Blood pressure measurement model training method and pulse blood pressure measurement system

Abstract

The invention provides a blood pressure measurement model training method and a pulse blood pressure measurement system, a pulse wave measurement module obtains a pulse wave signal, and pulse waves of the wrist of a user act on pressure of a pressurization module to obtain a pressure change signal; the signal denoising module is used for denoising the pulse wave signals and the pressure change signals and transmitting the signals to the blood pressure measurement module, and the blood pressure measurement module executes a blood pressure measurement model training method; an initial neural network comprising a time sequence prediction network, a feature extraction network and a decoding network is trained through a pulse wave signal data set and a pressure change signal data set, the time sequence prediction network and the feature extraction network output personal blood vessel features and high-dimensional pulse features respectively, and a blood pressure data prediction result is output through the decoding network. And minimizing the deviation between a blood pressure data prediction result and an actual result, performing parameter iteration updating on the initial neural network model to obtain a blood pressure measurement model, and obtaining blood pressure data through the blood pressure measurement model to realize continuous measurement of blood pressure.

Description

Technical Field

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

[0002] In today's society, people pay little attention to blood pressure management, the measurement process is cumbersome, and blood pressure quality management is insufficient. The general public usually does not measure or rarely measures their blood pressure, resulting in low awareness, treatment and control rates of hypertension. The overall prevalence of hypertension remains high and is rising year by year. With the continuous advancement of medical technology, the application of blood pressure in clinical monitoring has received more and more attention. Traditional intermittent blood pressure measurement can no longer meet the needs of accurately monitoring the dynamic changes of patients' blood pressure during treatment. Dynamic blood pressure measurement has gradually become an important part of clinical practice because it can provide more detailed and accurate data, blood pressure trends and help discover potential blood pressure problems.

[0003] In current medical practice, dynamic blood pressure monitoring uses a pressurized measurement method. The cuff generates pressure in the artery by inflating, and then releases the pressure to measure blood pressure. This method applies excessive pressure and noise during the inflation process, which will interfere with the user's normal work and life, making it impossible to perform dynamic monitoring comfortably and affecting the user experience. In addition, the measurement results of some patients in hospital environments are not accurate enough. In order to be able to dynamically monitor the patient's blood pressure comfortably, relevant researchers have proposed a method for measuring blood pressure without feeling. Although the non-feeling measurement method improves the user's comfort, there are still some shortcomings. First, it cannot be measured continuously. During measurement, the patient still needs to maintain a certain posture within the specified range, and it is impossible to monitor when the patient moves; second, the measurement range is too small and the accuracy is too low. The measurement method using wireless signals will be affected by signal attenuation. The accuracy will drop significantly after exceeding a certain distance. Even within the operating range of the device, its measurement accuracy is still far behind that of medical blood pressure measurement devices. The accuracy of the measured accuracy that can meet the international standard error is low, and the results are relatively unreliable. It cannot replace medical equipment for the time being; how to continuously pay attention to the dynamic changes of blood pressure in a portable and comfortable way has become an important challenge for public health. Summary of the invention

[0004] In view of this, the embodiments of the present invention provide a blood pressure measurement model training method and a pulse blood pressure measurement system to eliminate or improve one or more defects existing in the prior art, and solve the problem that the prior art cannot simply and comfortably continuously measure dynamic blood pressure.

[0005] One aspect of the present invention provides a blood pressure measurement model training method, the method comprising the following steps:

[0006] Acquire a training sample set, the training sample set comprising a pulse wave signal data set and a pressure change signal data set obtained by the pulse wave of the user's wrist acting on the pressure applied to the user's wrist; add corresponding blood pressure measurement results obtained by full-time pressurization as labels to samples in the training sample set; the full-time pressurization means that a low-pressure signal and a high-pressure signal are applied to the user's wrist by using a blood pressure measurement device during the whole process of blood pressure measurement;

[0007] Acquire an initial neural network model, the initial neural network model includes a time series prediction network, a feature extraction network and a decoding network; the time series prediction network takes the pulse wave signal data set and the pressure change signal data set as input, outputs a predicted pulse wave signal and a predicted full-course pressure change signal of the blood pressure measurement, and obtains personal vascular characteristics according to a preset measurement model; the feature extraction network takes the pulse wave signal data set as input, outputs a high-dimensional pulse feature; the personal vascular characteristics and the high-dimensional pulse features are fused and input into a decoding network, the decoding network learns the relationship between the personal vascular characteristics and the high-dimensional pulse characteristics and the blood pressure data, and outputs the blood pressure data prediction result of the user;

[0008] The parameters of the initial neural network model are iteratively updated by minimizing the deviation between the predicted results and the actual results of the blood pressure data, thereby obtaining a blood pressure measurement model.

[0009] In some embodiments, the timing prediction network includes a serial block layer, a first attention and normalization layer, a first feedforward fully connected and normalized layer, and a fully connected mapping layer; a residual short connection is established between the input of the feedforward fully connected and normalized layer and the input of the fully connected mapping layer.

