检测电路、柯氏音检测电路、脉搏波检测电路、第一处理器和第一无线通信单元；基于鸿蒙操作系统的移动智能设备,包括输入单元、第二处理器、第二无线通信单元；第一处理器根据第二处理器的控制信号,控制压力检测电路、柯氏音检测电路、充气单元和/或放气单元的运行；输入单元接收用户的控制输入,第二处理器根据控制输入生成控制信号；第二处理器利用深度学习模型、根据袖带压力信息、脉搏波信息和柯氏音信息确定动脉搏动过程中连续变化的血压。本发明检测精度高且能够获取实时连续的血压。(The invention provides an arm-worn blood pressure monitor based on a Hongmon operating system, and belongs to the technical field of artificial intelligence. The method comprises the following steps: the measurement module comprises a cuff, an air bag, an inflation unit, an deflation unit, a pressure detection circuit, a Korotkoff sound detection circuit, a pulse wave detection circuit, a first processor and a first wireless communication unit, wherein the air bag, the inflation unit, the deflation unit, the pressure detection circuit, the Korotkoff sound detection circuit, the pulse wave detection circuit, the first processor and the first wireless communication unit are arranged on the cuff; the mobile intelligent device based on the Hongmon operating system comprises an input unit, a second processor and a second wireless communication unit; the first processor controls the operation of the pressure detection circuit, the Korotkoff sound detection circuit, the inflation unit and/or the deflation unit according to the control signal of the second processor; the input unit receives control input of a user, and the second processor generates a control signal according to the control input; the second processor determines continuously changing blood pressure during the arterial pulse according to the cuff pressure information, the pulse wave information and the Korotkoff sound information by using the deep learning model. The invention has high detection precision and can acquire real-time continuous blood pressure.)

1. An arm-worn blood pressure monitor based on hongmeng operating system, comprising:

the system comprises a measurement module and a mobile intelligent device based on a Hongmon operating system;

the measurement module comprises a cuff, and an air bag, an inflation unit, an deflation unit, a pressure detection circuit, a Korotkoff sound detection circuit, a pulse wave detection circuit, a first processor and a first wireless communication unit which are arranged on the cuff; the mobile intelligent equipment comprises an input unit, a second processor, a second wireless communication unit and a display unit;

the inflation unit and the deflation unit are used for inflating and deflating the air bag in the blood pressure detection process according to the control of the first processor; the pressure detection circuit is used for detecting cuff pressure information, the korotkoff sound detection circuit is used for detecting korotkoff sound information, the pulse wave detection circuit is used for detecting pulse wave information, and the first processor sends the cuff pressure information, the pulse wave information and the korotkoff sound information to the second wireless communication unit through the first wireless communication unit; the first processor also controls the operation of the pressure detection circuit, the Korotkoff sound detection circuit, the inflation unit and/or the deflation unit according to control signals transmitted by the second processor through the second wireless communication unit and the first wireless communication unit;

the input unit is used for receiving control input of a user, the second processor is used for generating the control signal according to the control input and sending the control signal to the first wireless communication unit through the second wireless communication unit, and the control input at least comprises starting and stopping; the second processor is further used for determining the continuously changed blood pressure in the artery pulse process according to the cuff pressure information, the pulse wave information and the Korotkoff sound information received by the second wireless communication unit by using a deep learning model and displaying a continuous blood pressure waveform in the display unit.

2. The blood pressure monitor of claim 1, wherein the second processor determines a time interval between each korotkoff sound and a nearest pulse wave rise start point based on the korotkoff sound information and the pulse wave information; inputting the pulse wave information, the Korotkoff sound information and the cuff pressure information into a first model branch of the deep learning model, inputting the output of the first model branch and the time interval into a second model branch of the deep learning model, and determining the continuously changing blood pressure in the artery beating process according to the output of the second model branch.

3. The sphygmomanometer according to claim 2, wherein the first model branch includes a first neural network combination and a second neural network combination, an input of the first neural network combination is the pulse wave information, an input of the second neural network combination is the korotkoff sound information and the cuff pressure information, and outputs of the first neural network combination and the second neural network combination are weighted and summed to obtain an output of the first model branch.

4. The blood pressure monitor according to claim 3, wherein the pulse wave information inputted to the first neural network combination is pulse wave information subjected to segmentation processing and preprocessing;

the segmentation processing adopts a sliding overlapping sampling strategy to obtain a pulse wave information segment;

the preprocessing comprises the steps of carrying out continuous wavelet decomposition, denoising and demodulation processing on the pulse wave information segment to obtain a wavelet time-frequency graph;

the first neural network combination comprises a convolutional neural network and a cyclic neural network, and the convolutional neural network comprises a first convolutional layer, a first pooling layer, a second convolutional layer and a second pooling layer which are sequentially connected; the recurrent neural network comprises a gated recurrent layer, a fully connected layer and an identification layer.

