Non-inductive blood pressure continuous monitoring method and device based on single calibration and storage medium

Abstract

The invention provides a non-inductive blood pressure continuous monitoring method and device based on single calibration and a storage medium, and the method comprises the steps: in an equipment calibration stage, in a measurement period of a pressurized sphygmomanometer, obtaining a human body monitoring signal; calculating periodic intensities of signals at different monitoring distance units based on the human body monitoring signals, and determining a target monitoring position; continuously acquiring human body monitoring signals corresponding to the target monitoring position, extracting phase change signals from the human body monitoring signals, further extracting pulse wave signals, and inputting the signals into a blood vessel feature extraction network to obtain blood vessel feature parameters; in the independent monitoring stage, continuously acquiring a human body monitoring signal corresponding to the target monitoring position; and segmenting the human body monitoring signal of the target monitoring position into multiple groups of signal segments in parallel, and inputting the signal segments and the blood vessel characteristic parameters into a pre-trained blood pressure prediction model to obtain a time sequence prediction result representing the systolic pressure and the diastolic pressure of the blood vessel. According to the invention, nighttime non-inductive blood pressure measurement can be realized, and meanwhile, the accuracy and robustness of nighttime blood pressure continuous tracking are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of blood pressure measurement, and in particular to a non-invasive blood pressure continuous monitoring method based on single calibration, a device and a storage medium. BACKGROUND

[0002] In the field of cardiovascular monitoring, nighttime blood pressure is defined as being measured at intervals of ≤30 minutes during 22:00-6:00 and having ≥20 valid readings. Normal “dipper” blood pressure refers to a 10-20% decrease in nighttime systolic pressure compared to daytime, reflecting normal autonomic nervous function; abnormalities include non-dipper (decrease <10%), reverse dipper (nighttime blood pressure higher than daytime), and extreme dipper (decrease >20%). Circadian blood pressure rhythm is an independent predictor of cardiovascular events, but conventional clinic and home monitoring cannot accurately identify it, resulting in a high rate of missed diagnosis and exposing patients to high risk for a long time. Therefore, there is an urgent need to develop an accurate nighttime blood pressure measurement method that meets the quantitative definition.

[0003] Traditional blood pressure measurement methods include those based on the principle of oscillography. By gradually inflating the cuff, the artery is compressed to stop blood flow, and an acoustic or optical sensor is used to observe the changes in the pulse waveform during the inflation process. When the cuff pressure is lower than the diastolic pressure, the detected pulse is very weak. As the cuff pressure gradually increases, the pulse amplitude first increases and then decreases. When the cuff pressure is equal to the systolic pressure, blood continues to pass through the occluded artery, but only the highest arterial pressure can be detected. The mean pressure is determined by detecting the position of the maximum pulse amplitude, and the systolic and diastolic pressures are determined by the peak amplitude ratio.

[0004] However, traditional blood pressure measurement methods require inflation every 30 minutes at night and pressurization above systolic pressure, which produces noise that can easily wake up users at night. Long-term over-pressurization can cause redness and local congestion of the user's skin, affecting the normal flow of blood, not only causing errors in measurement results, but also potentially damaging the user's vascular health. It is not suitable for nighttime blood pressure measurement and can only provide instantaneous measurement results, making it difficult to continuously monitor comfortably. SUMMARY

[0005] In view of this, the embodiments of the present application provide a non-invasive blood pressure continuous monitoring method based on single calibration, a device and a storage medium to eliminate or improve one or more defects in the prior art. It can solve the problem that the traditional blood pressure measurement method is complicated and cannot be continuously monitored at night.

[0006] One aspect of the present application provides a non-invasive blood pressure continuous monitoring method based on single calibration, which comprises the following steps: In the device calibration phase, in the measurement period of the pressurized sphygmomanometer, the human body monitoring signal of the target human body based on the millimeter wave device is obtained; Based on the human body monitoring signal, the periodic intensity of the signal at different monitoring distance units is calculated, and the distance unit with the maximum periodic intensity is determined as the target monitoring position; The human body monitoring signal corresponding to the target monitoring position is continuously acquired, a phase change signal is extracted therefrom, and the phase change signal is processed to extract a pulse wave signal representing arterial pulsation; The pulse wave signal is input into a pre-trained blood vessel feature extraction network to obtain blood vessel feature parameters representing blood vessel elasticity and peripheral resistance; In the independent monitoring stage, the human body monitoring signal corresponding to the target monitoring position is continuously acquired; and the millimeter wave device performs signal acquisition at a preset sampling rate; Based on predetermined different time scales, the human body monitoring signal at the target monitoring position is divided into multiple groups of signal segments in parallel, and the blood vessel feature parameters are input into a pre-trained blood pressure prediction model to obtain time series prediction results representing systolic and diastolic blood pressure; wherein the blood pressure prediction model includes a feature extraction path corresponding to each group of signal segments.

[0007] In some embodiments of the present application, the blood pressure prediction model includes a first network branch, a second network branch, a cross-attention mechanism layer connected to the first network branch and the second network branch respectively, and a fully connected output layer connected to the cross-attention mechanism; The first network branch includes multiple feature extraction paths for extracting a joint feature tensor representing pulse waveforms, target human heart rate and respiration based on multiple groups of signal segments; The second network branch is used to receive the blood vessel feature parameters; The cross-attention mechanism layer is used to fuse the blood vessel feature parameters as a condition vector with the joint feature tensor to obtain a fusion feature, and input the fusion feature into the fully connected output layer; The fully connected output layer is used to output the time series prediction results based on the fusion feature.

