Pulse signal data processing method and device based on multi-stage pressurization

Abstract

The application discloses a pulse signal data processing method and device based on multi-stage pressurization, and relates to the technical field of wearable medical devices. The method comprises the following steps: acquiring pulse signal data; extracting intrinsic mode functions for representing different frequency components of the pulse signal from the pulse signal data; screening target intrinsic mode functions from the intrinsic mode functions, and then reconstructing the screened target intrinsic mode functions into target pulse signal data; acquiring cycle data corresponding to the target pulse signal data, and extracting time-frequency features corresponding to the cycle data; inputting the cycle data and the time-frequency features into a neural network model, and determining a pulse condition classification result according to prior knowledge formed by the neural network model in a training stage. The application solves the technical problem of low signal data processing efficiency caused by the fact that the processing of pulse signal data in the prior art mainly depends on artificial experience judgment.

Description

Technical Field

[0001] This application relates to the field of wearable medical device technology, and more specifically, to a method and apparatus for processing pulse signal data based on multi-stage pressure. Background Technology

[0002] With the development of modern sensing technology, although electronic pulse diagnosis devices capable of collecting pulse signals have emerged, current technologies still generally follow a manual intervention process when analyzing the collected pulse signal data: professionals must manually annotate the pulse cycle, identify peaks and troughs, extract time-frequency features, and compare and classify them against literature. This process is not only time-consuming, labor-intensive, and highly repetitive, but also prone to low efficiency in pulse signal processing due to inconsistent subjective judgment standards.

[0003] There is currently no effective solution to the above problems. Summary of the Invention

[0004] This application provides a pulse signal data processing method and apparatus based on multi-stage pressure, which at least solves the technical problem in the prior art that the processing of pulse signal data mainly relies on human experience judgment, resulting in low signal data processing efficiency.

[0005] According to one aspect of the embodiments of this application, a pulse signal data processing method based on multi-stage pressure is provided, comprising: acquiring pulse signal data collected at multiple preset pressure levels; extracting intrinsic mode functions (IMFs) to characterize different frequency components of the pulse signal for each pressure level; selecting target IMFs from the IMFs whose contribution to the reconstructed pulse signal is greater than a preset threshold, and then reconstructing the selected target IMFs into target pulse signal data; acquiring periodic data corresponding to the target pulse signal data and extracting time-frequency features corresponding to the periodic data; inputting the periodic data and time-frequency features into a neural network model, and determining the pulse pattern classification result based on the prior knowledge formed by the neural network model during the training phase, wherein the pulse pattern classification result includes probability distribution information of the pulse signal data belonging to various pulse patterns.

[0006] Optionally, for each pressure level, the intrinsic mode functions (EMFs) used to characterize different frequency components of the pulse signal are extracted from the pulse signal data. This includes: decomposing the pulse signal data into multiple EMFs and a residual component according to preset conditions, wherein the residual component is the remaining trend term after the pulse signal data decomposition, used to characterize the changing trend of the pulse signal data; the EMFs are signal components that satisfy preset conditions, wherein the preset conditions include: the number of extreme points and the number of zero crossings in the signal sequence corresponding to the pulse signal data are equal or differ by no more than one, and the local mean of the envelope determined by the local maxima and the envelope determined by the local minima is zero.

[0007] Optionally, selecting target intrinsic mode functions (IMFs) from the IMFs whose contribution to the reconstructed pulse signal is greater than a preset threshold includes: obtaining the power spectrum of the IMFs; determining the frequency component with the highest power in the IMFs based on the power spectrum of the IMFs; accumulating the power of the IMFs in descending order of the power corresponding to the frequency component with the highest power to obtain the cumulative power contribution rate of the IMFs, wherein the cumulative power contribution rate of the IMFs is equal to the contribution of the IMFs to the reconstructed pulse signal; and determining the IMFs with a cumulative power contribution rate exceeding the preset threshold as target IMFs.

[0008] Optionally, after acquiring pulse signal data collected at multiple preset pressure levels, the method further includes: acquiring the airbag pressure value corresponding to each pressure level, wherein the airbag pressure value is the internal pressure value of the airbag obtained in real time by an airbag pressure sensor; and performing a correction operation on the pulse signal data based on the airbag pressure value, wherein the correction operation is used to adjust the pulse signal data at each pressure level to the same pressure reference.

[0009] Optionally, obtaining the periodic data corresponding to the target pulse signal data includes: in the time series corresponding to the target pulse signal data, sequentially comparing the amplitude of each data point with the amplitude of its adjacent data points; determining the data points with amplitudes greater than the two adjacent data points as peak points, and the data points with amplitudes less than the two adjacent data points as trough points; and determining the periodic data corresponding to the target pulse signal data based on the detected peak points and trough points.

[0010] Optionally, the periodic data corresponding to the target pulse signal data is determined based on the detected peak points and trough points, including: removing pseudo-peak points and pseudo-trough points that do not conform to the morphological characteristics of the pulse signal from the peak points and trough points, and taking the remaining peak points and trough points as the target peak points and target trough points; and determining multiple periodic data corresponding to the target pulse signal data based on the target peak points and target trough points.

[0011] Optionally, determining the periodic data corresponding to the target pulse signal data based on the target peak point and the target trough point further includes: arranging the target peak point and the target trough point in chronological order to form a peak sequence and a trough sequence; sequentially traversing each target peak point in the peak sequence, and finding the first trough point appearing after the target peak point in the trough sequence as the matching trough point for that target peak point; and combining the matching target peak points and target trough points in chronological order to form periodic data.

[0012] Optionally, the method further includes: if no target trough point is detected after the i-th target peak point, then the i-th target peak point is removed from the peak sequence, where i is an integer greater than or equal to 1; if no target peak point is detected before the j-th target trough point, then the j-th target trough point is removed from the trough sequence, where j is an integer greater than or equal to 1.

[0013] Optionally, pseudo-peaks and pseudo-valleys that do not conform to the morphological characteristics of the pulse signal are removed from the peaks and valleys. This includes: based on the morphological characteristics of the pulse waveform, for each peak, calculating the amplitude difference between the peak and the adjacent valley; if the amplitude difference is less than a preset amplitude threshold, then the peak and the corresponding adjacent valley are removed as pseudo-peaks and pseudo-valleys; detecting the time interval between adjacent peaks; if the time interval is less than a preset time threshold, then the peak with the largest amplitude among the adjacent peaks is selected and retained, and the remaining peaks are removed.

[0014] Optionally, extracting time-frequency features corresponding to the periodic data includes: extracting at least one of the following features from the pulse waveform corresponding to the periodic data: maximum rising slope, rising time, target amplitude, relative amplitude, and rising branch area as time-frequency features; wherein, the maximum rising slope represents the maximum slope of the rising segment of the pulse waveform, the rising time represents the time required for the pulse waveform to rise from the trough to the peak, the target amplitude represents the amplitude height of the pulse waveform, the relative amplitude represents the ratio of the target amplitude to a preset reference value, and the rising branch area represents the area between the rising segment of the pulse waveform and the baseline.

[0015] Optionally, inputting periodic data and time-frequency features into a neural network model and determining the pulse classification result based on the prior knowledge formed by the neural network model during the training phase further includes: performing linear interpolation on the periodic data to unify the number of data points for each complete periodic data, wherein linear interpolation is used to increase the number of data points by performing linear fitting based on adjacent data points, and unifying the number of data points of different periodic data to a target number; determining the average waveform of the periodic data after linear interpolation; using the average waveform as a feature curve; and inputting the feature curve and time-frequency features into the neural network model and determining the pulse classification result based on the prior knowledge formed by the neural network model during the training phase.