[0010] In some embodiments, the feature extraction network includes a convolutional layer, a second attention and normalization layer, a second feedforward fully connected and normalized layer, a third attention and normalization layer, a third feedforward fully connected and normalized layer, a fourth attention and normalization layer, a fourth feedforward fully connected and normalized layer and a category label extraction layer arranged in succession; a residual short connection is established between the input and output of the second attention and normalization layer; a residual short connection is established between the input and output of the second feedforward fully connected and normalized layer; a residual short connection is established between the input and output of the third attention and normalization layer; a residual short connection is established between the input and output of the third feedforward fully connected and normalized layer; a residual short connection is established between the input and output of the fourth attention and normalization layer; a residual short connection is established between the input and output of the fourth feedforward fully connected and normalized layer.

[0011] In some embodiments, the decoding network includes a first fully connected layer, a first non-linear activation function layer, a second fully connected layer, a second non-linear activation function layer and a third fully connected layer that are arranged in succession.

[0012] In some embodiments, the decoding network adopts a convolutional neural network, a multi-layer perceptron or a deep learning model.

[0013] In some embodiments, fusing the personal blood vessel feature with the high-dimensional pulse feature comprises:

[0014] Mapping the personal blood vessel features and the high-dimensional pulse features to the same feature space through a preset mapping function; the personal blood vessel features include blood vessel diameter, blood vessel wall thickness, blood flow velocity and blood vessel compliance parameters; the high-dimensional pulse features include pulse wave waveform features, pulse wave conduction features, pulse wave frequency features, pulse wave dynamic features and pulse wave pulse features;

[0015] The personal blood vessel features and the high-dimensional pulse features mapped to the same feature space are weighted according to feature importance and directly concatenated to obtain fused features.

[0016] In some embodiments, the method further includes: the pulse wave signal data set and the pressure change signal data set are input into a timing prediction network to obtain the predicted pulse wave signal and the full-range pressure change signal, the predicted pulse wave signal responds to the full-range pressure change signal and obtains a pulse wave change signal, and personal vascular characteristics are obtained according to a preset measurement model.

[0017] On the other hand, the present invention also provides a pulse blood pressure measurement system, which is deployed on a wrap-type device worn on a wrist, and the system comprises:

[0018] A pulse wave measurement module, including a PPG sensor component and a filter component, is used to collect the PPG signal of the user's wrist and obtain a pulse wave signal after filtering through the filter component;

[0019] A pressurizing module, including a micro air pump assembly and a high-precision air pressure sensor, for controlling the pressure applied to the wrist of the user, and detecting a pressure change signal obtained by the pulse wave signal acting on the pressure through the high-precision air pressure sensor;

[0020] A signal denoising module receives the pulse wave signal and the pressure change signal and performs denoising processing through a microcontroller;

[0021] A signal transmission module, including a Bluetooth component and a wireless transmission component, for transmitting the pulse wave signal and the pressure change signal after denoising to a blood pressure measurement module;

[0022] A blood pressure measurement module, the module is used to execute any one of the above-mentioned blood pressure measurement model training methods to obtain a blood pressure measurement model, and obtain blood pressure data through the pulse wave signal and the pressure change signal;

[0023] A storage module is used to store the blood pressure data, the pulse wave signal after denoising, and the pressure change signal.

[0024] In some embodiments, collecting the PPG signal of the user includes:

[0025] The light emitting diode deployed in the PPG sensor emits light of a specific wavelength toward the skin of the user's wrist; the light of the specific wavelength can penetrate the skin;

[0026] After the light penetrates the skin, a portion of the light is absorbed by arteries, veins and other tissues, and the light that is not absorbed is reflected to the photodiode deployed in the PPG sensor;

[0027] When the amount of blood in the artery increases, the light intensity detected by the photodiode decreases; when the amount of blood in the artery decreases, the light intensity detected by the photodiode increases;

[0028] The photodiode converts the change in light intensity into a PPG signal.

[0029] In some embodiments, performing denoising processing by the microcontroller includes reducing noise in the pulse wave signal and the pressure change signal by beamforming, wavelet transform and variational mode decomposition filtering.

[0030] The beneficial effects of the present invention are at least:

[0031] In the blood pressure measurement model training method and pulse blood pressure measurement system of the present invention, the pulse wave measurement module obtains a pulse wave signal, the pulse wave of the user's wrist acts on the pressure of the pressurizing module to obtain a pressure change signal, the pressurizing module accurately controls the pressure applied to the user's wrist so that the pressure level does not cause discomfort to the user, the blood pressure measurement module executes the blood pressure measurement model training method, trains an initial neural network including a timing prediction network, a feature extraction network and a decoding network through a pulse wave signal data set and a pressure change signal data set, the timing prediction network and the feature extraction network output personal blood vessel characteristics and high-dimensional pulse characteristics respectively, obtains a predicted pulse wave signal and a full-process pressure change signal through the timing prediction network, reduces the discomfort and damage caused by long-term pressure on the wrist during blood pressure measurement, and is suitable for all-weather blood pressure measurement; after fusion, the decoding network outputs the blood pressure data prediction result, minimizes the deviation between the blood pressure data prediction result and the actual result, and performs parameter iterative update on the initial neural network model to obtain a blood pressure measurement model; the fusion of personal blood vessel characteristics and high-dimensional pulse characteristics can achieve accurate and robust blood pressure measurement and obtain blood pressure data.