[0008] In some embodiments of the present application, the first network branch includes a first path, a second path, a third path, a shared weight residual block connected to the first path, the second path and the third path respectively, an intra-path attention layer, an inter-channel attention layer and a feature concatenation layer; The first path is used to extract waveform features of the pulse wave based on the signal segment; the second path is used to extract heart rate features of the target human based on the signal segment; and the third path is used to extract respiration features of the target human based on the signal segment; The input end of the intra-path attention layer is connected to the output end of the shared weight residual block, and is used to weight the waveform features, the heart rate features and the respiration features; The input end of the inter-path attention layer is connected with the output end of the intra-path attention layer, and is used for fusing the weighted waveform features, the heart rate features and the respiration features; The input end of the feature splicing layer is connected with the output end of the inter-path attention layer, and is used for fusing the features into a joint feature tensor.

[0009] In some embodiments of the present application, the human body monitoring signals from the target monitoring position are segmented into multiple groups of signal segments in parallel based on different time scales, including: The human body monitoring signals from the target monitoring position are segmented into a plurality of first signal segments according to a first preset time length, and are input into a first path; The human body monitoring signals from the target monitoring position are segmented into a plurality of second signal segments according to a second preset time length, and are input into a second path; The human body monitoring signals from the target monitoring position are segmented into a plurality of third signal segments according to a third preset time length, and are input into a third path; wherein the first preset time length is shorter than the second preset time length, and the second preset time length is shorter than the third preset time length.

[0010] In some embodiments of the present application, the blood pressure prediction model is trained by the following steps: Obtain training data, the training data including a plurality of batches of sample signal segments and corresponding sample time series prediction results; each batch of sample signal segments includes a sample first signal segment, a sample second signal segment and a sample third signal segment; Input the training data in batches into an initial blood pressure prediction model to obtain a training result; the model structure of the initial blood pressure prediction model is the same as that of the blood pressure prediction model; Input the training result and the sample time series prediction result into a preset loss function to obtain a loss value; Iteratively train the initial blood pressure prediction model using the loss value to obtain the blood pressure prediction model.

[0011] In some embodiments of the present application, the blood vessel feature extraction network is trained by the following steps: Obtain training data, the training data including a sample pulse wave signal and a sample pulse wave signal corresponding supervision signal; the supervision signal includes a sample cuff pressure signal synchronously collected by a pressure type sphygmomanometer during collection of the sample pulse wave signal; Input the sample pulse wave signal into an initial feature extraction network to obtain a training result; Input the training result and the supervision signal into a preset loss function to obtain a loss value; Iteratively train the feature extraction network using the loss value until the model converges to obtain the blood vessel feature extraction network.

[0012] In some embodiments of the present application, the human body monitoring signal is obtained by mixing the transmission signal of the millimeter wave device with the reflection signal corresponding to the transmission signal to obtain an intermediate frequency signal, and then processing the intermediate frequency signal by fast Fourier transform; Based on the human body monitoring signal, the periodic intensity of the signal at different monitoring distance units is calculated, and the distance unit with the maximum periodic intensity is determined as the target monitoring position, including: Based on each peak value in the human body monitoring signal, a distance unit is determined and the intermediate frequency signal is mapped to each distance unit; Based on the signal corresponding to each distance unit and a preset autocorrelation function, an autocorrelation function value corresponding to each distance unit is calculated; the size of the autocorrelation function value is positively correlated with the periodic intensity; According to a preset heart rate range, the distance unit corresponding to the maximum autocorrelation function value is searched from the distance units, and is taken as the target distance unit; Based on the target distance unit, the target monitoring position is determined.

[0013] In some embodiments of the present application, the phase change signal is processed to extract a pulse wave signal representing arterial pulsation, including: The phase change signal is subjected to band-pass filtering processing to remove noise; The noise-removed phase change signal is subjected to wavelet transform processing to obtain a pulse wave signal.

[0014] Another aspect of the present application provides a single calibration based non-invasive blood pressure continuous monitoring device, including a processor, a memory and a computer program / instruction stored on the memory, the processor being configured to execute the computer program / instruction, and when the computer program / instruction is executed, the system implements the steps of the single calibration based non-invasive blood pressure continuous monitoring method as described above; the device is in communication connection with the millimeter wave device to receive the human body monitoring signal of the target human body collected by the millimeter wave device in real time; and the device is in communication connection with the pressurized sphygmomanometer to receive the cuff pressure signal collected by the pressurized sphygmomanometer in real time.

[0015] Another aspect of the present application provides a computer readable storage medium having a computer program / instruction stored thereon, and the computer program / instruction is executed by a processor to implement the steps of the single calibration based non-invasive blood pressure continuous monitoring method as described above.