[0016] Optionally, the periodic data and time-frequency features are input into the neural network model, and the pulse classification result is determined based on the prior knowledge formed by the neural network model during the training phase. This includes: determining heart rate variability features based on the target pulse signal data, wherein the heart rate variability features include at least one of statistical features used to describe the degree of heart rate fluctuation and frequency domain features used to describe the spectral distribution of heart rate fluctuation; and inputting the periodic data, time-frequency features, and heart rate variability features into the neural network model, and determining the pulse classification result based on the prior knowledge formed by the neural network model during the training phase.

[0017] Optionally, after determining the pulse classification result based on the prior knowledge formed during the training phase of the neural network model, the method further includes: acquiring blood pressure data measured by oscillometric method using an airbag pressure sensor, wherein the oscillometric method is a method of measuring blood pressure by detecting airbag pressure oscillation waves; and correcting the blood pressure data based on the pulse classification result and / or heart rate variability characteristics.

[0018] Optionally, pulse signal data and period data are not normalized during feature extraction to preserve the amplitude information of the pulse signal data.

[0019] According to another aspect of the embodiments of this application, a blood pressure pulse diagnosis device is also provided, including: an airbag wristband for wearing on a user's wrist, with multiple thin-film pressure sensors disposed on the inner side of the airbag wristband; an air pump assembly connected to the airbag wristband for multi-stage inflation and deflation of the airbag wristband; an airbag pressure sensor for real-time detection of the airbag pressure inside the airbag wristband; and a control module connected to the air pump assembly, the airbag pressure sensor, and the multiple thin-film pressure sensors, for performing the following steps: acquiring pulse signal data collected at multiple preset pressure levels; extracting intrinsic mode functions (IMFs) to characterize different frequency components of the pulse signal for the pulse signal data at each pressure level; selecting target IMFs from the IMFs whose contribution to the reconstructed pulse signal is greater than a preset threshold, and then reconstructing the selected target IMFs into target pulse signal data; acquiring periodic data corresponding to the target pulse signal data and extracting time-frequency features corresponding to the periodic data; inputting the periodic data and time-frequency features into a neural network model, and determining the pulse classification result based on the prior knowledge formed by the neural network model during the training phase.

[0020] Optionally, an airbag is also provided on the inner side of the airbag wristband, and a thin-film pressure sensor is provided on the surface of the airbag near the wrist to collect pulse signals when the airbag is inflated and fits against the user's wrist.

[0021] Optionally, the multiple thin-film pressure sensors include four thin-film pressure sensors, corresponding to the cun, guan, chi, and posterior chi positions, respectively, for acquiring pulse signals at the cun, guan, chi, and posterior chi positions.

[0022] Optionally, the air pump assembly includes a diaphragm pump, and the multi-stage inflation includes multiple preset pressure levels, maintaining a fixed pressure for a preset duration at each pressure level.

[0023] Optionally, the blood pressure pulse diagnosis device also includes a signal amplification module. The input end of the signal amplification module is connected to multiple thin-film pressure sensors, and the output end is connected to the control module. The signal amplification module is used to amplify the pulse signal collected by the thin-film pressure sensors and transmit the amplified pulse signal to the control module.

[0024] Optionally, the control module includes a microcontroller, a power supply, and a communication module. The microcontroller is connected to the air pump assembly, the airbag pressure sensor, multiple thin-film pressure sensors, and the communication module, respectively. The communication module includes a wireless communication module or a wired communication interface for sending the collected data or the determined pulse classification results to the terminal device.

[0025] In this embodiment, pulse signal data is collected in stages under multi-stage pressure. Empirical mode decomposition is performed on the original pulse signal at each pressure level to extract its intrinsic mode function. Target intrinsic mode functions with a cumulative contribution greater than a preset threshold are selected to reconstruct the target pulse signal data. This achieves the goal of preserving the physiologically significant dominant frequency components in the pulse signal and avoiding the loss of amplitude and morphological features due to global normalization or over-filtering. Thus, it achieves the technical effect of obtaining a higher quality pulse waveform without losing the original signal strength information, thereby solving the technical problem of low signal data processing efficiency in the prior art, which mainly relies on manual experience judgment for pulse signal data processing. Attached Figure Description

[0026] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0027] Figure 1 This is a flowchart of an optional pulse signal data processing method based on multi-stage pressurization according to an embodiment of this application;

[0028] Figure 2 This is a schematic diagram of an optional blood pressure pulse diagnosis device according to an embodiment of this application;

[0029] Figure 3 This is a schematic diagram of an optional arrangement of pressure sensors inside the airbag according to an embodiment of this application;

[0030] Figure 4 This is a schematic diagram of an optional wrist-type blood pressure pulse diagnosis device according to an embodiment of this application. Detailed Implementation

[0031] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0032] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0033] According to an embodiment of this application, a method embodiment for pulse signal data processing based on multi-stage pressure is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0034] According to the embodiments of this application, a pulse signal data processing system based on multi-stage pressurization (hereinafter referred to as the system) can be used as the execution subject of the pulse signal data processing method based on multi-stage pressurization in the embodiments of this application. The pulse signal data processing system based on multi-stage pressurization can be a software system or an embedded system combining software and hardware. Of course, the execution subject of the method in the embodiments of this application can also be other forms of execution subject, such as devices or equipment. Those skilled in the art should know that this application does not particularly limit the specific form of the execution subject of the method.

[0035] Figure 1This is a flowchart of an optional pulse signal data processing method based on multi-stage inflation according to an embodiment of this application, such as... Figure 1 As shown, the method includes the following steps:

[0036] Step S101: Acquire pulse signal data collected at multiple preset pressure levels.

[0037] Optionally, the preset multiple different pressure levels refer to the system applying different sizes of airbag pressure to the wrist in stages according to a preset sequence during the measurement process. For example, it can be set to 6 pressure gradients, gradually increasing from light pressure to heavy pressure, with each pressure value corresponding to a data acquisition stage.

[0038] Optionally, after the system starts, the miniature diaphragm pump sequentially inflates the airbag according to the program. Each time a preset pressure value is reached, that pressure is maintained for 15 seconds. During this period, four thin-film sensors continuously record pulse fluctuation signals at the same sampling rate. A set of raw signal data is independently saved for each pressure level, forming a total of 6 sets of time-series data, each corresponding to a different pressure environment. This ensures that the acquired signals can accurately reflect the dynamic response of the pulse signal as pressure changes, avoiding the loss of deep or superficial pulse signals due to fixed pressure at a single point.

[0039] Step S102: Extract intrinsic mode functions from the pulse signal data at each pressure level to characterize different frequency components of the pulse signal.

[0040] Optionally, the intrinsic mode function is an oscillation mode adaptively separated from the original non-stationary signal through an empirical mode decomposition algorithm. Each intrinsic mode function represents a local fluctuation component in a specific frequency range of the signal. Its frequency and amplitude change with time, and it can capture complex nonlinear characteristics in the pulse wave, such as the main wave, diphtheria wave, clutter interference, etc.

[0041] Optionally, the raw pulse signal at each pressure level is decomposed using an empirical mode decomposition (EMD) algorithm. The EMD algorithm iterates by continuously extracting local extrema of the signal, constructing upper and lower envelopes, and calculating their mean values ​​until components satisfying the definition of intrinsic mode functions (EMFs) are obtained. The signal at each pressure level is decomposed into multiple EEMs, corresponding to different physiological or interference components such as high-frequency noise, mid-frequency pulse wave, and low-frequency respiratory interference.