[0032] Furthermore, the present invention denoises the pulse wave signal and the pressure change signal, reduces the noise in the pulse wave signal and the pressure change signal through beamforming, wavelet transform and variational mode decomposition filtering, improves the clarity and accuracy of the signal, and provides a high-quality input signal for blood pressure measurement.

[0033] Additional advantages, purposes, and features of the present invention will be described in part in the following description, and will become apparent to those skilled in the art after studying the following, or may be learned from the practice of the present invention. The purposes and other advantages of the present invention may be achieved and obtained by the structures specifically indicated in the specification and the accompanying drawings.

[0034] Those skilled in the art will appreciate that the objectives and advantages that can be achieved with the present invention are not limited to the above specific description, and the above and other objectives that can be achieved by the present invention will be more clearly understood from the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The drawings described herein are used to provide a further understanding of the present invention, constitute a part of the present application, and do not constitute a limitation of the present invention. In the drawings:

[0036] Figure 1 The figure is a flow chart of a blood pressure measurement model training method according to an embodiment of the present invention.

[0037] Figure 2 It is a structural schematic diagram of a blood pressure measurement model according to an embodiment of the present invention.

[0038] Figure 3 FIG. 1 is a schematic diagram of the structure of a pulse blood pressure measurement system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments and the accompanying drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0040] It should also be noted that, in order to avoid obscuring the present invention due to unnecessary details, only structures and / or processing steps closely related to the solutions according to the present invention are shown in the accompanying drawings, while other details that are not closely related to the present invention are omitted.

[0041] It should be emphasized that the term “include / comprises” when used herein refers to the presence of features, elements, steps or components, but does not exclude the presence or addition of one or more other features, elements, steps or components.

[0042] It should also be noted that, unless otherwise specified, the term “connection” herein may refer not only to a direct connection but also to an indirect connection involving an intermediate.

[0043] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In the accompanying drawings, the same reference numerals represent the same or similar components, or the same or similar steps.

[0044] In the prior art, dynamic blood pressure monitoring adopts a pressurized measurement method, which generates pressure in the artery by inflating the cuff, and then releases the pressure to measure blood pressure. In this method, the excessive pressure and noise exerted by the device during the inflation process will interfere with the user's normal work and life, making it impossible to perform dynamic monitoring comfortably, affecting the user experience; the non-contact measurement method improves the user's comfort, but there are still disadvantages such as inability to measure continuously, too small a measurement range, and too low accuracy. It is impossible to monitor when the patient moves, and the accuracy will decrease with the increase of distance. Moreover, even within the operating range of the device, its measurement accuracy is still far behind that of medical blood pressure measurement devices. The accuracy of the measurement that can meet the international standard error is low, the result is relatively unreliable, and it cannot temporarily replace medical equipment; the present invention proposes a blood pressure measurement model A training method and a pulse blood pressure measurement system, wherein a pulse wave measurement module obtains a pulse wave signal, the pulse wave of the user's wrist acts on the pressure of a pressurizing module to obtain a pressure change signal, the pulse wave signal and the pressure change signal are denoised by a signal denoising module and transmitted to a blood pressure measurement module, the blood pressure measurement module executes a blood pressure measurement model training method, an initial neural network including a timing prediction network, a feature extraction network and a decoding network is trained by a pulse wave signal data set and a pressure change signal data set, the timing prediction network and the feature extraction network output personal vascular features and high-dimensional pulse features respectively, after fusion, the decoding network outputs the blood pressure data prediction result, the deviation between the blood pressure data prediction result and the actual result is minimized, the parameters of the initial neural network model are iteratively updated to obtain a blood pressure measurement model, and blood pressure data is obtained through the blood pressure measurement model.

[0045] Figure 1 The figure is a flow chart of a blood pressure measurement model training method according to an embodiment of the present invention. Figure 2 Schematic diagram of the structure of a pressurized blood pressure measurement model according to an embodiment of the present invention. Specifically, one aspect of the present invention provides a blood pressure measurement model training method, the method comprising the following steps S101 to S103:

[0046] Step S101: Obtain a training sample set, which includes a pulse wave signal data set and a pressure change signal data set obtained by the pulse wave of the user's wrist acting on the pressure applied to the user's wrist; add the corresponding blood pressure measurement results obtained after full-time pressurization to the samples in the training sample set as labels; full-time pressurization means that low-pressure signals and high-pressure signals are applied to the user's wrist using a blood pressure measurement device throughout the blood pressure measurement process.