[0016] The single calibration based non-invasive blood pressure continuous monitoring method and device can solve the problem that the traditional blood pressure measurement method is complicated and cannot be used for continuous monitoring at night; by synchronously acquiring the cuff pressure signal and the human body monitoring signal of the millimeter wave device during the short pressurization process of the pressurized sphygmomanometer, combining the blood vessel feature extraction network, only single calibration is needed to obtain the personalized blood vessel feature parameters reflecting the blood vessel elasticity and peripheral resistance characteristics, which significantly reduces the dependence on frequent calibration, avoids the user sleep disturbance and potential blood vessel health risk caused by multiple pressurization in the traditional scheme.

[0017] In addition, during the night continuous monitoring stage, the user only needs to wear the millimeter wave monitoring device to realize non-invasive and non-disturbing blood pressure continuous monitoring, which ensures the comfort of the wearable device while meeting the clinical level accuracy requirement, and can be applied to various scenes requiring long-term monitoring of blood pressure fluctuation rules.

[0018] Additional advantages, objects, and features of the application will be set forth in part in the description which follows, and in part will become apparent to those skilled in the art upon examination of the following or can be learned by practice of the application. The objects and other advantages of the application can be realized and attained by the structure particularly pointed out in the description and claims hereof as well as the appended drawings.

[0019] It will be understood by those skilled in the art that the objects and advantages of the present application can be realized and attained by the structure particularly pointed out in the appended claims and carried out as hereinafter described. BRIEF DESCRIPTION OF DRAWINGS

[0020] The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings: Figure 1 The method flow chart of the single calibration based non-invasive blood pressure continuous monitoring method provided by an embodiment of the application.

[0021] Figure 2 The schematic diagram of the device correction stage and the independent monitoring stage provided by an embodiment of the application.

[0022] Figure 3 The training process schematic diagram of the blood vessel feature extraction network provided by an embodiment of the application.

[0023] Figure 4 The feature extraction flow chart of the first network branch in the blood pressure prediction model provided by an embodiment of the application.

[0024] Figure 5 This is a schematic diagram of the prediction process of a blood pressure prediction model provided in an embodiment of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the embodiments and accompanying drawings. Here, the illustrative embodiments and descriptions of this invention are used to explain the invention, but are not intended to limit the invention.

[0026] It should also be noted that, in order to avoid obscuring the invention with unnecessary details, only the structures and / or processing steps closely related to the solution according to the invention are shown in the accompanying drawings, while other details that are not closely related to the invention are omitted.

[0027] It should be emphasized that the term "including / comprises" as used herein refers to the presence of a feature, element, step, or component, but does not exclude the presence or addition of one or more other features, elements, steps, or components.

[0028] It should also be noted that, unless otherwise specified, the term "connection" in this article can refer not only to a direct connection, but also to an indirect connection involving an intermediary.

[0029] In the following description, embodiments of the invention will be illustrated with reference to the accompanying drawings. In the drawings, the same reference numerals represent the same or similar parts, or the same or similar steps.

[0030] The following is a detailed description of the non-intrusive continuous blood pressure monitoring method based on single calibration provided in this application.

[0031] like Figure 1 As shown, the embodiments of this application provide a method for continuous, non-intrusive blood pressure monitoring based on a single calibration. The implementation of this method can rely on a computer program, which can run on computer devices such as smartphones, tablets, and personal computers, or run on a server. This embodiment does not limit the operating entity of the method.

[0032] The non-contact continuous blood pressure monitoring method based on single-calibration provided in this embodiment is based on single-device calibration. It uses a millimeter-wave device to independently detect the pulse wave signal of the target human body in a non-contact manner, thereby predicting the target human body's nighttime blood pressure. Specifically, the method includes at least steps S101 to S106: Step S101: During the equipment calibration phase, the target human body's human body monitoring signal based on the millimeter-wave device is acquired during the measurement cycle of the pressurized blood pressure monitor.

[0033] Among them, the pressure blood pressure monitor includes, but is not limited to, a pressure blood pressure monitor worn on the wrist of the target human body, or a pressure blood pressure monitor worn on the arm of the target human body. This embodiment does not limit the implementation method of the pressure blood pressure monitor.

[0034] Traditional cuffless blood pressure monitoring technology (based on optical sensors or millimeter-wave radar to extract pulse wave features) improves portability and comfort, but has significant calibration limitations due to changes in physiological characteristics such as blood vessel volume and elasticity at night: it requires real-time pressure calibration (which interferes with sleep continuity) and relies on long-term personalized data collection or stimulating experiments (such as cold pressurization tests), which leads to a deterioration in user experience and makes it difficult to meet the accurate monitoring needs of natural sleep at night.

[0035] To address the aforementioned technical challenges, this embodiment simultaneously acquires millimeter-wave radar signals and cuff pressure signals during the brief inflation calibration of the pressure monitor. A personalized vascular model is established through a single calibration to support subsequent long-term, continuous, and non-intrusive blood pressure monitoring. This single inflation calibration captures the unique vascular physiological characteristics of each individual, overcoming the limitations of traditional methods that require frequent calibration or rely on stimulating experiments, significantly improving user experience and monitoring accuracy.

[0036] like Figure 2 As shown, during the device calibration phase, by simultaneously acquiring the cuff pressure signal during the inflation process of the pressurized blood pressure monitor and the human body monitoring signal of the millimeter-wave device, more vascular characteristic parameters (such as pulse wave velocity, pulse pressure, etc.) for characterizing vascular elasticity and vascular characteristic parameters (such as total peripheral resistance, etc.) for characterizing peripheral vascular resistance are obtained through a single calibration.