5. The blood pressure monitor of claim 4, wherein the weight sequence loss function of the first neural network combination is:

wherein, x'_j＝x_j-max(x₁，…，x_m) M is the number of training data, W_jIs the weight of the weight loss function, x_jIs a wavelet time-frequency diagram matrix of the jth training data.

6. The sphygmomanometer according to claim 3, wherein the first neural network combination comprises a depth noise reduction automatic encoder and a support vector machine, the depth noise reduction automatic encoder comprises a plurality of layers of automatic encoders, a hidden layer of a previous layer of automatic encoder serves as an input layer of a next layer of automatic encoder, and the value of each node of the hidden layer of each layer of automatic encoder is calculated by outputting each node value of the input layer to an excitation function through linear weighted connection summation;

the pulse wave information input to the first neural network combination is pulse wave information subjected to variation modal decomposition processing, and the pulse wave information is decomposed to different frequency bands through the variation modal decomposition processing to obtain modal components belonging to the different frequency bands.

7. The sphygmomanometer according to claim 3, wherein the korotkoff sound information and the cuff pressure information input to the second neural network combination are a plurality of korotkoff sound information pieces each including 1/n korotkoff sounds, and a cuff pressure information piece in accordance with a time period thereof, n being a positive integer;

the second neural network combination comprises a deep belief neural network and a Light GBM network; the Light GBM network determines a corresponding blood pressure value according to the characteristic information of the Korotkoff sound information segment and the cuff pressure information segment;

the construction method of the deep belief neural network comprises the following steps: stacking d limited Boltzmann machines, constructing a vector S for a Korotkoff sound information section corresponding to one blood pressure value, using the vector S as a presentation layer neuron of the first limited Boltzmann machine, coding the vector S by an unsupervised learning method, and outputting a vector Q1 as a first characteristic extraction result of the Korotkoff sound information section to form a hidden layer neuron; training the deep confidence neural network by adopting a large amount of unlabeled Korotkoff sound information section training set data through a Gibbs sampling and contrast divergence algorithm to obtain a weight matrix W1, taking Q1 as the input of the next limited Boltzmann machine, repeating the previous step, and performing secondary feature extraction on the Korotkoff sound information section; and d limited boltzmann machines are used for obtaining a final monitoring signal feature extraction result and the weight W of the depth confidence network, wherein the weight W is { W1, W2 and … … Wd }.

8. The blood pressure monitor of claim 2, wherein the second model branch comprises an inclusion-v 3 neural network and an extreme learning machine; the Incep-v 3 neural network determines the systolic pressure and the diastolic pressure of the detected object according to the cuff pressure value and the Korotkoff sound information; and the extreme learning machine corrects the blood pressure value output by the first model branch according to the time interval, the systolic pressure and the diastolic pressure.

9. A blood pressure monitor according to claim 2, wherein the output of the second model branch is a discrete blood pressure value; the second processor predicts missing blood pressure values between the discrete blood pressure values using a residual network and a prediction filter;

the input of the residual error network is the discrete blood pressure value, and the output is the coefficient of the prediction filter;

the prediction filter predicts by the following formula:

wherein P (t) is the blood pressure value at time t, P' (t + t)₀) To predicted (t + t)₀) Blood pressure value at time, A (t)₀，t+t₀) Is (t + t)₀) T th corresponding to time₀Coefficient of, t₀-r, -r +1, …, r, r being the radius of the prediction filter.

10. A blood pressure monitor according to claim 2, further comprising a heart rate detection unit and/or a respiration detection unit;

the input of the second model branch further comprises first heart rate information and second heart rate information, and/or first breathing frequency and second breathing frequency;

the first heart rate information is heart rate information of a detected object in a calm state, and the second heart rate information is heart rate information detected by the heart rate detection unit in a detection process;

the first respiratory frequency is the respiratory frequency of a detected object in a calm state, and the second respiratory frequency is the respiratory frequency information detected by the respiratory detection unit in the detection process.

Technical Field

The invention relates to the technical field of artificial intelligence, in particular to an arm-worn blood pressure monitor based on a Hongmon operating system.