[0047] Step S102: Obtain an initial neural network model, which includes a time series prediction network, a feature extraction network and a decoding network; the time series prediction network takes the pulse wave signal data set and the pressure change signal data set as input, outputs a predicted pulse wave signal and a predicted full-process pressure change signal of the blood pressure measurement, and obtains personal vascular characteristics according to a preset measurement model; the feature extraction network takes the pulse wave signal data set as input, outputs a high-dimensional pulse feature; the personal vascular features and the high-dimensional pulse features are fused and input into the decoding network, the decoding network learns the relationship between the personal vascular features and the high-dimensional pulse features and the blood pressure data and outputs the user's blood pressure data prediction result.

[0048] Step S103: Iteratively update the parameters of the initial neural network model by minimizing the deviation between the predicted results and the actual results of the blood pressure data, thereby obtaining a blood pressure measurement model.

[0049] In step S101, when the pulse wave of the user's wrist in the present invention acts on the pressure applied to the user's wrist to obtain a pressure change signal data set, the pressure applied to the user's wrist is low pressure, which avoids affecting the user's experience due to excessive pressure and long pressure application time during long-term blood pressure measurement, and making it impossible to measure comfortably for a long time; the label of the training sample is the measurement result obtained by the blood pressure measuring device through the full pressurization process. The blood pressure measuring device of the present invention adopts the gold standard blood pressure device, and the full pressurization process is a process of gradually increasing the pressure applied to the user's wrist from low pressure to high pressure.

[0050] In step S102, an initial neural network model for measuring blood pressure is trained, and the applied pulse wave signal and pressure change signal are input to the timing prediction network. The pulse wave of the user's wrist affects the pressure applied to the wrist to obtain a pressure change signal. The pressure is applied and controlled by the pressurization module. The pulse wave signal is obtained by the pulse wave measurement module and processed by the signal denoising module. The timing prediction network obtains the pulse wave change signal by predicting the pulse wave signal's response to the pressure change signal throughout the process. The pulse wave signal changes and the pressure change signal are analyzed during the entire pressurization process to understand the mutual influence relationship between blood vessels and pressure, and to measure the user's personal blood vessel characteristics.

[0051] In some embodiments, the timing prediction network includes a serially arranged sequence block layer, a first attention and normalization layer, a first feedforward fully connected and normalized layer, and a fully connected mapping layer; a residual short connection is established between the input of the feedforward fully connected and normalized layer and the input of the fully connected mapping layer.

[0052] Specifically, the sequence blocking layer divides the input pulse wave signal and pressure change signal into segments, thereby improving the model's ability to understand and express data features, while reducing computational complexity to improve the model's training efficiency. Furthermore, the sequence is processed in blocks at different levels to enhance the model's prediction ability and accuracy. Furthermore, the input pulse wave signal and pressure change signal finally pass through the fully connected mapping layer, and the predicted pulse wave signal and full-process pressure change signal are obtained through an autoregressive prediction mechanism.

[0053] In some embodiments, the method also includes: inputting the pulse wave signal data set and the pressure change signal data set into a timing prediction network to obtain a predicted pulse wave signal and a full-range pressure change signal, predicting that the pulse wave signal responds to the full-range pressure change signal and obtains a pulse wave change signal, and obtaining personal vascular characteristics according to a preset measurement model.

[0054] Furthermore, in some embodiments, fusing the personal blood vessel feature with the high-dimensional pulse feature includes steps S1021 to S1022:

[0055] Step S1021: Map personal vascular characteristics and high-dimensional pulse characteristics to the same feature space through a preset mapping function; personal vascular characteristics include vascular diameter, vascular wall thickness, blood flow velocity and vascular compliance parameters; high-dimensional pulse characteristics include pulse wave waveform characteristics, pulse wave conduction characteristics, pulse wave frequency characteristics, pulse wave dynamic characteristics and pulse wave pulse characteristics.

[0056] Step S1022: The personal blood vessel features and high-dimensional pulse features mapped to the same feature space are weighted according to feature importance, and directly concatenated to obtain fusion features.

[0057] Specifically, the pulse wave change signal is obtained by using the predicted pulse wave signal and the full-process pressure change signal output by the timing prediction network, and the influence of different pressurization conditions on the pulse wave signal is analyzed; the blood vessel diameter, blood vessel wall thickness, blood flow velocity and blood vessel compliance parameters are calculated by measuring the speed at which the pulse wave signal propagates in the blood vessel, and the change in blood vessel diameter is analyzed through the periodic change of the pulse wave signal; the pulse wave waveform characteristics analyze the peak time, peak width and peak area of ​​the pulse wave, the pulse wave conduction characteristics analyze the conduction time and speed of the pulse wave signal in the blood vessel, the pulse wave frequency characteristics analyze the heart rate, pulse wave number and frequency distribution, the pulse wave dynamic characteristics analyze the trend and fluctuation of the pulse wave over time, and the pulse wave pulse characteristics analyze the amplitude of the pulse peak and the time interval between the peaks, so as to better understand the user's pulse state.