[0037] The human body monitoring signal is obtained by mixing the transmitted signal of the millimeter-wave device with the corresponding reflected signal to obtain the intermediate frequency signal, and then processing it through Fast Fourier Transform (FFT).

[0038] Specifically, a millimeter-wave device (such as an FMCW millimeter-wave radar) worn on the wrist or placed at a designated location transmits a frequency-modulated continuous wave (FMCW) signal and receives the echo reflected from the target human body. The transmitted and received signals are down-converted by a mixer to obtain an intermediate frequency signal containing vital signs information of the target human body. This signal is modulated by periodic micro-movements on the target human body surface and carries rich respiratory and pulse characteristics.

[0039] The transmitted signal of a millimeter-wave device can be expressed by the following formula: In the formula, This represents a complex exponential signal emitted at time t; denotes the amplitude of the transmitted signal of the millimeter wave device; denotes the base of the complex exponential function; denotes the imaginary unit; denotes the center carrier frequency of the millimeter wave device; denotes the bandwidth of the chirp signal; denotes the period of a chirp signal.

[0040] The reflected signal can be represented by the following formula: wherein, denotes the signal received by the millimeter wave device at time t; denotes the amplitude of the transmitted signal of the millimeter wave device; denotes the base; denotes the imaginary unit; denotes the center carrier frequency of the millimeter wave device; denotes the bandwidth of the chirp signal; denotes the period of a chirp signal. denotes the time delay between the transmitted signal and the received signal.

[0041] By performing a fast Fourier transform on the intermediate frequency signal, the time domain signal can be converted to the frequency domain to generate a spectrum with distinct and separated spectral peaks, thereby obtaining the human body monitoring signal. In this spectrum, the breathing and heartbeat frequencies are usually presented in the form of distinct and distinguishable peaks, thereby achieving non-contact monitoring and extraction of human vital signs.

[0042] wherein the intermediate frequency signal can be represented by the following formula: wherein, denotes the amplitude of the intermediate frequency signal; denotes the frequency of the intermediate frequency signal, wherein, denotes the propagation speed of the millimeter wave in air; denotes the phase of the intermediate frequency signal, wherein, denotes the wavelength; denotes the base; denotes the imaginary unit.

[0043] In step S102, based on the human body monitoring signal, the period intensity of the signal at different monitoring distance units is calculated, and the distance unit with the maximum period intensity is determined as the target monitoring position.

[0044] Since the frequency of the transmitted signal of the millimeter wave device changes over time, a continuous frequency-modulated signal is formed. Therefore, by measuring the frequency difference between the reflected signal reflected by the target human body and the transmitted signal, the distance between the millimeter wave device and the monitoring position can be determined.

[0045] Specifically, the distance between the millimeter wave device and the monitoring position can be represented by the following formula: In the formula, represents the distance between the millimeter wave device and the monitoring position; represents the frequency of the intermediate frequency signal; represents the propagation speed of the millimeter wave in the air; represents the period of a chirp signal; represents the bandwidth of the linear frequency modulation pulse.

[0046] In the millimeter wave device working project, the monitoring positions at different distances will cause different time delays of the reflected signals. After mixing these time delays with the transmitted signals, multiple intermediate frequency signal components with different frequencies will be formed at the baseband. By performing fast Fourier transform on the composite intermediate frequency signal, multiple separate peaks can be generated in the frequency spectrum, each peak appearing at a specific distance unit corresponding to a reflection point or a reflecting object at a specific distance.

[0047] Through the linear mapping relationship between the frequency and the distance, the radial distance of each monitoring position relative to the millimeter wave device can be accurately calculated, that is, different detection positions can be mapped to different monitoring distance units according to the performance of the intermediate frequency signal in the frequency domain and the characteristics of the millimeter wave device.

[0048] Meanwhile, the pulse wave time series presents significant periodicity due to its rhythmic physiological activity. Based on this, the periodicity of the signal intensity time series can be measured by quantifying the offset degree of the time series and itself at different time offsets, that is, by calculating the autocorrelation function to quantify the periodicity and measure the similarity of the intermediate frequency signal and itself at different time delays.

[0049] In the formula, In the formula, represents the signal intensity of the intermediate frequency signal at time t; represents the signal intensity after time has passed; represents the length of the entire time series.

[0050] In this embodiment, the size of the autocorrelation function value is positively correlated with the signal periodicity. Based on this relationship, by calculating the autocorrelation function of the signal corresponding to each distance unit, the distance unit with the maximum autocorrelation function value is determined as the target distance unit, and then the target monitoring position, i.e., the region where the target artery is located, is accurately positioned.

[0051] Meanwhile, in order to accurately extract the pulse wave signal, the calculation of the autocorrelation function needs to be constrained in a physiological reasonable heart rate range. In view of the fact that the normal variation range of the heart rate is 50-160 bpm, the corresponding time frequency range is converted, and the global maximum value of the autocorrelation function in the range is searched to determine the spatial coordinates of the wrist artery.