[0042] Optionally, the empirical mode decomposition algorithm does not require a preset frequency range and can accurately reproduce the non-periodic and nonlinear fluctuation characteristics of the pulse signal.

[0043] Step S103: Select target intrinsic mode functions from the intrinsic mode functions whose contribution to the reconstructed pulse signal is greater than a preset threshold, and then reconstruct the selected target intrinsic mode functions into target pulse signal data.

[0044] Optionally, the contribution refers to the proportion of energy included in a certain intrinsic mode function to the total energy of the original signal, which is obtained by calculating the ratio of the square integral value of the intrinsic mode function to the total power of the original signal.

[0045] Optionally, the target intrinsic mode function is a set of intrinsic mode functions whose contribution exceeds a set threshold, representing the main physiological components in the pulse signal.

[0046] Optionally, the target pulse signal data is a denoised pulse waveform that retains core features after the key intrinsic mode functions are re-superimposed and synthesized.

[0047] Optionally, the above steps can remove sensor noise, electromyographic interference, and respiratory baseline drift, while retaining key morphological information such as the main pulse wave, rising limb, and dicrotic wave, avoiding distortion of pulse strength information caused by direct normalization or threshold filtering.

[0048] Step S104: Obtain the periodic data corresponding to the target pulse signal data, and extract the time-frequency features corresponding to the periodic data.

[0049] Optionally, time-frequency characteristics refer to quantifiable parameters in the time and frequency domains, such as the rising slope of each cycle, i.e., the rate of pressure change per unit time, the amplitude of the main wave, i.e., the difference between the peak and the trough, the area under the rising branch, i.e., the area under the curve before the peak, the cycle length, etc., thereby reflecting the strength, speed, slowness, and smoothness of the pulse signal.

[0050] Optionally, the system first employs a peak detection algorithm based on local maximum search to identify all peaks and troughs in the target pulse signal. Then, outliers are removed using peak-trough pairing rules, forming a continuous and aligned periodic sequence. Linear interpolation is performed on each period, standardizing the sampling points to 400 to eliminate individual rhythm differences. Subsequently, the time-frequency characteristics of each period, such as the rising slope, main wave amplitude, and rising support area, are calculated to form a fixed-length feature vector.

[0051] Optionally, the above steps transform the originally continuous pulse waveforms of varying lengths into standardized, comparable feature vectors, making the pulse data of different pressure levels and different subjects comparable.

[0052] Step S105: Input the periodic data and time-frequency features into the neural network model, and determine the pulse classification result based on the prior knowledge formed by the neural network model during the training phase. The pulse classification result includes the probability distribution information of the pulse signal data belonging to various pulse types.

[0053] Optionally, the neural network model is a deep learning model based on the Transformer architecture, which has learned a large number of labeled samples during the training phase, such as the periodic data and feature combinations corresponding to 28 types of pulse patterns.

[0054] Optionally, prior knowledge refers to the mapping relationship between pulse patterns and features that the model automatically learns from the data during the training process.

[0055] Optionally, the probability distribution information represents the confidence level corresponding to each pulse category in the model output.

[0056] Alternatively, the structure of the neural network model is as follows:

[0057] 1. Input Layer: Multimodal Feature Fusion Input. The model receives three types of feature inputs: periodic data and time-frequency features. All inputs retain the absolute amplitude of the physical signal without any normalization or standardization processing, in order to fully preserve the various key amplitude information upon which the pulse depends. The three types of features are:

[0058] Average waveform characteristics: A representative waveform obtained by aligning and averaging multiple complete pulse cycles, reflecting the overall morphological trend of the pulse;

[0059] Time-frequency characteristics: Dynamic indicators related to the cardiac cycle extracted from the pulse waveform, such as rise slope, rise time, target amplitude, relative amplitude, and rise branch area, are used to quantify the speed, strength, and energy distribution of the pulse.

[0060] Heart rate variability characteristics: physiological parameters that reflect the state of the heart's autonomous regulation of rhythm, including statistical characteristics describing the degree of fluctuation in heart rate intervals and frequency domain characteristics describing the spectral distribution of fluctuations.

[0061] The three types of features are concatenated into a unified composite vector at the input layer, which serves as the initial input to the model.

[0062] 2. Feature Embedding Layer: To ensure comparability and synergy among different types of features in the semantic space, the model uses a linear transformation layer to map composite vectors to a unified semantic embedding space. This process does not change the physical meaning of the features; it only completes dimensionality adaptation and semantic encoding, facilitating deeper interaction modeling in subsequent modules.

[0063] 3. Location Encoding Layer: To enable the model to perceive the relative role of each input feature in the overall diagnostic logic, such as the average waveform as the main body, time-frequency features as supplementary features, and heart rate variability features as the adjustment background, a location encoding mechanism is introduced.

[0064] 4. Transformer Encoder Module: The core of the model is a multi-layered stacked Transformer encoder structure, with each layer containing the following sub-modules:

[0065] Multi-head self-attention mechanism: enables the model to dynamically calculate the interdependencies between different dimensions of the input features.

[0066] Feedforward neural networks: perform nonlinear transformations on the attention output to enhance the model's ability to express complex nonlinear relationships.

[0067] Residual connections and layer normalization contribute to the stability of deep network training, prevent gradient vanishing, and improve convergence efficiency.

[0068] 5. Classification Output Layer: Dual-task structure. The first output branch is used to identify six major categories of pulse patterns, and the output is a probability distribution of a category. The second output branch is used to identify 28 specific subcategories of pulse patterns, and the output is a probability distribution of the subcategories of pulse patterns.

[0069] In one optional embodiment, extracting intrinsic mode functions (IMFs) to characterize different frequency components of the pulse signal for each pressure level includes: decomposing the pulse signal data into multiple IMFs and a residual component according to preset conditions, wherein the residual component is the remaining trend term after decomposition of the pulse signal data, used to characterize the changing trend of the pulse signal data; the IMFs are signal components that satisfy preset conditions, wherein the preset conditions include: the number of extreme points and the number of zero crossings in the signal sequence corresponding to the pulse signal data are equal or differ by no more than one, and the local mean of the envelope determined by the local maxima and the envelope determined by the local minima is zero.

[0070] Optionally, the residual component is the part remaining after the empirical mode decomposition is completed, representing the long-term trend or low-frequency drift of the original signal, such as slow changes caused by respiration, body position changes or sensor temperature drift, excluding the periodic information of the pulse.

[0071] Optionally, the system performs empirical mode decomposition on the raw pulse signal at each pressure level: First, it identifies all local maxima and local minima in the signal sequence and generates upper and lower envelopes using cubic spline interpolation. Next, it calculates the average of the upper and lower envelopes, i.e., the local mean, and randomly determines whether this local mean is close to zero. If not, it subtracts this mean from the original signal to obtain a new sequence. This process is repeated until the sequence satisfies the condition that the number of extreme points is equal to or differs from the number of zero-crossing points by no more than one. The resulting sequence is an intrinsic mode function. Then, the above process is repeated on the remaining signal until the residual signal no longer contains decomposable oscillatory components. The final remaining part is the residual component.

[0072] Optionally, during the entire decomposition process, each intrinsic mode function is conditionally screened, which helps to separate the physiological components directly related to the pulse, eliminate slowly changing residual components, and avoid interfering with subsequent period detection and feature extraction.