Background

The measurement principles of the currently used blood pressure meters mainly include the following two types: the Korotkoff sound method and the proportional coefficient method of cuff oscillation wave (also called oscillography or oscillation method) both require a cuff air bag to be wound on the arm for inflating and pressurizing, the pressure is added above the systolic pressure, and then the blood pressure is judged in the process of deflation and pressure relief. The korotkoff sound method comprises an artificial korotkoff sound method and an electronic korotkoff sound method, the artificial korotkoff sound method requires experienced medical staff to adopt a stethoscope, a mercury manometer and a cuff, an inflation/deflation bag is used for binding the cuff at a proper position on the upper arm of a subject, the stethoscope is close to the brachial artery, the inflation/deflation bag is used for inflating the cuff to increase pressure until the blood flow of the arm is blocked, then the cuff pressure is gradually reduced through the inflation/deflation bag to recover the blood flow of the arm, the arterial blood flow pulsation of the arm can generate a korotkoff sound change from small to large and then from large to small in the process, and the change of the korotkoff sound can be heard by means of the stethoscope and the mercury manometer to determine the systolic pressure and the diastolic pressure. The basic principle of the electronic Korotkoff sound method is to replace manual work with electronic technology, for example, inflating and deflating the cuff with an air pump, an electrically controlled valve, etc., and listening with an electronic sound pick-up and a processor.

The basic process of the proportional coefficient method of the cuff oscillation wave is very similar to that of the Korotkoff sound method, and the cuff is inflated to increase the pressure so as to block the blood flow of the arm, then gradually deflating the cuff to reduce the pressure to restore the blood flow of the arm, and monitoring the static pressure in the cuff and the pressure pulse wave generated by the pulsation of the arterial blood, but the calculation method is that the pressure pulse wave generated in the cuff and the corresponding cuff pressure are transmitted by detecting the arterial blood flow pulsation change of the arm in the deflation process, can detect a group of pressure pulse waves with the amplitude from small to large and then the pressure pulse waves with the amplitude from small to large and the corresponding cuff pressure from large to small, and the cuff pressure corresponding to the maximum value of the pressure pulse wave is taken as the average pressure, and the systolic pressure and the diastolic pressure are calculated according to the amplitude proportional coefficient of the pressure pulse wave with an empirical value.

However, the artificial korotkoff sound method has high requirements on detection personnel, otherwise, the artificial korotkoff sound method is likely to cause large errors and is not suitable for household daily detection, the electronic korotkoff sound method is easily influenced by other external sounds, and the pulse intensity of different people also has certain influence on the measurement result. In contrast, in the proportional coefficient method of the cuff oscillation wave, since the proportional coefficient is generally an empirical value (obtained through a large number of experiments), an error due to individual difference occurs. In addition, the systolic pressure and the diastolic pressure measured by the korotkoff sound method are blood pressures at two moments in the measurement process, the proportional coefficient method of the cuff oscillation wave is to obtain the systolic pressure and the diastolic pressure based on the average blood pressure in the measurement process, and both methods cannot obtain real-time continuous blood pressure.

Disclosure of Invention

Therefore, the technical problem to be solved by the embodiments of the present invention is to overcome the defects that the blood pressure detection method in the prior art has limited accuracy and cannot acquire real-time continuous blood pressure, thereby providing a boom-worn blood pressure monitor based on the hongmeng operating system.

Therefore, the invention provides an arm-worn blood pressure monitor based on a hongmeng operating system, which comprises:

the system comprises a measurement module and a mobile intelligent device based on a Hongmon operating system;

the measurement module comprises a cuff, and an air bag, an inflation unit, an deflation unit, a pressure detection circuit, a Korotkoff sound detection circuit, a pulse wave detection circuit, a first processor and a first wireless communication unit which are arranged on the cuff; the mobile intelligent equipment comprises an input unit, a second processor, a second wireless communication unit and a display unit;

the inflation unit and the deflation unit are used for inflating and deflating the air bag in the blood pressure detection process according to the control of the first processor; the pressure detection circuit is used for detecting cuff pressure information, the korotkoff sound detection circuit is used for detecting korotkoff sound information, the pulse wave detection circuit is used for detecting pulse wave information, and the first processor sends the cuff pressure information, the pulse wave information and the korotkoff sound information to the second wireless communication unit through the first wireless communication unit; the first processor also controls the operation of the pressure detection circuit, the Korotkoff sound detection circuit, the inflation unit and/or the deflation unit according to control signals transmitted by the second processor through the second wireless communication unit and the first wireless communication unit;

the input unit is used for receiving control input of a user, the second processor is used for generating the control signal according to the control input and sending the control signal to the first wireless communication unit through the second wireless communication unit, and the control input at least comprises starting and stopping; the second processor is further used for determining the continuously changed blood pressure in the artery pulse process according to the cuff pressure information, the pulse wave information and the Korotkoff sound information received by the second wireless communication unit by using a deep learning model and displaying a continuous blood pressure waveform in the display unit.