[0058] Furthermore, the preset measurement model is a personalized vascular feature model of the user, which is used to accurately analyze and judge the user's personal vascular characteristics based on the user's predicted pulse wave signal and the predicted full-process pressure change signal of the blood pressure measurement; mapping the personal vascular characteristics and high-dimensional pulse characteristics to the same feature space through a preset mapping function solves the problem of dimensional and distribution differences of different features during feature fusion, and weights the personal vascular characteristics and high-dimensional pulse wave features to divide the importance and influence of the two features; using direct splicing to obtain fused features is conducive to simplifying the processing flow and obtaining new feature vectors simply and directly.

[0059] In some embodiments, the feature extraction network includes a convolutional layer, a second attention and normalization layer, a second feedforward fully connected and normalized layer, a third attention and normalization layer, a third feedforward fully connected and normalized layer, a fourth attention and normalization layer, a fourth feedforward fully connected and normalized layer and a category label extraction layer arranged in succession; a residual short connection is established between the input and output of the second attention and normalization layer; a residual short connection is established between the input and output of the second feedforward fully connected and normalized layer; a residual short connection is established between the input and output of the third attention and normalization layer; a residual short connection is established between the input and output of the third feedforward fully connected and normalized layer; a residual short connection is established between the input and output of the fourth attention and normalization layer; a residual short connection is established between the input and output of the fourth feedforward fully connected and normalized layer.

[0060] Specifically, the feature extraction network converts the pulse wave signal into a data representation that can reflect the essential characteristics of the pulse wave signal. The high-dimensional pulse feature contains detailed information of the pulse wave signal for analysis and processing. Furthermore, after the input pulse wave signal passes through the convolution layer, a learnable category label is added to identify and distinguish features of different categories to improve the recognition accuracy and classification ability of the model. By learning labels of different categories, richer and more accurate information support is provided for subsequent decision-making and prediction. When a softmax function layer is set after the category label extraction layer, the probability distribution of each category can be output to perform multi-category classification tasks and obtain classification results.

[0061] In some embodiments, the decoding network includes a first fully connected layer, a first nonlinear activation function layer, a second fully connected layer, a second nonlinear activation function layer, and a third fully connected layer that are continuously arranged. In some embodiments, the decoding network uses a convolutional neural network, a multilayer perceptron, or a deep learning model. The decoding network structure senses the fusion of personal vascular features and high-dimensional pulse features and calculates the user's blood pressure data to generate accurate blood pressure measurement results.

[0062] In step S103, by minimizing the predicted results and the actual results and continuously updating the parameters of the initial neural network model, the blood pressure measurement model obtained by continuously adjusting the parameters can more accurately predict the blood pressure data and improve the efficiency of blood pressure data measurement.

[0063] Figure 3 FIG. 1 is a schematic diagram of the structure of a pulse blood pressure measurement system according to an embodiment of the present invention. In another aspect, the present invention further provides a pulse blood pressure measurement system, which is deployed on a wrap-around device worn on a user's wrist, and comprises:

[0064] The pulse wave measurement module includes a PPG sensor component and a filter component, which is used to collect the PPG signal from the user's wrist and obtain the pulse wave signal after filtering through the filter component. Specifically, the PPG sensor component collects the PPG signal. The PPG signal is due to the part of the light sent by the light emitting diode that is not absorbed by the tissue returning to the photodiode. The blood volume in the arterial blood vessels will affect the light intensity returned to the photodiode. After obtaining the light intensity change, the photoelectric detection will be converted into a PPG signal. The PPG signal changes periodically with the contraction and relaxation of the heartbeat. The pulse wave is sensed through the PPG signal and filtered to obtain the pulse wave signal; the photodiode converts the reflected light into a digital PPG signal.

[0065] In some embodiments, collecting the user's PPG signal includes steps S201 to S204:

[0066] Step S201: The light emitting diode deployed in the PPG sensor emits light of a specific wavelength to the skin of the user's wrist; the light of the specific wavelength can penetrate the skin.

[0067] Step S202: After the light penetrates the skin, part of the light is absorbed by the arteries, veins and other tissues, and the light that is not absorbed is reflected to the photodiode deployed in the PPG sensor.

[0068] Step S203: When the amount of blood in the artery increases, the light intensity detected by the photodiode decreases; when the amount of blood in the artery decreases, the light intensity detected by the photodiode increases.

[0069] Step S204: The photodiode converts the change in light intensity into a PPG signal.