[0073] In one optional embodiment, selecting target intrinsic mode functions (IMFs) from the IMFs whose contribution to the reconstructed pulse signal is greater than a preset threshold includes: obtaining the power spectrum of the IMFs; determining the frequency component with the highest power in the IMFs based on the power spectrum of the IMFs; accumulating the power of the IMFs in descending order of the power corresponding to the frequency component with the highest power to obtain the cumulative power contribution rate of the IMFs, wherein the cumulative power contribution rate of the IMFs is equal to the contribution of the IMFs to the reconstructed pulse signal; and determining the IMFs with a cumulative power contribution rate exceeding the preset threshold as target IMFs.

[0074] Optionally, the power spectrum refers to the energy distribution of the intrinsic mode functions (IMFs) in the frequency domain. It is obtained by performing a Fourier transform on each IMF and calculating the square of its amplitude. It is used to quantify the energy magnitude of the components separated from the original pulse signal at different frequencies. The frequency at which the peak of the power spectrum occurs is the dominant frequency of that component. The cumulative power contribution rate refers to the percentage of power contributed by the IMFs to the overall energy of the original signal, starting from the IMF with the highest energy and accumulating the power values ​​of each component in descending order until a certain proportion of the total energy is reached.

[0075] Optionally, for all intrinsic mode functions (IMFs) obtained from decomposition at each pressure level, the power spectrum of each IMF is first calculated. For each power spectrum, the frequency component corresponding to the maximum power is determined as the dominant frequency, used to assist in judging physiological relevance, but not directly used for screening. Subsequently, all IMFs are sorted from high to low total power. Starting with the IMF with the highest power, its power value is accumulated one by one, and the percentage of the current accumulated power to the total power of the original pulse signal is calculated, which is the power accumulation contribution rate. When the accumulation contribution rate first exceeds a preset threshold, the accumulation stops, and all previously selected IMFs are defined as target IMFs. For example, if the accumulation contribution rate of the first five IMFs reaches 92% (assuming the preset threshold is 90%), only these five are retained as target IMFs, and the remaining low-power components are discarded. Finally, the target IMFs are directly superimposed to reconstruct the target pulse signal data for subsequent cycle recognition and feature extraction.

[0076] In one optional embodiment, after acquiring pulse signal data collected at multiple preset pressure levels, the method further includes: acquiring the airbag pressure value corresponding to each pressure level, wherein the airbag pressure value is the internal pressure value of the airbag obtained in real time by an airbag pressure sensor; and performing a correction operation on the pulse signal data based on the airbag pressure value, wherein the correction operation is used to adjust the pulse signal data at each pressure level to the same pressure reference.

[0077] Optionally, the airbag pressure value refers to the gas pressure value inside the airbag that is collected in real time by the airbag pressure sensor installed inside the airbag during each pressure level maintenance period.

[0078] Optionally, the correction operation uses the airbag pressure value as a reference and linearly shifts the pulse signal data collected at each pressure level, adjusting the pulse signals of all pressure levels to the same reference pressure benchmark. This reduces signal offset caused by errors in the inflation step length, airbag elastic hysteresis, or individual differences in wrist tissue stiffness, facilitating lateral comparison of pulse morphology at different pressure levels. Note that this correction is not normalization, but rather alignment based on physical pressure. The key lies in the relative change of pulse amplitude when pressure changes. If the airbag pressure is not accurately recorded, even if the pulse signal is accurately collected, it is impossible to determine which pulse type label the signal corresponds to.

[0079] Optionally, for the pulse signal data at each pressure level, i.e., the original pressure sequence, the pulse signal data is bound to the corresponding actual airbag pressure value to form a pressure-pulse signal pair. Then, a unified target pressure benchmark is selected. For the pulse signal at each pressure level, the difference between the actual airbag pressure value and the target benchmark is calculated, and linear offset compensation, i.e., correction, is performed on the pulse signal based on this difference. After correction, the pulse signals at all pressure levels are equivalent to those acquired under the unified target pressure benchmark, forming a set of pressure benchmark-aligned pulse signal sequences for intrinsic mode function decomposition, period extraction, and classification.

[0080] In one optional embodiment, obtaining the periodic data corresponding to the target pulse signal data includes: in the time series corresponding to the target pulse signal data, sequentially comparing the amplitude of each data point with the amplitude of its adjacent data points; determining the data points with amplitudes greater than the two adjacent data points as peak points, and the data points with amplitudes less than the two adjacent data points as trough points; and determining the periodic data corresponding to the target pulse signal data based on the detected peak points and trough points.

[0081] Optionally, the target pulse signal data is used as input. Starting from the second data point, the process iterates sequentially to the second-to-last data point, comparing the amplitude of each intermediate point with the amplitude of the preceding and following points: if the amplitude of the intermediate point is greater than the amplitude of the preceding point and greater than the amplitude of the following point, it is marked as a peak point; if the amplitude of the intermediate point is less than the amplitude of the preceding point and less than the amplitude of the following point, it is marked as a trough point; if none of the above conditions are met, it is ignored. All detected peak points and trough points are arranged in chronological order to form an alternating sequence, thus constituting a complete periodic data.

[0082] In one optional embodiment, determining the periodic data corresponding to the target pulse signal data based on the detected peak points and trough points includes: removing pseudo-peak points and pseudo-trough points that do not conform to the morphological characteristics of the pulse signal from the peak points and trough points, and using the remaining peak points and trough points as target peak points and target trough points; and determining multiple periodic data corresponding to the target pulse signal data based on the target peak points and target trough points.

[0083] Optionally, for each peak and trough point, pulse signal morphological features are sequentially applied for matching and judgment. The criteria for these features include minimum amplitude difference, minimum cycle duration, reasonableness of the rising / falling slope, consistency of the diphthoplasty, and waveform continuity. If any criterion is not met, the point is marked as a false point and discarded. Only points that meet all the feature criteria are retained as target peaks and target troughs. The selected target peaks and target troughs are then arranged in chronological order to form an alternating sequence, constituting multiple periods of data.

[0084] Optionally, the system introduces a pseudopoint elimination mechanism based on pulse morphology features to address the shortcomings of existing pulse diagnosis equipment that rely solely on mathematical extreme value detection and neglect physiological rationality.

[0085] In an optional embodiment, determining the periodic data corresponding to the target pulse signal data based on the target peak point and the target trough point further includes: arranging the target peak point and the target trough point in chronological order to form a peak sequence and a trough sequence; sequentially traversing each target peak point in the peak sequence, and finding the first trough point appearing after the target peak point in the trough sequence as the matching trough of the target peak point; and combining the matching target peak points and target trough points in chronological order to form periodic data.

[0086] Optionally, the system arranges all retained target peaks in the order of their appearance in the time series, forming an ordered list: Target Peak Sequence = [P1, P2, P3, ..., P n ], where P1 <P2<...<P n(Incrementing by time). Similarly, all target trough points are arranged in chronological order as follows: Target trough sequence = [V1, V2, V3, ..., V...]. m ], where V1 <V2<...<V m Iterate through each point P in the target wave crest sequence. i (i from 1 to n), in the target trough sequence, starting from the first time greater than P i Start the search from the time point P, and find the first time point located at P. i The subsequent trough point V j And satisfy V j >P i If found, then pair Pi with V. j A valid match is identified, forming a periodic data point; if no match is found, i.e., P... i If there is no subsequent trough, the peak is skipped and considered invalid or a residual at the end. All successfully matched "target peak-target trough" pairs are arranged in chronological order to form a continuous, non-overlapping, and non-intersecting set of periodic data.