[0052] wherein the global maximum value of the autocorrelation function in the time frequency range corresponding to the normal variation range of the heart rate can be represented by the following formula: wherein, denotes the minimum time delay used when calculating the autocorrelation function, corresponding to the period represented by the highest heart rate (for example, 160 bpm); denotes the maximum time delay used when calculating the autocorrelation function, corresponding to the period represented by the lowest heart rate (for example, 50 bpm).

[0053] wherein, , denotes the lowest heart rate, denotes the signal sampling rate; , denotes the highest heart rate.

[0054] The spatial coordinates of the wrist artery can be determined by the following formula: wherein, denotes the target distance unit; denotes searching on all distance units d to find the d that makes the function value maximum; denotes the maximum value of the autocorrelation function corresponding to the distance unit d in the set heart rate range.

[0055] Specifically, based on the human body monitoring signal, the period intensity of the signal at different monitoring distance units is calculated, and the distance unit with the maximum period intensity is determined as the target monitoring position, including: based on each peak value in the human body monitoring signal, the distance unit is determined and the intermediate frequency signal is mapped into each distance unit; based on the signal corresponding to each distance unit and the preset autocorrelation function, the autocorrelation function value corresponding to each distance unit is calculated; according to the preset heart rate range, the distance unit corresponding to the maximum autocorrelation function value is searched from the distance unit, as the target distance unit; and the target monitoring position is determined based on the target distance unit.

[0056] In some embodiments of the present application, the target distance unit includes the distance unit where the main target artery is located and the distance unit corresponding to the adjacent reference area thereof.

[0057] Accordingly, the target monitoring location includes the location of the target artery and the location of the adjacent reference area of ​​the target artery. Among them, the adjacent reference area of ​​the target artery is located in the same scanning area as the target artery and has no obvious large blood vessels or significant periodic activity.

[0058] Step S103: Continuously acquire human body monitoring signals corresponding to the target monitoring location, extract phase change signals from them, and perform signal processing on the phase change signals to extract pulse wave signals that characterize arterial pulsation.

[0059] The phase signal at the target location is continuously monitored over time, and its phase change sequence is recorded. Then, a pulse wave signal representing arterial pulsation is extracted through a signal processing workflow. First, bandpass filtering is used to suppress low-frequency interference in the environment and high-frequency noise from the radar system. Next, wavelet transform is applied to accurately separate the periodic pulse wave signal from the processed signal.

[0060] Specifically, signal processing is performed on the phase change signal to extract the pulse wave signal that represents arterial pulsation, including: performing bandpass filtering on the phase change signal to remove noise; and performing wavelet transform on the noise-removed phase change signal to obtain the pulse wave signal.

[0061] Step S104: Input the pulse wave signal into the pre-trained vascular feature extraction network to obtain vascular feature parameters that characterize vascular elasticity and peripheral resistance.

[0062] The blood vessel feature extraction network was pre-trained through the following steps: Step S1041: Obtain training data.

[0063] like Figure 3 As shown, the training data includes sample pulse wave signals and corresponding supervisory signals; the supervisory signals include sample cuff pressure signals synchronously acquired by the compression blood pressure monitor during the acquisition of sample pulse wave signals.

[0064] Step S1042: Input the sample pulse wave signal into the initial feature extraction network to obtain the training result.

[0065] In this embodiment, the feature extraction network is a convolutional temporal network, whose input is the millimeter-wave pulse wave signal during the calibration process. The convolutional temporal network captures vascular motion characteristics, using synchronously acquired sample cuff pressure signals as supervisory values. It learns the pressure magnitude corresponding to different pulse waveform shapes and amplitudes, reflecting vascular changes under different pressure states. Finally, it extracts personalized vascular feature parameters reflecting vascular elasticity and peripheral resistance, providing a dynamic modeling basis for continuous nighttime blood pressure monitoring.

[0066] Step S1043: Input the training results and supervision signals into a preset loss function to obtain the loss value. Here, the preset loss function refers to a loss function that has been pre-set.

[0067] In some embodiments of the present invention, the preset loss function includes, but is not limited to, the mean squared error loss function or the smoothing L1 loss function. This embodiment does not limit the implementation of the preset loss function.

[0068] Step S1044: Use the loss value to iteratively train the feature extraction network until the model converges, and obtain the blood vessel feature extraction network.

[0069] Specifically, based on the loss value, the parameters of the feature extraction network are iteratively updated using the backpropagation algorithm until the model converges, ultimately yielding a trained vascular feature extraction network.

[0070] Step S105: During the independent monitoring phase, continuously acquire human body monitoring signals corresponding to the target monitoring location; the millimeter-wave device collects signals at a preset sampling rate.

[0071] like Figure 2 As shown, during the independent nighttime monitoring phase, a millimeter-wave device is worn on the wrist of the target human body to collect signals at a preset sampling rate (e.g., 10Hz, 30Hz, or 100Hz). The acquired transmitted signal is mixed with the corresponding transmitted signal to obtain an intermediate frequency signal, which is then subjected to a fast Fourier transform to obtain the human body monitoring signal.

[0072] Step S106: Based on predetermined different time scales, the human monitoring signal at the target monitoring location is segmented into multiple signal segments in parallel, and these segments, along with vascular feature parameters, are input into a pre-trained blood pressure prediction model to obtain time-series prediction results characterizing systolic and diastolic blood pressure. The blood pressure prediction model includes feature extraction pathways that correspond one-to-one with each of the multiple signal segments.