[0070] The pressurization module includes a micro air pump assembly and a high-precision air pressure sensor, which are used to control the pressure applied to the user's wrist and detect the pressure change signal obtained by the pulse wave acting on the pressure through the high-precision air pressure sensor. Specifically, the pressurization module is deployed on a wrap-around pressurization device worn on the user's wrist. The wrap-around pressurization device applies wrap-around pressure to the user's wrist, controls the pressure applied to the user's wrist and ensures that the user's measurement experience is not affected by the pressurization process. The micro air pump assembly is used to provide pressure applied to the wrist, and the high-precision air pressure sensor is used to measure the pressure change signal obtained by the pulse wave affecting the pressure.

[0071] The signal denoising module receives the pulse wave signal and the pressure change signal and performs denoising processing through the microcontroller. In some embodiments, the denoising processing through the microcontroller includes: reducing the noise in the pulse wave signal and the pressure change signal through beamforming, wavelet transform and variational mode decomposition filtering. Specifically, the high-definition and high-precision pulse wave signal and pressure change signal obtained after the denoising process provide high-quality input signals for the blood pressure measurement module to measure blood pressure.

[0072] The signal transmission module includes a Bluetooth component and a wireless transmission component, which is used to transmit the pulse wave signal and pressure change signal after noise reduction processing to the blood pressure measurement module.

[0073] The blood pressure measurement module executes any one of the above-mentioned blood pressure measurement model training methods to obtain blood pressure data through pulse wave signals and pressure change signals.

[0074] The storage module is used to store blood pressure data, pulse wave signals after denoising, and pressure change signals. Specifically, the blood pressure data, pulse wave signals after denoising, and pressure change signals are stored as a vascular feature database to analyze and obtain the relationship between blood pressure, pulse, and externally applied pressure, and to integrate and process data from different situations, so as to facilitate subsequent review, analysis, and provide a basis for decision-making, thereby optimizing the measurement process.

[0075] The present invention is described below in conjunction with a specific embodiment:

[0076] The present invention proposes a method for using a pulse blood pressure measurement system to calculate personal vascular characteristics, and then accurately obtain the user's blood pressure data prediction results. The first thing to be solved is how to design a pulse blood pressure measurement system. In the pulse blood pressure measurement system, a PPG sensor component is used to collect pulse wave signals, and the pressurization module uses a wrapped pressurization device to apply wrapped pressure to the user's wrist. The next core technology is how to use the pulse blood pressure measurement system to calculate personal vascular characteristics and accurately measure blood pressure. A timing prediction network is designed to predict the pulse wave signal under full pressurization and the pulse wave change signal that responds to pressure based on the pulse wave signal under pressurization, and then obtain personal vascular characteristics, and obtain high-dimensional pulse features through the pulse wave signal through the feature extraction network, and then input the personal vascular characteristics and high-dimensional pulse features into the decoding network to achieve accurate and robust blood pressure measurement.

[0077] 1. The overall process of pulse blood pressure measurement.

[0078] (1) Wear the wrap-around device on the subject’s wrist to collect data.

[0079] (2) The collected signals are processed and the pulse wave signals are obtained by filtering.

[0080] (3) The pulse wave signal during the pressurization process is input into the timing prediction network to obtain individual vascular characteristics.

[0081] (4) The pulse wave signal and personal vascular characteristics are input into the decoding network to measure blood pressure.

[0082] 2. The pulse and blood pressure measurement system includes a pulse measurement module, a pressurization module, a signal denoising module, a signal transmission module, a blood pressure measurement module and a storage module.

[0083] The pulse measurement module uses a PPG sensor to sense changes in blood volume in blood vessels and collect rich and detailed data from the subject's wrist. The data will be further fed into the signal denoising module to provide a high-quality input source for subsequent neural networks.

[0084] The pressurization module includes a micro air pump and a high-precision air pressure sensor to accurately control the pressure applied to the user. The pressure change signal obtained by the wrist artery's response to the pressure will eventually be detected by the high-precision air pressure sensor and further sent to the signal denoising module. The data is synchronized and processed by the microcontroller to prepare for the next step of blood pressure measurement.

[0085] The signal denoising module uses signal enhancement technology through a microcontroller integrated in the wristband to reduce the noise in the perceived signal through advanced signal processing methods including but not limited to beamforming, wavelet transform and variational mode decomposition filtering, thereby improving the clarity and accuracy of the signal.

[0086] The signal transmission module realizes the communication between the signal and related subsequent tools, including the Bluetooth module and the wireless transmission module, and further sends the measured signal to the subsequent blood pressure measurement module.

[0087] The blood pressure measurement module executes the blood pressure measurement model training method to obtain the blood pressure measurement model, and obtains the blood pressure data through the pulse wave signal and the pressure change signal.

[0088] The storage module is used to store blood pressure data, pulse wave signals after denoising, and pressure change signals.

[0089] 3. Blood pressure measurement combines the pulse wave signal during the pressurization process, and realizes efficient representation and utilization of physiological signals through the collaborative work of the timing prediction network, feature extraction network and decoding network. First, define the timing prediction network, obtain the pulse wave change signal after the full pressurization through the pulse wave signal under pressurization, and obtain the predicted pulse wave signal response to the complete pressurization process by inputting the pulse wave signal's response to the depressurization process, and then calculate the user's personal vascular characteristics.