[0087] Optionally, the system uses a time sequence matching mechanism to achieve accurate, orderly, one-to-one pairing of target peak points and target trough points, thus solving the common mismatch and omission phenomena in pulse signal cycle extraction.

[0088] In an optional embodiment, the pulse signal data processing method based on multi-stage pressure further includes: if no target trough point is detected after the i-th target peak point, then the i-th target peak point is removed from the peak sequence, where i is an integer greater than or equal to 1; if no target peak point is detected before the j-th target trough point, then the j-th target trough point is removed from the trough sequence, where j is an integer greater than or equal to 1.

[0089] Optionally, if no target trough appears after the i-th target peak, it indicates that the target peak lacks a corresponding descent process and is a false signal caused by anomalies or noise; therefore, this target peak is removed from the peak sequence. Similarly, if there is no valid target peak before the j-th target trough, it indicates that the target trough lacks a corresponding rise process, which does not conform to the basic physiological law of pulse rise and fall, and is also an invalid point, thus being removed as well. Through this removal operation, the system retains only peaks and troughs with corresponding relationships before and after them, helping each cycle to consist of complete, paired pulses, reducing misjudgments of pulse types, and improving the accuracy of blood pressure calibration.

[0090] In one optional embodiment, pseudo-peaks and pseudo-valleys that do not conform to the morphological characteristics of the pulse signal are removed from the peaks and valleys. This includes: based on the morphological characteristics of the pulse waveform, for each peak, calculating the amplitude difference between the peak and the adjacent valley; if the amplitude difference is less than a preset amplitude threshold, then the peak and the corresponding adjacent valley are removed as pseudo-peaks and pseudo-valleys; detecting the time interval between adjacent peaks; if the time interval is less than a preset time threshold, then the peak with the largest amplitude among the adjacent peaks is selected and retained, and the remaining peaks are removed.

[0091] Optionally, for each peak, the amplitude difference between the peak and the nearest trough is calculated. If the amplitude difference is less than the preset amplitude threshold, it means that the change between the peak and the trough is too weak and does not conform to the strength and rhythm of the pulsation. It may only be a pseudo signal formed by muscle tremor or electromagnetic interference. Therefore, the peak and trough are removed together.

[0092] Secondly, the time interval between two adjacent peaks is detected. If two peaks appear too close together, it may be due to noise superposition or a single pulse being mistakenly split into two peaks. In this case, the peak with the largest amplitude is retained, while other peaks with smaller amplitudes and similar positions are removed. This preserves the main characteristics of the pulse while filtering out redundant points caused by signal jitter or misjudgment of diphthoplasty. Ultimately, the retained peaks and troughs conform to the pulse-related characteristics in terms of amplitude, timing, and morphology, improving the stability and diagnostic reliability of the pulse diagnostic instrument in complex operating environments.

[0093] In one optional embodiment, extracting time-frequency features corresponding to the periodic data includes: extracting at least one of the following features from the pulse waveform corresponding to the periodic data: maximum rising slope, rising time, target amplitude, relative amplitude, and rising branch area as time-frequency features; wherein, the maximum rising slope represents the maximum slope of the rising segment of the pulse waveform, the rising time represents the time required for the pulse waveform to rise from the trough to the peak, the target amplitude represents the amplitude height of the pulse waveform, the relative amplitude represents the ratio of the target amplitude to a preset reference value, and the rising branch area represents the area between the rising segment of the pulse waveform and the baseline.

[0094] Optionally, the maximum upward slope refers to the slope value corresponding to the segment with the largest upward rate of the curve during the process of the pulse waveform rising from the trough to the peak within one cycle.

[0095] Optionally, the rise time refers to the duration from the trough of the wave to the peak of the wave, which can reflect the speed of the pulsation.

[0096] Optionally, the target amplitude refers to the vertical distance between the peak and the corresponding trough, i.e., the absolute amplitude of the pulse, which reflects the strength of the pulse.

[0097] Optionally, the relative amplitude is the ratio of the target amplitude to the preset reference value of the current pressurization phase, used to eliminate amplitude drift caused by changes in airbag pressure.

[0098] Optionally, the rising limb area refers to the geometric area enclosed by the bottom of the curve and the horizontal baseline between the trough and the peak of the pulse waveform, reflecting the kinetic energy and energy accumulation process contained in the pulse.

[0099] Optionally, the above five types of time-frequency features are extracted without signal normalization, which helps to preserve the key information such as the strength, speed, and energy of the original pulsation.

[0100] In one optional embodiment, the periodic data and time-frequency features are input into a neural network model, and the pulse classification result is determined based on the prior knowledge formed by the neural network model during the training phase. The method further includes: performing linear interpolation on the periodic data to unify the number of data points for each complete periodic data set; wherein linear interpolation is used to increase the number of data points by performing linear fitting based on adjacent data points, thereby unifying the number of data points for different periodic data sets to a target number; determining the average waveform of the periodic data after linear interpolation; using the average waveform as a feature curve; and inputting the feature curve and time-frequency features into the neural network model, and determining the pulse classification result based on the prior knowledge formed by the neural network model during the training phase.

[0101] Optionally, before inputting the periodic data and time-frequency features into the neural network model, the system first performs linear interpolation. That is, for each filtered and paired complete periodic data point, based on its original sampling point sequence, intermediate values ​​are inserted proportionally between adjacent data points, so that each period is ultimately expanded to a fixed number of target data points, for example, a uniform total of four hundred points. This process does not change the physical trend of the waveform morphology; it only smoothly fills gaps through linear fitting. This helps align periods across different individuals, pressure levels, and pulse types in the time dimension, thereby reducing ambiguity in model input caused by inconsistent sampling lengths.

[0102] Optionally, after interpolation, the system aligns all periodic data belonging to the same pressure level along the time axis and calculates the average value point by point to form an average waveform representing the typical pulsation pattern under that pressure, which is used to reflect the common characteristics of most cycles under that pressure. Subsequently, the system uses the average waveform as a feature curve and together with the extracted time-frequency features to construct a multi-dimensional input vector, which is then fed into a neural network model trained with pulse pattern labeled data.

[0103] Optionally, the neural network model has learned the implicit correlation between the characteristic curve morphology and numerical features of various pulse types during the training phase. The model automatically determines the pulse type corresponding to the current input as one or a combination of preset pulse types based on the relevant features.

[0104] Optionally, the system introduces average waveform as the morphological backbone and time-frequency features as a quantization supplement, enabling the neural network to simultaneously capture the overall morphological trend and local dynamic parameters, thereby improving the accuracy and robustness of classification. In particular, it can maintain stable output even in practical application scenarios such as slight changes in patient position and poor sensor contact.

[0105] In one optional embodiment, periodic data and time-frequency features are input into a neural network model, and the pulse classification result is determined based on the prior knowledge formed by the neural network model during the training phase. This includes: determining heart rate variability features based on target pulse signal data, wherein the heart rate variability features include at least one of statistical features used to describe the degree of heart rate fluctuation and frequency domain features used to describe the spectral distribution of heart rate fluctuation; and inputting periodic data, time-frequency features, and heart rate variability features into a neural network model, and determining the pulse classification result based on the prior knowledge formed by the neural network model during the training phase.

[0106] Optionally, heart rate variability refers to quantitative indicators reflecting dynamic changes in cardiac rhythm obtained by analyzing the time intervals between adjacent pulse cycles. This characteristic does not reflect pulse strength or morphology, but rather reveals the neural regulatory background accompanying pulse formation. Specifically, it includes two categories:

[0107] One type is statistical characteristics, such as the population standard deviation, the root mean square of the difference between adjacent time intervals, and the percentage of time intervals exceeding 50 milliseconds. These indicators are used to characterize the overall variability in heart rate duration.