[0073] In some embodiments of the present invention, the blood pressure prediction model includes a first network branch, a second network branch, a cross-attention mechanism layer connected to the first network branch and the second network branch respectively, and a fully connected output layer connected to the cross-attention mechanism.

[0074] The first network branch includes multiple feature extraction pathways, which are used to extract joint feature tensors representing pulse wave waveforms, target human heart rate and respiration based on multiple signal segments.

[0075] In the traditional cuffless blood pressure monitoring technology, the pulsation of the artery is perceived by optical sensors, millimeter wave radar, etc., from which the amplitude of the pulse wave, pulse wave transmission time, reflected wave transmission time, systolic and diastolic time ratio and other blood pressure related features are extracted, which can improve the portability and comfort, but ignores the vascular characteristic information of the oscillogram envelope response in the pressurization process, resulting in failure in the subsequent long-term measurement.

[0076] To solve the above technical problems, as shown in the embodiment, according to different time scales, the night human body monitoring signal is divided into multiple signal segments, which are input into different paths respectively, and the convolutional neural network is used to process and extract physiological features of different time scales, including pulse waveform features (such as main wave / beat wave amplitude ratio) extracted by short-term path, heart rate features (such as heart rate variability features) extracted by medium-term path, and respiratory features (such as 0.1-0.3Hz frequency band energy respiratory rhythm coupling effect) extracted by long-term path. At the same time, each path adopts a shared weight residual block structure, and the output features are fused through the intra-path and inter-path attention mechanism to form a "pulse wave-heart rate-breathing" joint feature tensor. Figure 4 Specifically, the first network branch includes a first path, a second path, a third path, a shared weight residual block connected with the first path, the second path and the third path respectively, an intra-path attention layer, an inter-channel attention layer and a feature splicing layer;

[0077] The first path is used to extract the waveform features of the pulse wave based on the signal segment; the second path is used to extract the heart rate features of the target human body based on the signal segment; and the third path is used to extract the respiratory features of the target human body based on the signal segment. The input end of the intra-path attention layer is connected with the output end of the shared weight residual block, and is used to weight the waveform features, the heart rate features and the respiratory features. Specifically, based on the predetermined different time scales, the human body monitoring signal from the target monitoring position is segmented into multiple signal segments in parallel, and the vascular feature parameters are input into the pre-trained blood pressure prediction model, including: the human body monitoring signal at the target monitoring position is segmented into a plurality of first signal segments according to a first preset time length, and is input into the first path; the human body monitoring signal at the target monitoring position is segmented into a plurality of second signal segments according to a second preset time length, and is input into the second path; the human body monitoring signal at the target monitoring position is segmented into a plurality of third signal segments according to a third preset time length, and is input into the third path; wherein the first preset time length is shorter than the second preset time length, and the second preset time length is shorter than the third preset time length; the vascular feature parameters are input into the second network branch.

[0078]

[0079] ​In some embodiments of the present invention, the first preset duration is 2 seconds; the second preset duration is 4 seconds; and the third preset duration is 32 seconds.

[0080] In actual implementation, the first preset duration can be 1.5 seconds or 2.5 seconds, etc.; the second preset duration can be 3.5 seconds or 4.5 seconds, etc.; and the third preset duration can be 31 seconds or 33 seconds, etc. The values ​​of the first, second, or third preset durations can be adjusted according to actual needs. The blood pressure prediction model may also include other feature extraction pathways. This embodiment does not limit the values ​​of the first, second, or third preset durations, nor the implementation method of the blood pressure prediction model.

[0081] In this embodiment, the input of the inter-path attention layer is connected to the output of the intra-path attention layer, and is used to fuse weighted waveform features, heart rate features, and respiratory features. The input of the feature splicing layer is connected to the output of the inter-path attention layer, and is used to integrate the fused features into a joint feature tensor.

[0082] like Figure 5 As shown, the second network branch is used to receive vascular feature parameters. By using the vascular feature parameters obtained during the device calibration phase as a conditional vector, and fusing them with the joint feature tensor through an attention mechanism, the final output layer predicts the time series results of systolic and diastolic blood pressure.

[0083] Specifically, the cross-attention mechanism layer is used to fuse vascular feature parameters as conditional vectors with joint feature tensors to obtain fused features, which are then input into the fully connected output layer; the fully connected output layer is used to output temporal prediction results based on the fused features.