[0090] In addition, the user's pulse wave signal is encoded through the feature extraction network to extract high-dimensional features related to hemodynamics, and the hemodynamic features are spliced ​​with personal vascular features to obtain a reliable and robust high-dimensional representation. Finally, the decoding network consists of three fully connected layers separated by nonlinear activation functions. The decoding network fully perceives the extracted fusion features and calculates the user's blood pressure value. Through this end-to-end processing flow, more accurate measurement results can be generated for each user, personal information and physiological signals can be organically combined, and the overall performance and applicability of the blood pressure measurement model can be improved. During the training process of the blood pressure measurement model, the measurement results of the gold standard blood pressure device are used as sample labels. The process of blood pressure measurement by the blood pressure measurement model includes:

[0091] (1) The pulse wave signal and pressure change signal are transmitted to the timing prediction network to obtain the individual vascular characteristics.

[0092] (2) The pulse wave signal is passed into the feature extraction network to obtain high-dimensional pulse features.

[0093] (3) Combine personal vascular features with high-dimensional pulse features.

[0094] (4) Predict the user's blood pressure data through the decoding network.

[0095] In summary, the present invention provides a blood pressure measurement model training method and a pulse blood pressure measurement system, wherein a pulse wave measurement module obtains a pulse wave signal, the pulse wave of the user's wrist acts on the pressure of a pressurizing module to obtain a pressure change signal, the pulse wave signal and the pressure change signal are denoised by a signal denoising module and transmitted to the blood pressure measurement module, the blood pressure measurement module executes the blood pressure measurement model training method, and an initial neural network is trained by a pulse wave signal data set and a pressure change signal data set, the initial neural network model includes a timing prediction network, a feature extraction network and a decoding network, the timing prediction network and the feature extraction network respectively output personal vascular features and high-dimensional pulse features, and after fusion, the blood pressure data prediction result is output through the decoding network, the deviation between the blood pressure data prediction result and the actual result is minimized, the parameters of the initial neural network model are iteratively updated to obtain a blood pressure measurement model, and blood pressure data is obtained through the blood pressure measurement model.

[0096] Furthermore, the present invention denoises the pulse wave signal and the pressure change signal, reduces the noise in the pulse wave signal and the pressure change signal through beamforming, wavelet transform and variational mode decomposition filtering, improves the clarity and accuracy of the signal, and provides a high-quality input signal for blood pressure measurement.

[0097] An embodiment of the present invention further provides a computer device, which may include a processor and a memory, wherein the processor and the memory may be connected via a bus or in other ways.

[0098] The processor may be a central processing unit (CPU). The processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or a combination of the above chips.

[0099] The memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer executable programs and modules, such as the program instructions / modules corresponding to the key shielding method of the vehicle-mounted display device in the embodiment of the present invention. The processor executes various functional applications and data processing of the processor by running the non-transitory software programs, instructions and modules stored in the memory.

[0100] The memory may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created by the processor, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0101] The one or more modules are stored in the memory, and when executed by the processor, the method described in this embodiment is performed.

[0102] The embodiment of the present invention also provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the steps of the aforementioned edge computing server deployment method are implemented. The computer-readable storage medium can be a tangible storage medium, such as a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a floppy disk, a hard disk, a removable storage disk, a CD-ROM, or any other form of storage medium known in the technical field.

[0103] It should be understood by those skilled in the art that the exemplary components, systems and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software or a combination of the two. Whether it is performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present invention are programs or code segments used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link via a data signal carried in a carrier.

[0104] It should be clear that the present invention is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present invention.

[0105] In the present invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with features of other embodiments or replace features of other embodiments.

[0106] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the embodiments of the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A blood pressure measurement model training method, characterized in that: The method comprises the following steps: Acquire a training sample set, the training sample set comprising a pulse wave signal data set and a pressure change signal data set obtained by the pulse wave of the user's wrist acting on the pressure applied to the user's wrist; add corresponding blood pressure measurement results obtained by full-time pressurization as labels to samples in the training sample set; the full-time pressurization means that a low-pressure signal and a high-pressure signal are applied to the user's wrist by using a blood pressure measurement device during the whole process of blood pressure measurement; Acquire an initial neural network model, the initial neural network model includes a time series prediction network, a feature extraction network and a decoding network; the time series prediction network takes the pulse wave signal data set and the pressure change signal data set as input, outputs a predicted pulse wave signal and a predicted full-course pressure change signal of the blood pressure measurement, and obtains personal vascular characteristics according to a preset measurement model; the feature extraction network takes the pulse wave signal data set as input, outputs a high-dimensional pulse feature; the personal vascular characteristics and the high-dimensional pulse features are fused and input into a decoding network, the decoding network learns the relationship between the personal vascular characteristics and the high-dimensional pulse characteristics and the blood pressure data, and outputs the blood pressure data prediction result of the user; The parameters of the initial neural network model are iteratively updated by minimizing the deviation between the predicted results and the actual results of the blood pressure data, thereby obtaining a blood pressure measurement model.