[0108] Another type is frequency domain features, such as total power, high-frequency power, low-frequency power, and the ratio of low-frequency to high-frequency power. The time interval sequence is decomposed into different frequency components by Fourier transform, which correspond to slow regulation dominated by the parasympathetic nervous system and medium-speed regulation dominated by the sympathetic nervous system, respectively.

[0109] Optionally, in this application, the system first calculates the time interval sequences from the periodic data that has been pseudo-point removed and paired, and then extracts at least one of statistical features and frequency domain features as a third type of input dimension. Subsequently, the system combines the periodic data, time-frequency features, and heart rate variability features to form a composite input vector, which is then input into the neural network model. By introducing heart rate variability features, the system overcomes the limitation of traditional pulse diagnosis relying solely on waveform morphology, and integrates pulse characteristics with cardiac rhythm regulation mechanisms at the algorithmic level, thereby improving the accuracy, anti-interference ability, and clinical interpretability of pulse classification.

[0110] In an optional embodiment, after determining the pulse classification result based on the prior knowledge formed during the training phase of the neural network model, the method further includes: acquiring blood pressure data measured by an oscillometric method using an airbag pressure sensor, wherein the oscillometric method is a method for measuring blood pressure by detecting airbag pressure oscillation waves; and correcting the blood pressure data based on the pulse classification result and / or heart rate variability characteristics.

[0111] Alternatively, the oscillometric method is a method to estimate systolic and diastolic blood pressure by detecting the change in the oscillation amplitude of the pressure inside the wristband over time. The principle is based on the fact that during the process of blood flow from complete blockage to restoration, the response intensity of the blood vessel wall to the wristband pressure fluctuations shows a trend of first increasing and then decreasing. The maximum oscillation point corresponds to the average pressure, and specific proportional points before and after correspond to systolic and diastolic blood pressure, respectively.

[0112] Optionally, given the known pulse classification results, the original blood pressure value output by the oscillometric method can be specifically corrected: for example, the system can adjust the original blood pressure data upward based on the blood pressure correction coefficient corresponding to the pulse in the training data to correct the underestimation caused by the weak pulse signal; or adjust it downward based on the negative correction coefficient corresponding to the pulse to avoid overestimation caused by excessively strong pulse.

[0113] In addition, the system further optimizes the correction accuracy by incorporating heart rate variability characteristics: if the root mean square of the difference between adjacent time intervals and the percentage of time interval differences exceeding 50 milliseconds are low, it indicates excessive sympathetic nerve activation and increased vascular tension, and the system will slightly increase the blood pressure value based on the pulse correction; if the overall standard deviation is high and the proportion of high-frequency power is large, it indicates vagal nerve dominance and vasodilation, and the system will appropriately decrease the blood pressure value. Finally, the system uses the corrected blood pressure value as the final output result, and simultaneously marks the pulse calibration status.

[0114] In one alternative embodiment, the pulse signal data and period data are not normalized during feature extraction to preserve the amplitude information of the pulse signal data.

[0115] Optionally, normalization refers to compressing or stretching the original signal by a uniform ratio so that all data fall within a certain fixed value range. Although normalization can improve the convergence speed of the model in general machine learning, it essentially erases the absolute amplitude information of the signal, causing "strong pulse" and "weak pulse" to lose their distinction in digital representation. Therefore, the system is prohibited from performing any form of normalization when extracting features from pulse signal data and the periodic data formed after processing.

[0116] Figure 2This is a schematic diagram of an optional blood pressure pulse diagnosis device according to an embodiment of this application. According to another embodiment of this application, a blood pressure pulse diagnosis device is also provided, including: an airbag wristband, an air pump assembly, an airbag pressure sensor, an airbag pressure data transmission line, and a control module.

[0117] The system comprises an airbag wristband, worn on the user's wrist, with multiple thin-film pressure sensors on its inner side; an air pump assembly, connected to the airbag wristband, for multi-stage inflation and deflation of the airbag wristband; an airbag pressure sensor for real-time detection of the airbag pressure within the airbag wristband; and a control module, connected to the air pump assembly, the airbag pressure sensor, and the multiple thin-film pressure sensors, for performing the following steps: acquiring pulse signal data collected at multiple preset pressure levels; extracting intrinsic mode functions (IMFs) to characterize different frequency components of the pulse signal for each pressure level; selecting target IMFs from the IMFs whose contribution to the reconstructed pulse signal is greater than a preset threshold, and then reconstructing the target IMFs into target pulse signal data; acquiring periodic data corresponding to the target pulse signal data and extracting the time-frequency features corresponding to the periodic data; and inputting the periodic data and time-frequency features into a neural network model, determining the pulse classification result based on the prior knowledge formed during the training phase of the neural network model.

[0118] Optionally, an airbag is also provided on the inner side of the airbag wristband, and a thin-film pressure sensor is provided on the surface of the airbag near the wrist to collect pulse signals when the airbag is inflated and fits against the user's wrist.

[0119] Optionally, the multiple thin-film pressure sensors include four thin-film pressure sensors, corresponding to the cun, guan, chi, and posterior chi positions, respectively, for acquiring pulse signals at the cun, guan, chi, and posterior chi positions.

[0120] Optionally, the air pump assembly includes a diaphragm pump, and the multi-stage inflation includes multiple preset pressure levels, maintaining a fixed pressure for a preset duration at each pressure level.

[0121] Optionally, the blood pressure pulse diagnosis device also includes a signal amplification module. The input end of the signal amplification module is connected to multiple thin-film pressure sensors, and the output end is connected to the control module. The signal amplification module is used to amplify the pulse signal collected by the thin-film pressure sensors and transmit the amplified pulse signal to the control module.

[0122] Optionally, the control module includes a microcontroller, a power supply, and a communication module. The microcontroller is connected to the air pump assembly, the airbag pressure sensor, multiple thin-film pressure sensors, and the communication module, respectively. The communication module includes a wireless communication module or a wired communication interface for sending the collected data or the determined pulse classification results to the terminal device.

[0123] Optionally, the control module is also used to: acquire the airbag pressure value corresponding to each pressure level, wherein the airbag pressure value is the internal pressure value of the airbag obtained in real time by the airbag pressure sensor; and perform a correction operation on the pulse signal data according to the airbag pressure value, wherein the correction operation is used to adjust the pulse signal data at each pressure level to the same pressure reference.

[0124] Optionally, the control module extracts intrinsic mode functions (IMFs) to characterize different frequency components of the pulse signal, including: decomposing the pulse signal data into multiple IMFs and a residual component according to preset conditions, wherein the residual component is the remaining trend term after the pulse signal data decomposition, used to characterize the changing trend of the pulse signal data; the IMFs are signal components that satisfy preset conditions, wherein the preset conditions include: the number of extreme points and the number of zero crossings in the signal sequence corresponding to the pulse signal data are equal or differ by no more than one, and the local mean of the envelope determined by the local maxima and the envelope determined by the local minima is zero.

[0125] Optionally, the control module selects target intrinsic mode functions (IMFs) from the IMFs whose contribution to the reconstructed pulse signal is greater than a preset threshold, including: obtaining the power spectrum of the IMFs; determining the frequency component with the highest power in the IMFs based on the power spectrum; accumulating the power of the IMFs in descending order of the power corresponding to the frequency component with the highest power to obtain the cumulative power contribution rate of the IMFs; and determining the IMFs with a cumulative power contribution rate exceeding the preset threshold as target IMFs.