Optionally, the second processor determines a time interval between each korotkoff sound and a nearest pulse wave rising start point according to the korotkoff sound information and the pulse wave information; inputting the pulse wave information, the Korotkoff sound information and the cuff pressure information into a first model branch of the deep learning model, inputting the output of the first model branch and the time interval into a second model branch of the deep learning model, and determining the continuously changing blood pressure in the artery beating process according to the output of the second model branch.

Optionally, the first model branch includes a first neural network combination and a second neural network combination, the input of the first neural network combination is the pulse wave information, the input of the second neural network combination is the korotkoff sound information and the cuff pressure information, and the outputs of the first neural network combination and the second neural network combination are weighted and summed to obtain the output of the first model branch.

Optionally, the pulse wave information input to the first neural network combination is pulse wave information subjected to segmentation processing and preprocessing;

the segmentation processing adopts a sliding overlapping sampling strategy to obtain a pulse wave information segment;

the preprocessing comprises the steps of carrying out continuous wavelet decomposition, denoising and demodulation processing on the pulse wave information segment to obtain a wavelet time-frequency graph;

the first neural network combination comprises a convolutional neural network and a cyclic neural network, and the convolutional neural network comprises a first convolutional layer, a first pooling layer, a second convolutional layer and a second pooling layer which are sequentially connected; the recurrent neural network comprises a gated recurrent layer, a fully connected layer and an identification layer.

Optionally, the weight sequence loss function of the first neural network combination is:

wherein, x'_j＝x_j-max(x₁，…，x_m) M is the number of training data, W_jIs the weight of the weight loss function, x_jIs a wavelet time-frequency diagram matrix of the jth training data.

Optionally, the first neural network combination includes a depth noise reduction automatic encoder and a support vector machine, the depth noise reduction automatic encoder includes multiple layers of automatic encoders, a hidden layer of a previous layer of automatic encoder serves as an input layer of a next layer of automatic encoder, and values of nodes of a hidden layer of each layer of automatic encoder are calculated by summing values of nodes of the input layer through linear weighted connection and outputting to an excitation function;

the pulse wave information input to the first neural network combination is pulse wave information subjected to variation modal decomposition processing, and the pulse wave information is decomposed to different frequency bands through the variation modal decomposition processing to obtain modal components belonging to the different frequency bands.

Optionally, the korotkoff sound information and the cuff pressure information input to the second neural network combination are a plurality of korotkoff sound information segments and cuff pressure information segments consistent with time segments of the korotkoff sound information segments, each korotkoff sound information segment includes 1/n korotkoff sounds, and n is a positive integer;

the second neural network combination comprises a deep belief neural network and a Light GBM network; the Light GBM network determines a corresponding blood pressure value according to the characteristic information of the Korotkoff sound information segment and the cuff pressure information segment;

the construction method of the deep belief neural network comprises the following steps: the construction method of the deep belief neural network comprises the following steps: stacking d limited Boltzmann machines, constructing a vector S for a Korotkoff sound information section corresponding to one blood pressure value, using the vector S as a presentation layer neuron of the first limited Boltzmann machine, coding the vector S by an unsupervised learning method, and outputting a vector Q1 as a first characteristic extraction result of the Korotkoff sound information section to form a hidden layer neuron; training the deep confidence neural network by adopting a large amount of unlabeled Korotkoff sound information section training set data through a Gibbs sampling and contrast divergence algorithm to obtain a weight matrix W1, taking Q1 as the input of the next limited Boltzmann machine, repeating the previous step, and performing secondary feature extraction on the Korotkoff sound information section; and d limited boltzmann machines are used for obtaining a final monitoring signal feature extraction result and the weight W of the depth confidence network, wherein the weight W is { W1, W2 and … … Wd }.

Optionally, the second model branch comprises an inclusion-v 3 neural network and a limit learning machine; the Incep-v 3 neural network determines the systolic pressure and the diastolic pressure of the detected object according to the cuff pressure value and the Korotkoff sound information; and the extreme learning machine corrects the blood pressure value output by the first model branch according to the time interval, the systolic pressure and the diastolic pressure.

Optionally, the output of the second model branch is a discrete blood pressure value; the second processor predicts missing blood pressure values between the discrete blood pressure values using a residual network and a prediction filter;

the input of the residual error network is the discrete blood pressure value, and the output is the coefficient of the prediction filter;

the prediction filter predicts by the following formula:

wherein P (t) is the blood pressure value at time t, P' (t + t)₀) To predicted (t + t)₀) Blood pressure value at time, A (t)₀，t+t₀) Is (t + t)₀) T th corresponding to time₀Coefficient of, t₀-r, -r +1, …, r, r being the radius of the prediction filter.