2. The blood pressure measurement model training method according to claim 1, characterized in that: The time series prediction network includes a sequence block layer, a first attention and normalization layer, a first feedforward fully connected and normalized layer and a fully connected mapping layer that are continuously arranged; the input of the feedforward fully connected and normalized layer and the input of the fully connected mapping layer establish a residual short connection.

3. The blood pressure measurement model training method according to claim 1, characterized in that: The feature extraction network includes a convolutional layer, a second attention and normalization layer, a second feedforward fully connected and normalized layer, a third attention and normalization layer, a third feedforward fully connected and normalized layer, a fourth attention and normalization layer, a fourth feedforward fully connected and normalized layer, and a category label extraction layer that are arranged in succession; The input and output of the second attention and normalization layer establish a residual short connection; The second feedforward full connection establishes a residual short connection with the input and output of the normalization layer; the third attention establishes a residual short connection with the input and output of the normalization layer; The third feedforward full connection establishes a residual short connection with the input and output of the normalization layer; the fourth attention establishes a residual short connection with the input and output of the normalization layer; the fourth feedforward full connection establishes a residual short connection with the input and output of the normalization layer.

4. The blood pressure measurement model training method according to claim 1, characterized in that: The decoding network includes a first fully connected layer, a first nonlinear activation function layer, a second fully connected layer, a second nonlinear activation function layer and a third fully connected layer which are arranged in succession.

5. The blood pressure measurement model training method according to claim 1, characterized in that: The decoding network adopts a convolutional neural network, a multi-layer perceptron or a deep learning model.

6. The blood pressure measurement model training method according to claim 1, characterized in that: The fusion of the personal blood vessel feature and the high-dimensional pulse feature comprises: Mapping the personal blood vessel features and the high-dimensional pulse features to the same feature space through a preset mapping function; the personal blood vessel features include blood vessel diameter, blood vessel wall thickness, blood flow velocity and blood vessel compliance parameters; the high-dimensional pulse features include pulse wave waveform features, pulse wave conduction features, pulse wave frequency features, pulse wave dynamic features and pulse wave pulse features; The personal blood vessel features and the high-dimensional pulse features mapped to the same feature space are weighted according to feature importance and directly concatenated to obtain fused features.

7. The blood pressure measurement model training method according to claim 1, characterized in that: The method also includes: the pulse wave signal data set and the pressure change signal data set are input into a timing prediction network to obtain the predicted pulse wave signal and the full-range pressure change signal, the predicted pulse wave signal responds to the full-range pressure change signal and obtains a pulse wave change signal, and the personal vascular characteristics are obtained according to a preset measurement model.

8. A pulse blood pressure measurement system, characterized in that: The system is deployed on a wrap-type device worn on the wrist of a user, and the system includes: A pulse wave measurement module, including a PPG sensor component and a filter component, is used to collect the PPG signal of the user's wrist and obtain a pulse wave signal after filtering through the filter component; A pressurizing module, including a micro air pump assembly and a high-precision air pressure sensor, for controlling the pressure applied to the wrist of the user, and detecting a pressure change signal obtained by the pulse wave acting on the pressure through the high-precision air pressure sensor; A signal denoising module receives the pulse wave signal and the pressure change signal and performs denoising processing through a microcontroller; A signal transmission module, including a Bluetooth component and a wireless transmission component, for transmitting the pulse wave signal and the pressure change signal after denoising to a blood pressure measurement module; A blood pressure measurement module, the module being used to execute the blood pressure measurement model training method according to any one of claims 1 to 7 to obtain a blood pressure measurement model, and to obtain blood pressure data through the pulse wave signal and the pressure change signal; A storage module is used to store the blood pressure data, the pulse wave signal after denoising, and the pressure change signal.

9. The pulse blood pressure measurement system according to claim 8, characterized in that: Collecting the PPG signal of the user includes: The light emitting diode deployed in the PPG sensor emits light of a specific wavelength toward the skin of the user's wrist; the light of the specific wavelength can penetrate the skin; After the light penetrates the skin, a portion of the light is absorbed by arteries, veins and other tissues, and the light that is not absorbed is reflected to the photodiode deployed in the PPG sensor; When the amount of blood in the artery increases, the light intensity detected by the photodiode decreases; when the amount of blood in the artery decreases, the light intensity detected by the photodiode increases; The photodiode converts the change in light intensity into a PPG signal.

10. The pulse blood pressure measurement system according to claim 8, characterized in that: The denoising process performed by the microcontroller includes: reducing the noise in the pulse wave signal and the pressure change signal by beamforming, wavelet transform and variational mode decomposition filtering.

Images