[0126] Optionally, the control module acquires the periodic data corresponding to the target pulse signal data, including: in the time series corresponding to the target pulse signal data, sequentially comparing the amplitude of each data point with the amplitude of its adjacent data points; determining the data points with amplitudes greater than the two adjacent data points as peak points, and the data points with amplitudes less than the two adjacent data points as trough points; and determining the periodic data corresponding to the target pulse signal data based on the detected peak points and trough points.

[0127] Optionally, the control module determines the periodic data corresponding to the target pulse signal data based on the detected peaks and troughs, including: removing pseudo-peaks and pseudo-troughs that do not conform to the morphological characteristics of the pulse signal, and using the remaining peaks and troughs as target peaks and target troughs; and determining multiple periodic data corresponding to the target pulse signal data based on the target peaks and target troughs.

[0128] Optionally, the control module determines the periodic data corresponding to the target pulse signal data based on the target peak point and the target trough point, and further includes: arranging the target peak point and the target trough point in chronological order to form a peak sequence and a trough sequence; sequentially traversing each target peak point in the peak sequence, and finding the first trough point that appears after the target peak point in the trough sequence as the matching trough point for the target peak point; and combining the matching target peak points and target trough points in chronological order to form periodic data.

[0129] Optionally, the control module is further configured to: if no target trough point is detected after the i-th target peak point, remove the i-th target peak point from the peak sequence, where i is an integer greater than or equal to 1; if no target peak point is detected before the j-th target trough point, remove the j-th target trough point from the trough sequence, where j is an integer greater than or equal to 1.

[0130] Optionally, the control module removes pseudo-peaks and pseudo-valleys that do not conform to the morphological characteristics of the pulse signal from the peaks and valleys. This includes: based on the morphological characteristics of the pulse waveform, for each peak, calculating the amplitude difference between the peak and the adjacent valley; if the amplitude difference is less than a preset amplitude threshold, then the peak and the corresponding adjacent valley are removed as pseudo-peaks and pseudo-valleys; detecting the time interval between adjacent peaks; if the time interval is less than a preset time threshold, then the peak with the largest amplitude among the adjacent peaks is selected and retained, while the remaining peaks are removed.

[0131] Optionally, the control module extracts time-frequency features corresponding to the periodic data, including: extracting at least one of the following features from the pulse waveform corresponding to the periodic data: maximum rising slope, rising time, amplitude, relative amplitude, and rising branch area as time-domain features; wherein, the maximum rising slope represents the maximum slope of the rising segment of the pulse waveform, the rising time represents the time required for the pulse waveform to rise from the trough to the peak, the amplitude represents the amplitude height of the pulse waveform, the relative amplitude represents the ratio of the amplitude to the preset reference value, and the rising branch area represents the area between the rising segment of the pulse waveform and the baseline.

[0132] Optionally, the control module inputs periodic data and time-frequency features into the neural network model, and determines the pulse classification result based on the prior knowledge formed by the neural network model during the training phase. This also includes: performing linear interpolation on the periodic data to unify the number of data points for each complete period; wherein linear interpolation is used to increase the number of data points by linear fitting based on adjacent data points, thus unifying the number of data points for different periodic data to a target number; determining the average waveform of the periodic data after linear interpolation; using the average waveform as a feature curve; and inputting the feature curve and time-frequency features into the neural network model, determining the pulse classification result based on the prior knowledge formed by the neural network model during the training phase.

[0133] Optionally, the control module inputs periodic data and time-frequency features into the neural network model, and determines the pulse classification result based on the prior knowledge formed by the neural network model during the training phase. This includes: determining heart rate variability features based on the target pulse signal data, whereby the heart rate variability features include at least one of statistical features used to describe the degree of heart rate fluctuation and frequency domain features used to describe the spectral distribution of heart rate fluctuation; inputting periodic data, time-frequency features, and heart rate variability features into the neural network model, and determining the pulse classification result based on the prior knowledge formed by the neural network model during the training phase.

[0134] Optionally, the control module is also configured to: acquire blood pressure data measured by an oscillometric method using an airbag pressure sensor, wherein the oscillometric method measures blood pressure by detecting airbag pressure oscillations; and correct the blood pressure data based on pulse classification results and / or heart rate variability characteristics. Pulse signal data and cycle data are not normalized during feature extraction to preserve the amplitude information of the pulse signal data.

[0135] Figure 3 This is a schematic diagram of an optional arrangement of pressure sensors inside the airbag according to an embodiment of this application. Figure 3 As shown, four thin-film pressure sensors are sequentially installed along the radial artery on the inner side of the airbag wristband (the side that fits against the skin). Their positions correspond to the four pulse positions of Cun, Guan, Chi, and Chi-posterior, respectively, and are used to synchronously collect pulse wave signals from each position.

[0136] Figure 4 This is a schematic diagram of an optional wrist-type blood pressure pulse diagnosis device according to an embodiment of this application. Figure 4 As shown, a control unit is provided on the main body of the airbag wristband, including a display screen, a power button and a start button, which is used to control the inflation and deflation of the airbag, signal acquisition and pulse analysis, and display the measurement results.

[0137] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0138] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0139] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0140] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0141] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0142] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0143] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A pulse signal data processing method based on multi-stage pressure administration, characterized in that, include: Acquire pulse signal data collected at multiple preset pressure levels; For each pressure level, intrinsic mode functions (EMFs) are extracted from the pulse signal data to characterize different frequency components of the pulse signal. Target intrinsic mode functions that contribute more than a preset threshold to the reconstructed pulse signal are selected from the intrinsic mode functions, and then the selected target intrinsic mode functions are reconstructed into target pulse signal data. Obtain the periodic data corresponding to the target pulse signal data, and extract the time-frequency features corresponding to the periodic data; The periodic data and the time-frequency features are input into a neural network model. The pulse classification result is determined based on the prior knowledge formed by the neural network model during the training phase. The pulse classification result includes the probability distribution information of the pulse signal data when it belongs to various pulse types.

2. The method according to claim 1, characterized in that, For pulse signal data at each pressure level, intrinsic mode functions (EMFs) are extracted to characterize different frequency components of the pulse signal, including: According to preset conditions, the pulse signal data is decomposed into multiple intrinsic mode functions and a residual component, wherein the residual component is the trend term remaining after the pulse signal data is decomposed, which is used to characterize the changing trend of the pulse signal data. The intrinsic mode function is a signal component that satisfies the preset conditions, wherein the preset conditions include: the number of extreme points and the number of zero crossings in the signal sequence corresponding to the pulse signal data are equal or differ by no more than one, and the local mean of the envelope determined by the local maxima and the envelope determined by the local minima is zero.

3. The method according to claim 1, characterized in that, Selecting target intrinsic mode functions (IMFs) from the IMFs whose contribution to the reconstructed pulse signal is greater than a preset threshold includes: Obtain the power spectrum of the intrinsic mode functions; The frequency component with the highest power in the intrinsic mode function is determined based on the power spectrum of the intrinsic mode function; The power of the intrinsic mode function is accumulated sequentially from largest to smallest according to the power corresponding to the frequency component with the largest power to obtain the cumulative power contribution rate of the intrinsic mode function, wherein the cumulative power contribution rate of the intrinsic mode function is equal to the contribution of the intrinsic mode function to the reconstructed pulse signal; The intrinsic mode function whose cumulative power contribution rate exceeds a preset threshold is determined as the target intrinsic mode function.