[0084] In summary, the non-contact blood pressure continuous monitoring method based on single calibration provided in this embodiment acquires the human body monitoring signal of the target human body based on the millimeter-wave device during the measurement cycle of the pressurized blood pressure monitor during the device calibration stage; based on the human body monitoring signal, the periodic intensity of the signal at different monitoring distance units is calculated, and the distance unit with the largest periodic intensity is determined as the target monitoring position; The human monitoring signal corresponding to the target monitoring position is continuously acquired, a phase change signal is extracted therefrom, and the phase change signal is processed to extract a pulse wave signal representing arterial pulsation; the pulse wave signal is input into a pre-trained blood vessel feature extraction network to obtain a blood vessel feature parameter representing blood vessel elasticity and peripheral resistance; in the independent monitoring stage, the human monitoring signal corresponding to the target monitoring position is continuously acquired; the millimeter wave device acquires signals at a preset sampling rate; based on the predetermined different time scales, the human monitoring signal of the target monitoring position is segmented into multiple signal segments in parallel, and the blood vessel feature parameter is input into a pre-trained blood pressure prediction model to obtain a time series prediction result representing systolic and diastolic blood pressure; the method can solve the problem that the traditional blood pressure measurement method is complicated and cannot be continuously monitored at night; by synchronously acquiring the cuff pressure signal and the human monitoring signal of the millimeter wave device during the short pressurization process of the pressurized sphygmomanometer, and combining the blood vessel feature extraction network, the personalized blood vessel feature parameter reflecting the characteristics of blood vessel elasticity and peripheral resistance can be obtained only through single calibration, which significantly reduces the dependence on frequent calibration and avoids the sleep disturbance and potential blood vessel health risk of the user caused by multiple pressurization in the traditional scheme. At the same time, a multi-time scale analysis mechanism is adopted to adaptively segment and process the human monitoring signal, and the blood pressure prediction model is used to extract features through multiple channels, so as to sensitively perceive the dynamic change law of the vascular characteristics at night, and significantly improve the accuracy and robustness of the continuous tracking of the blood pressure at night.

[0085] In addition, in the night continuous monitoring stage, the user only needs to wear the millimeter wave monitoring device to realize the continuous monitoring of blood pressure without awareness and disturbance, which ensures the comfort of the wearable device while meeting the clinical level accuracy requirement, and can be applied to various scenes requiring long-term monitoring of blood pressure fluctuation law, including automatic discrimination of dipper / non-dipper blood pressure rhythm of hypertensive patients, monitoring of blood pressure compensation changes related to sleep apnea syndrome, and other night blood pressure evaluation requirements in chronic disease management, etc., thereby providing an effective technical means for night blood pressure health management.

[0086] Corresponding to the above method, the application also provides a single calibration based non-aware continuous blood pressure monitoring device, which comprises a computer device comprising a processor and a memory, and the memory stores computer instructions, and the processor is used to execute the computer instructions stored in the memory, and when the computer instructions are executed by the processor, the single calibration based non-aware continuous blood pressure monitoring method device realizes the steps of the method as described above. The device is in communication connection with the millimeter wave device to receive the human monitoring signal of the target human body collected by the millimeter wave device in real time; the device is in communication connection with the pressurized sphygmomanometer to receive the cuff pressure signal collected by the pressurized sphygmomanometer in real time.

[0087] The embodiments of the present application also provide a computer readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the steps of the foregoing method. 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 memory disk, a CD-ROM, or any other form of storage medium known in the art.

[0088] Those skilled in the art should understand that the exemplary components, systems and methods described in connection with the embodiments disclosed herein can be implemented in hardware, software, or a combination thereof. The decision to implement particular embodiments in hardware or software depends on the particular application and design constraints imposed on the overall system. Skilled persons can use different methods to implement the described functions for each particular application, but such implementation should not be considered beyond the scope of the present application. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a functional card, etc. When implemented in software, the elements of the present application are program or code segments used to perform the required tasks. The program or code segments can be stored in a machine readable medium or transmitted through a data signal carried in a carrier wave over a transmission medium or communication link.

[0089] It should be noted that the present application is not limited to the specific configurations and processes described above and shown in the drawings. For the sake of brevity, detailed descriptions of known methods are omitted. In the above embodiments, several specific steps are described and shown as examples. However, the method processes of the present application are not limited to the specific steps described and shown, and skilled persons can make various changes, modifications and additions, or change the order of the steps, after understanding the spirit of the present application.

[0090] In the present application, the features described and / or illustrated for one embodiment can be used in the same way or in a similar way in one or more other embodiments, and / or in combination with or instead of features of other embodiments.

[0091] The above description is merely preferred embodiments of the present application, and is not intended to limit the present application. The embodiments of the present application can be variously changed and modified by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the scope of the present application.

Claims

1. A method for continuous, non-intrusive blood pressure monitoring based on a single calibration, characterized in that, The method includes the following steps: During the equipment calibration phase, the target human body's human monitoring signal based on the millimeter-wave device is acquired during the measurement cycle of the pressurized blood pressure monitor. Based on the human body monitoring signal, the periodic intensity of the signal at different monitoring distance units is calculated, and the distance unit with the largest periodic intensity is determined as the target monitoring location; The system continuously acquires human body monitoring signals corresponding to the target monitoring location, extracts phase change signals from them, and performs signal processing on the phase change signals to extract pulse wave signals that characterize arterial pulsation. The pulse wave signal is input into a pre-trained vascular feature extraction network to obtain vascular feature parameters characterizing vascular elasticity and peripheral resistance. During the independent monitoring phase, human body monitoring signals corresponding to the target monitoring location are continuously acquired; the millimeter-wave device collects signals at a preset sampling rate. Based on predetermined different time scales, the human monitoring signals at the target monitoring location are divided into multiple signal segments in parallel, and these segments are input into a pre-trained blood pressure prediction model along with the vascular feature parameters to obtain time-series prediction results characterizing systolic and diastolic blood pressure. The blood pressure prediction model includes feature extraction pathways that correspond one-to-one with the multiple signal segments.