Optionally, the blood pressure monitor further comprises a heart rate detection unit and/or a respiration detection unit;

the input of the second model branch further comprises first heart rate information and second heart rate information, and/or first breathing frequency and second breathing frequency;

the first heart rate information is heart rate information of a detected object in a calm state, and the second heart rate information is heart rate information detected by the heart rate detection unit in a detection process;

the first respiratory frequency is the respiratory frequency of a detected object in a calm state, and the second respiratory frequency is the respiratory frequency information detected by the respiratory detection unit in the detection process.

The technical scheme of the embodiment of the invention has the following advantages:

the arm-worn blood pressure monitor based on the Hongmon operating system provided by the embodiment of the invention determines the blood pressure of a detection object according to the pulse wave information and the Korotkoff sound information, thereby improving the detection precision. Moreover, the blood pressure meter can continuously and uninterruptedly measure the blood pressure, can provide continuous arterial pressure wave in the process of arterial pulsation, and can more accurately reflect the blood pressure condition of a detection object. In addition, the measurement module and the mobile intelligent device are in wireless communication connection, so that the mobile intelligent terminal can be independent, for example, the mobile intelligent terminal can be realized by a smart phone or a tablet computer in combination with a corresponding application program (APP), and the purchase cost of a user is reduced.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.

Fig. 1 is a schematic structural diagram of a specific example of an arm-worn blood pressure monitor based on a hongmeng operating system in the embodiment of the present invention;

FIG. 2 is a functional block diagram of a specific example of a measurement module in an embodiment of the present invention;

fig. 3 is a schematic block diagram of a specific example of a mobile intelligent device in the embodiment of the present invention.

Detailed Description

The technical solutions of the present invention will be described clearly and completely with reference to the accompanying drawings, and it should be understood that the described embodiments are some, but not all embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

In describing the present invention, it is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. The terms "comprises" and/or "comprising," when used in this specification, are intended to specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof. The term "and/or" includes any and all combinations of one or more of the associated listed items. The terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," "outer," and the like are used in the orientation or positional relationship indicated in the drawings for convenience in describing the invention and for simplicity in description, and do not indicate or imply that the referenced device or element must have a particular orientation, be constructed and operated in a particular orientation, and are therefore not to be construed as limiting the invention. The terms "first," "second," and "third" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance. The terms "mounted," "connected," and "coupled" are to be construed broadly and may, for example, be fixedly coupled, detachably coupled, or integrally coupled; can be mechanically or electrically connected; the two elements can be directly connected, indirectly connected through an intermediate medium, or communicated with each other inside; either a wireless or a wired connection. The specific meanings of the above terms in the present invention can be understood in specific cases to those skilled in the art.

In addition, the technical features involved in the different embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

Example 1

The present embodiment provides an arm-worn blood pressure monitor based on the hongmeng operating system, as shown in fig. 1, fig. 2, and fig. 3, including: the system comprises a measurement module 1 and a mobile intelligent device 2 based on a Hongmon operating system;

the measurement module 1 comprises a cuff 11, and an air bag (not shown in the figure) arranged on the cuff 11, an inflation unit 12, a deflation unit 13, a pressure detection circuit 14, a Korotkoff sound detection circuit 15, a pulse wave detection circuit 16, a first processor 17 and a first wireless communication unit 18; the mobile intelligent device 2 comprises an input unit 21, a second processor 22, a second wireless communication unit 23 and a display unit 24;

the specific layout of the inflation unit 12, the deflation unit 13, the pressure detection circuit 14, the korotkoff sound detection circuit 15, the pulse wave detection circuit 16, the first processor 17 and the first wireless communication unit 18 in the cuff 11 can be determined according to the detection requirement and the factors such as the size and shape of the device, which are not specifically shown in the figure.

The inflation unit 12 and the deflation unit 13 are used for inflating and deflating the air bag in the blood pressure detection process according to the control of the first processor 17; the pressure detection circuit 14 is configured to detect cuff pressure information, the korotkoff sound detection circuit 15 is configured to detect korotkoff sound information, the pulse wave detection circuit 16 is configured to detect pulse wave information, and the first processor 17 sends the cuff pressure information, the pulse wave information, and the korotkoff sound information to the second wireless communication unit 23 through the first wireless communication unit 18; the first processor 17 further controls the operation of the pressure detection circuit 14, the korotkoff sound detection circuit 15, the inflation unit 12 and/or the deflation unit 13 according to the control signals transmitted from the second processor 22 through the second wireless communication unit 23 and the first wireless communication unit 18;

the input unit 21 is configured to receive a control input of a user, the second processor 22 is configured to generate the control signal according to the control input and send the control signal to the first wireless communication unit 18 through the second wireless communication unit 23, and the control input at least includes starting and stopping; the second processor 22 is further configured to determine a continuously changing blood pressure during the arterial pulse according to the cuff pressure information, the pulse wave information and the korotkoff sound information received by the second wireless communication unit 23 by using a deep learning model and display a continuous blood pressure waveform in the display unit 24.