4. The method according to claim 1, characterized in that, After acquiring pulse signal data collected at multiple preset pressure levels, the method further includes: Obtain the airbag pressure value corresponding to each pressure level, wherein the airbag pressure value is the internal pressure value of the airbag obtained in real time by the airbag pressure sensor; The pulse signal data is corrected based on the airbag pressure value, wherein the correction operation is used to adjust the pulse signal data at each pressure level to the same pressure reference.

5. The method according to claim 1, characterized in that, Acquiring the period data corresponding to the target pulse signal data includes: In the time series corresponding to the target pulse signal data, the amplitude of each data point is compared with the amplitude of its neighboring data points in turn. Data points with amplitudes greater than those of the two adjacent data points are identified as peak points, and data points with amplitudes less than those of the two adjacent data points are identified as trough points. The period data corresponding to the target pulse signal data is determined based on the detected peak points and trough points.

6. The method according to claim 5, characterized in that, Based on the detected peak points and trough points, the period data corresponding to the target pulse signal data is determined, including: False peaks and troughs that do not conform to the morphological characteristics of pulse signals are removed from the peaks and troughs, and the remaining peaks and troughs are taken as target peaks and target troughs. Multiple cycle data corresponding to the target pulse signal data are determined based on the target peak point and the target trough point.

7. The method according to claim 6, characterized in that, Determining the periodic data corresponding to the target pulse signal data based on the target peak point and the target trough point also includes: The target peak points and the target trough points are arranged in chronological order to form a peak sequence and a trough sequence; Each target peak point in the peak sequence is traversed sequentially, and the first valley point that appears after the target peak point in the valley sequence is found as the matching valley of the target peak point. The target peaks and troughs that match each other are combined in chronological order to form the periodic data.

8. The method according to claim 7, characterized in that, The method further includes: If no target valley point is detected after the i-th target peak point, then the i-th target peak point is removed from the peak sequence, where i is an integer greater than or equal to 1; If no target peak is detected before the j-th target valley, the j-th target valley is removed from the valley sequence, where j is an integer greater than or equal to 1.

9. The method according to claim 6, characterized in that, The pseudo-peaks and pseudo-troughs that do not conform to the morphological characteristics of the pulse signal are removed, including: Based on the morphological characteristics of the pulse waveform, for each peak point, the amplitude difference between the peak point and the adjacent trough point is calculated. If the amplitude difference is less than a preset amplitude threshold, the peak point and the corresponding adjacent trough point are discarded as pseudo peak points and pseudo trough points. The time interval between adjacent peaks is detected. If the time interval is less than a preset time threshold, the peak with the largest amplitude is selected from the adjacent peaks and retained, while the other peaks are discarded.

10. The method according to claim 1, characterized in that, Extracting the time-frequency features corresponding to the periodic data includes: Extract at least one of the following features from the pulse waveform corresponding to the periodic data: maximum rising slope, rising time, target amplitude, relative amplitude, and rising support area, as the time-frequency feature; Wherein, the maximum rising slope represents the maximum slope of the rising segment of the pulse waveform, the rising time represents the time required for the pulse waveform to rise from the trough to the peak, the target amplitude represents the amplitude height of the pulse waveform, the relative amplitude represents the ratio of the target amplitude to a preset reference value, and the rising segment area represents the area between the rising segment of the pulse waveform and the baseline.

11. The method according to claim 1, characterized in that, The process further includes inputting the periodic data and the time-frequency features into a neural network model, determining the pulse classification result based on the prior knowledge formed during the training phase of the neural network model, and including: Linear interpolation is performed on the periodic data to unify the number of data points in each complete periodic data. The linear interpolation is used to unify the number of data points in different periodic data to a target number by increasing the number of data points through linear fitting based on adjacent data points. Determine the average waveform of the periodic data after the linear interpolation is completed; The average waveform is used as the characteristic curve; The characteristic curve and the time-frequency feature are input into the neural network model, and the pulse classification result is determined based on the prior knowledge formed by the neural network model during the training phase.

12. The method according to claim 1, characterized in that, The periodic data and the time-frequency features are input into a neural network model, and the pulse classification result is determined based on the prior knowledge formed by the neural network model during the training phase, including: Heart rate variability characteristics are determined based on the target pulse signal data, wherein the heart rate variability characteristics include at least one of statistical characteristics for describing the degree of heart rate fluctuation and frequency domain characteristics for describing the spectral distribution of heart rate fluctuation; The periodic data, the time-frequency features, and the heart rate variability features are input into a neural network model, and the pulse classification result is determined based on the prior knowledge formed by the neural network model during the training phase.

13. The method according to claim 12, characterized in that, After determining the pulse classification result based on the prior knowledge formed during the training phase of the neural network model, the method further includes: Blood pressure data is obtained by using an oscillometric method to measure blood pressure via an airbag pressure sensor, wherein the oscillometric method is a method of measuring blood pressure by detecting airbag pressure oscillation waves; The blood pressure data are corrected based on the pulse classification results and / or the heart rate variability characteristics.

14. The method according to claim 1, characterized in that, The pulse signal data and the period data are not normalized during feature extraction in order to preserve the amplitude information of the pulse signal data.

15. A blood pressure pulse diagnosis device, characterized in that, include: An airbag wristband is designed to be worn on a user's wrist, and the inner side of the airbag wristband is provided with multiple thin-film pressure sensors. An air pump assembly, connected to the airbag wristband, is used to inflate and deflate the airbag wristband in multiple stages. An airbag pressure sensor is used to detect the airbag pressure inside the airbag wristband in real time. The control module, connected to the air pump assembly, the airbag pressure sensor, and the plurality of thin-film pressure sensors respectively, is used to perform the following steps: Acquire pulse signal data collected at multiple preset pressure levels; extract intrinsic mode functions (IMFs) to characterize different frequency components of the pulse signal for each pressure level; select target IMFs from the IMFs whose contribution to the reconstructed pulse signal is greater than a preset threshold, and then reconstruct the selected target IMFs into target pulse signal data. Obtain the periodic data corresponding to the target pulse signal data, and extract the time-frequency features corresponding to the periodic data; input the periodic data and the time-frequency features into the neural network model, and determine the pulse classification result based on the prior knowledge formed by the neural network model during the training phase.

16. The apparatus according to claim 15, characterized in that, An airbag is also provided on the inner side of the airbag wristband, and the thin film pressure sensor is provided on the surface of the airbag near the wrist, which is used to fit against the user's wrist to collect pulse signals when the airbag is inflated.

17. The apparatus according to claim 15, characterized in that, The plurality of thin-film pressure sensors include four thin-film pressure sensors, corresponding to the cun, guan, chi, and chi-posterior positions respectively, for collecting pulse signals at the cun, guan, chi, and chi-posterior positions respectively.

18. The apparatus according to claim 15, characterized in that, The air pump assembly includes a diaphragm pump, and the multi-stage inflation includes multiple preset pressure levels, maintaining a fixed pressure for a preset duration at each pressure level.

19. The apparatus according to claim 15, characterized in that, It also includes a signal amplification module, the input of which is connected to the plurality of thin-film pressure sensors and the output of which is connected to the control module. The signal amplification module is used to amplify the pulse signal collected by the thin-film pressure sensors and transmit the amplified pulse signal to the control module.

20. The apparatus according to claim 15, characterized in that, The control module includes a microcontroller, a power supply, and a communication module. The microcontroller is connected to the air pump assembly, the airbag pressure sensor, the multiple thin-film pressure sensors, and the communication module. The communication module includes a wireless communication module or a wired communication interface for sending the collected data or the determined pulse classification results to the terminal device.

Images