2. The method according to claim 1, characterized in that, The blood pressure prediction model includes a first network branch, a second network branch, a cross-attention mechanism layer connected to the first network branch and the second network branch respectively, and a fully connected output layer connected to the cross-attention mechanism. The first network branch includes multiple feature extraction pathways, used to extract joint feature tensors representing pulse wave waveforms, target human heart rate, and respiration based on the multiple sets of signal segments; The second network branch is used to receive the vascular feature parameters; The cross-attention mechanism layer is used to fuse the vascular feature parameters as a conditional vector with the joint feature tensor to obtain fused features, and then input them into the fully connected output layer. The fully connected output layer is used to output the temporal prediction result based on the fused features.

3. The method according to claim 2, characterized in that, The first network branch includes a first path, a second path, a third path, a residual block with shared weights connected to the first path, the second path, and the third path respectively, an intra-path attention layer, an inter-channel attention layer, and a feature splicing layer; The first path is used to extract the waveform features of the pulse wave based on the signal segment; the second path is used to extract the heart rate features of the target human body based on the signal segment; and the third path is used to extract the respiratory features of the target human body based on the signal segment. The input of the attention layer within the pathway is connected to the output of the residual block with shared weights, and is used to weight the waveform features, the heart rate features, and the respiratory features. The input end of the inter-path attention layer is connected to the output end of the intra-path attention layer, and is used to fuse the weighted waveform features, heart rate features, and respiratory features; The input of the feature splicing layer is connected to the output of the inter-path attention layer, and is used to integrate the fused features into the joint feature tensor.

4. The method according to claim 3, characterized in that, The method involves dividing the human body monitoring signal from the target monitoring location into multiple signal segments in parallel based on predetermined different time scales, including: According to a first preset duration, the human body monitoring signal at the target monitoring location is divided into several first signal segments and input to the first channel; According to the second preset duration, the human body monitoring signal at the target monitoring location is divided into several second signal segments and input to the second channel; According to the third preset duration, the human body monitoring signal at the target monitoring location is divided into several third signal segments and input to the third channel; wherein, the first preset duration is shorter than the second preset duration, and the second preset duration is shorter than the third preset duration.

5. The method according to claim 4, characterized in that, The blood pressure prediction model was trained through the following steps: Acquire training data, which includes several batches of sample signal segments and corresponding sample time-series prediction results; each batch of sample signal segments includes a first sample signal segment, a second sample signal segment, and a third sample signal segment; The training data is input into the initial blood pressure prediction model in batches to obtain the training results; the model structure of the initial blood pressure prediction model is the same as that of the blood pressure prediction model. The training results and the sample time-series prediction results are input into a preset loss function to obtain the loss value; The initial blood pressure prediction model is iteratively trained using the loss value to obtain the blood pressure prediction model.

6. The method according to claim 1, characterized in that, The vascular feature extraction network was trained through the following steps: Acquire training data, which includes sample pulse wave signals and corresponding monitoring signals; the monitoring signals include sample cuff pressure signals synchronously acquired by the pressure-adjustable blood pressure monitor during the acquisition of the sample pulse wave signals. The sample pulse wave signal is input into the initial feature extraction network to obtain the training result; The training results and the supervision signal are input into a preset loss function to obtain the loss value; The feature extraction network is iteratively trained using the loss value until the model converges, thus obtaining the blood vessel feature extraction network.

7. The method according to claim 1, characterized in that, The human body monitoring signal is obtained by mixing the transmitted signal of the millimeter-wave device with the reflected signal corresponding to the transmitted signal to obtain an intermediate frequency signal, and then processing it through a fast Fourier transform. The step of calculating the periodic intensity of the signal at different monitoring distance units based on the human body monitoring signal, and determining the distance unit with the largest periodic intensity as the target monitoring location, includes: Based on each peak value in the human body monitoring signal, a distance unit is determined and the intermediate frequency signal is mapped to each distance unit; Based on the signal corresponding to each distance cell and the preset autocorrelation function, the autocorrelation function value corresponding to each distance cell is calculated; the magnitude of the autocorrelation function value is positively correlated with the period intensity. According to the preset heart rate range, the distance cell corresponding to the maximum autocorrelation function value is found from the distance cells and used as the target distance cell; The target monitoring location is determined based on the target distance unit.

8. The method according to claim 1, characterized in that, The step of processing the phase change signal to extract the pulse wave signal characterizing arterial pulsation includes: The phase-change signal is bandpass filtered to remove noise; The pulse wave signal is obtained by performing wavelet transform on the noise-removed reciprocal signal.

9. A non-contact continuous blood pressure monitoring device based on single calibration, comprising a processor, a memory, and a computer program / instructions stored in the memory, characterized in that, The processor is used to execute the computer program / instructions, and when the computer program / instructions are executed, the system implements the steps of the method as described in any one of claims 1 to 8; the device is communicatively connected to a millimeter-wave device to receive human body monitoring signals of the target human body collected in real time by the millimeter-wave device; the device is communicatively connected to a pressure blood pressure monitor to receive cuff pressure signals collected in real time by the pressure blood pressure monitor.

10. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method as described in any one of claims 1 to 8.

Images