That is, the second processor 22 can obtain the blood pressure that changes in real time during the detection process, and not only can display the blood pressure at a certain time or several times during the detection process.

Specifically, the first wireless communication unit 18 and the second wireless communication unit 23 may be near field communication units, that is, the first wireless communication unit 18 and the second wireless communication unit 23 may implement near field communication. The detection module also includes a power source, such as a battery or the like. The input unit 21 may include a key and/or a touch screen. The touch screen may also serve as the display unit 24. The inflation unit 12 comprises an electric inflator. The pressure sensing circuit 14 may comprise a capacitive pressure sensor, which may be located within the bladder or at a communication port of the bladder, and functions to convert changes in air pressure into changes in the capacitance of the capacitor. The korotkoff sound detection circuit 15 may include a microphone.

The arm-worn blood pressure monitor based on the Hongmon operating system determines the blood pressure of a detection object according to pulse wave information and Korotkoff sound information, and improves detection precision. Moreover, the blood pressure meter can continuously and uninterruptedly measure the blood pressure, can provide continuous arterial pressure wave in the process of arterial pulsation, and can more accurately reflect the blood pressure condition of a detection object. In addition, because the measurement module 1 and the mobile intelligent device 2 are in wireless communication connection, the mobile intelligent terminal can be independent, for example, the mobile intelligent terminal can be realized by a smart phone or a tablet computer in combination with a corresponding application program (APP), so that the purchase cost of a user is reduced.

Optionally, the second processor 22 determines a time interval between each korotkoff sound and the nearest pulse wave rising start point according to the korotkoff sound information and the pulse wave information; inputting the pulse wave information, the Korotkoff sound information and the cuff pressure information into a first model branch of the deep learning model, inputting the output of the first model branch and the time interval into a second model branch of the deep learning model, and determining the continuously changing blood pressure in the artery beating process according to the output of the second model branch.

Specifically, in the detection of one deflation, the time interval between each korotkoff sound and the nearest rising start point of the pulse wave is different, the time interval corresponding to the korotkoff sound appearing for the first time is the longest, and the time interval corresponding to the korotkoff sound appearing later is shorter and shorter.

Both the blood pressure measurement method based on Korotkoff sounds and the blood pressure measurement method based on pulse waves need multiple arterial pulsation across domains, and continuous blood pressure in each pulsation process cannot be acquired. In this embodiment, a plurality of discrete blood pressure values in the multi-beat process are estimated based on the pulse wave information and the korotkoff sound information, and are corrected based on the time interval, the cuff pressure value and the korotkoff sound information, and finally, the second processor 22 obtains a continuous blood pressure value according to the discrete blood pressure values.

Optionally, the first model branch includes a first neural network combination and a second neural network combination, the input of the first neural network combination is the pulse wave information, the input of the second neural network combination is the korotkoff sound information and the cuff pressure information, and the outputs of the first neural network combination and the second neural network combination are weighted and summed to obtain the output of the first model branch.

In this embodiment, the first neural network combination is configured to determine a blood pressure value varying in each beat based on the pulse wave information, and the second neural network combination is configured to determine a blood pressure value in each beat based on the korotkoff sound information and the cuff pressure information, and then perform weighted summation on outputs of the first neural network combination and the second neural network combination to obtain a blood pressure value varying in each beat.

Optionally, the pulse wave information input to the first neural network combination is pulse wave information subjected to segmentation processing and preprocessing;

the segmentation processing adopts a sliding overlapping sampling strategy to obtain a pulse wave information segment;

the preprocessing comprises the steps of carrying out continuous wavelet decomposition, denoising and demodulation processing on the pulse wave information segment to obtain a wavelet time-frequency graph;

the first neural network combination comprises a convolutional neural network and a cyclic neural network, and the convolutional neural network comprises a first convolutional layer, a first pooling layer, a second convolutional layer and a second pooling layer which are sequentially connected; the recurrent neural network comprises a gated recurrent layer, a fully connected layer and an identification layer.

Specifically, the first convolution layer performs convolution operation on the input, and the output value of the convolution kernel is c_i＝f(w·x_i+ a), where w is the convolution kernel weight, a is the offset, x_iIs a wavelet time-frequency graph vector matrix of the ith pulse wave information segment, and i represents the ith pulse wave information segment. In each feature vector output by the pooling layer to the convolutional layerA maximum value is extracted.

The gating cycle layer has two inputs, one is the characteristic information (extracted by the convolutional neural network) x corresponding to the current pulse wave information segment_tThe other is a hidden state h transmitted by the previous node_t-1This hidden state includes information about previous nodes, in combination with x_tAnd h_t-1Will get the output y of the current hidden node_tAnd a hidden state h passed to the next node_t. The gated loop layer has a reset gate (r-gate) and an update gate (z-gate).

The convolutional neural network is used for extracting the characteristics of the wavelet time-frequency graph of each pulse wave information segment, and the cyclic neural network is used for identifying the blood pressure value corresponding to each pulse wave information segment.

Optionally, the weight sequence loss function of the first neural network combination is:

wherein, x'_j＝x_j-max(x₁，…，x_m) M is the number of training data, W_jIs the weight of the weight loss function, x_jIs a wavelet time-frequency diagram matrix of the jth training data.

Optionally, the first neural network combination includes a depth noise reduction automatic encoder and a support vector machine, the depth noise reduction automatic encoder includes multiple layers of automatic encoders, a hidden layer of a previous layer of automatic encoder serves as an input layer of a next layer of automatic encoder, and values of nodes of a hidden layer of each layer of automatic encoder are calculated by summing values of nodes of the input layer through linear weighted connection and outputting to an excitation function;

the pulse wave information input to the first neural network combination is pulse wave information subjected to variation modal decomposition processing, and the pulse wave information is decomposed to different frequency bands through the variation modal decomposition processing to obtain modal components belonging to the different frequency bands.

The depth noise reduction auto-encoder may be trained by:

constructing a training sample (X, Y), wherein X is a section of modal component of different frequency bands of the pulse wave, and Y is a corresponding blood pressure value;

carrying out unsupervised greedy layer-by-layer training on the depth noise reduction automatic encoder, wherein the automatic encoders of the l-1 layer and the l-1 layer are trained through a contrast divergence algorithm;

carrying out supervised fine adjustment on the depth noise reduction automatic encoder: turning over the depth noise reduction automatic encoder subjected to unsupervised greedy training in the previous step to obtain an automatic encoder with doubled layer number; and training the automatic coding machine by using a BP algorithm to finely adjust the depth noise reduction automatic coder.

In the embodiment, the depth noise reduction automatic encoder can accurately extract the characteristic information in the pulse wave information to provide better data basis for the blood pressure identification of a subsequent support vector machine, reduce the identification difficulty and improve the identification precision.

Optionally, the korotkoff sound information and the cuff pressure information input to the second neural network combination are a plurality of korotkoff sound information segments and cuff pressure information segments consistent with time segments of the korotkoff sound information segments, each korotkoff sound information segment includes 1/n korotkoff sounds, and n is a positive integer;

the second neural network combination comprises a deep belief neural network and a Light GBM network; the Light GBM network determines a corresponding blood pressure value according to the characteristic information of the Korotkoff sound information segment and the cuff pressure information segment;

the construction method of the deep belief neural network comprises the following steps: stacking d limited Boltzmann machines, constructing a vector S for a Korotkoff sound information section corresponding to one blood pressure value, using the vector S as a presentation layer neuron of the first limited Boltzmann machine, coding the vector S by an unsupervised learning method, and outputting a vector Q1 as a first characteristic extraction result of the Korotkoff sound information section to form a hidden layer neuron; training the deep confidence neural network by adopting a large amount of unlabeled Korotkoff sound information section training set data through a Gibbs sampling and contrast divergence algorithm to obtain a weight matrix W1, taking Q1 as the input of the next limited Boltzmann machine, repeating the previous step, and performing secondary feature extraction on the Korotkoff sound information section; and d limited boltzmann machines are used for obtaining a final monitoring signal feature extraction result and the weight W of the depth confidence network, wherein the weight W is { W1, W2 and … … Wd }.

Specifically, the number of the limited boltzmann machines is determined according to the number order of the input korotkoff sound information segments and the number order of the output feature extraction results. The Light GBM model is a learning algorithm based on a decision tree, and has the advantages of higher training speed, higher accuracy and large data processing capacity.

In this embodiment, a real-time blood vessel pressure value is identified according to the korotkoff sound information and the cuff pressure information, where the blood vessel pressure identification is mainly based on the korotkoff sound information, but since korotkoff sounds are affected by the cuff pressure, that is, even if the cuff pressure is different, the korotkoff sounds have different blood vessel pressures, the blood vessel pressure value identified based on the korotkoff sound information is also corrected by using the cuff pressure information in this embodiment.

It should be noted here that, when performing weighted summation on the outputs of the first neural network combination and the second neural network combination, only the blood pressure values of the same time or the same time period (the time period intersection can be regarded as the same time period) are subjected to weighted summation, and if only one neural network combination at a certain time or time period has a blood pressure value corresponding to the output, the weighted summation processing is not required.

Optionally, the second model branch comprises an inclusion-v 3 neural network and a limit learning machine; the Incep-v 3 neural network determines the systolic pressure and the diastolic pressure of the detected object according to the cuff pressure value and the Korotkoff sound information; and the extreme learning machine corrects the blood pressure value output by the first model branch according to the time interval, the systolic pressure and the diastolic pressure.

In this embodiment, the input of the second model branch includes not only the output of the first model branch and the time interval, but also the cuff pressure value and the korotkoff sound information.

Specifically, the inclusion-v 3 neural network identifies a first time point corresponding to systolic pressure and a second time point corresponding to diastolic pressure according to the korotkoff sound information, and takes a pressure value corresponding to the first time point in the cuff pressure values as the systolic pressure, and takes a pressure value corresponding to the second time point in the cuff pressure values as the diastolic pressure.

Specifically, the second model branch corrects the blood pressure value at the time corresponding to the systolic pressure according to the systolic pressure, and the second model branch corrects the blood pressure value at the time corresponding to the diastolic pressure according to the diastolic pressure. For example, the blood pressure can be corrected by means of weighted summation, and the weight value can be obtained by training and learning.

The blood pressure value is determined according to the pulse wave information, which is influenced by the individual blood vessel condition, and the detection of the individual blood vessel condition requires professional detection instruments and some or even traumatic detection, which do not meet the daily blood pressure detection requirement.

In addition, before the korotkoff sound information is input into the inclusion-v 3 neural network, framing and windowing can be performed on the korotkoff sound information to obtain korotkoff sound information frames, the length of each korotkoff sound information frame is smaller than or equal to the duration of one korotkoff sound, then short-time fourier transform and the like are performed on the korotkoff sound information frames to obtain corresponding spectrogram data, and then the inclusion-v 3 neural network identifies whether the corresponding korotkoff sound information frame is a korotkoff sound information frame corresponding to systolic pressure or not and whether the corresponding korotkoff sound information frame is a korotkoff sound information frame corresponding to diastolic pressure or not based on the spectrogram data.

Optionally, the output of the second model branch is a discrete blood pressure value; the second processor 22 predicts the missing blood pressure values between the discrete blood pressure values using a residual network and a prediction filter;

the input of the residual error network is the discrete blood pressure value, and the output is the coefficient of the prediction filter;

the prediction filter predicts by the following formula:

wherein P (t) is the blood pressure value at time t, P' (t + t)₀) To predicted (t + t)₀) Blood pressure value at time, A (t)₀，t+t₀) Is (t + t)₀) T th corresponding to time₀Coefficient of, t₀-r, -r +1, …, r, r being the radius of the prediction filter.

In other alternative embodiments, the second processor 22 may also perform a curve fitting process, such as a polynomial fit, on the discrete blood pressure values.

Optionally, the blood pressure monitor further comprises a heart rate detection unit and/or a respiration detection unit;

the input of the second model branch further comprises first heart rate information and second heart rate information, and/or first breathing frequency and second breathing frequency;

the first heart rate information is heart rate information of a detected object in a calm state, and the second heart rate information is heart rate information detected by the heart rate detection unit in a detection process;

the first respiratory frequency is the respiratory frequency of a detected object in a calm state, and the second respiratory frequency is the respiratory frequency information detected by the respiratory detection unit in the detection process.

In the embodiment, the blood pressure value can be corrected based on the heart rate change and the respiratory frequency change of the detected object in the blood pressure detection process, so that the misjudgment of the hypertension caused by factors such as exercise and tension is reduced.

In other alternative embodiments, the first heart rate information and the second heart rate information, and/or the first breathing rate and the second breathing rate may not be used as input of the second model branch, that is, the blood pressure value is not corrected according to the heart rate change and the breathing rate change of the detected subject, but the second processor 22 outputs corresponding prompt information according to the first heart rate information and the second heart rate information, and/or the first breathing rate and the second breathing rate, and displays the prompt information while displaying the continuous blood pressure waveform, wherein the prompt information may indicate that the detected subject is not in a calm state currently, and the blood pressure value cannot be used as a diagnosis basis.

It should be understood that the above examples are only for clarity of illustration and are not intended to limit the embodiments. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. And obvious variations or modifications therefrom are within the scope of the invention.