Blood Pressure Measurement Method and Device Based on Multidimensional Transformation Neural Network Model

Abstract

This disclosure relates to a blood pressure measurement method and apparatus based on a multidimensional transform neural network model. The method includes: acquiring the pulse wave signal of a subject; inputting the pulse wave signal into a multidimensional transform neural network model to obtain a blood pressure value. The multidimensional transform neural network model includes an extremum-based dimensional transform block, a fast Fourier transform-based dimensional transform block, and a one-dimensional deep residual contraction network. The extremum-based dimensional transform block converts the pulse wave signal into a first signal, and then the fast Fourier transform-based dimensional transform block converts the first signal into a second signal, so that the blood pressure value is obtained based on the second signal through the one-dimensional deep residual contraction network. This method enables rapid and accurate blood pressure measurement and maintains continuous monitoring capabilities.

Description

Technical Field

[0001] This disclosure relates to the field of medical device technology, specifically to a blood pressure measurement method, device, and electronic device based on a multidimensional transformation neural network model. Background Technology

[0002] Hypertension is one of the most common chronic diseases worldwide, and a long-term state of high blood pressure significantly increases the risk of heart disease, stroke, and other cardiovascular diseases. Therefore, timely monitoring of blood pressure changes is crucial for the prevention and management of these diseases.

[0003] However, the data processing in related technologies can lead to certain errors in the final detection results. Summary of the Invention

[0004] The blood pressure measurement method, apparatus, and electronic device based on a multidimensional transformation neural network model provided in this exemplary embodiment can at least solve the above-mentioned technical problems and other technical problems not mentioned above.

[0005] According to one aspect of this disclosure, a blood pressure measurement method based on a multidimensional transform neural network model is provided. The method includes: acquiring a pulse wave signal of a subject; inputting the pulse wave signal into the multidimensional transform neural network model to obtain a blood pressure value, wherein the multidimensional transform neural network model includes an extremum-based dimensional transform block, a fast Fourier transform-based dimensional transform block, and a one-dimensional deep residual contraction network. The extremum-based dimensional transform block is used to convert the pulse wave signal into a first signal, and then the fast Fourier transform-based dimensional transform block converts the first signal into a second signal, so that the blood pressure value is obtained based on the second signal by the one-dimensional deep residual contraction network.

[0006] Optionally, the dimensional transformation block based on the extrema is used to perform the following operations: converting the pulse wave signal into a two-dimensional pulse wave signal based on the dimensional transformation of the extrema; inputting the two-dimensional pulse wave signal into a residual network with the last layer of connections removed to obtain a first output; performing pooling processing on the first output to convert it into a one-dimensional pulse wave signal; and performing residual linking on the one-dimensional pulse wave signal and the pulse wave signal to obtain the first signal.

[0007] Optionally, the dimensional transformation based on the maximum and minimum values ​​to convert the pulse wave signal into a two-dimensional pulse wave signal includes: identifying the maximum and minimum values ​​in each cycle of the pulse wave signal; taking the maximum or minimum value in each cycle as the starting point, extracting multiple signal segments of a length of a first preset threshold from the pulse wave signal; and stacking each of the signal segments as rows of a two-dimensional matrix to obtain the two-dimensional pulse wave signal.

[0008] Optionally, the dimension transformation block based on Fast Fourier Transform is used to perform the following operations: normalize the first signal output to obtain a second output; perform multiple dimension transformations based on Fast Fourier Transform on the second output to obtain a third output; and residually link the third output with the pulse wave signal to obtain the second signal.

[0009] Optionally, acquiring the pulse wave signal of the subject includes: acquiring the photoplethysmography (PPG) signal of the subject; performing data cleaning on the PPG signal to obtain a first pulse wave signal; and extracting a second pulse wave signal that meets preset conditions by performing a quality check on the first pulse wave signal; wherein, inputting the pulse wave signal into the multidimensional transform neural network model includes: inputting the second pulse wave signal into the multidimensional transform neural network model.

[0010] Optionally, the step of data cleaning the photoplethysmography (PPG) signal to obtain a first pulse signal includes: performing low-pass filtering on the PPG signal to obtain a third pulse signal; performing discrete wavelet transform on the third pulse signal and eliminating the low-frequency component coefficients of the first layer obtained during the discrete wavelet transform process to obtain the first pulse signal.

[0011] Optionally, the step of extracting a second pulse wave signal that meets preset conditions by performing a quality check on the first pulse wave signal includes: an outlier detection method based on kernel density estimation to detect outliers in the pulse wave signals within each window of the first pulse wave signal, wherein each window is obtained by pre-segmenting the first pulse wave signal; and obtaining the second pulse wave signal that meets the preset conditions by extracting the pulse wave signals from the windows of the first pulse wave signal that do not contain outliers.

[0012] Optionally, the outlier detection method based on kernel density estimation detects outliers in the pulse wave signal within each window of the first pulse wave signal, including: calculating the quality score of each data point within each window of the first pulse wave signal based on a kernel density estimation formula using a smooth and symmetric Gaussian kernel, wherein the quality score is the reciprocal of the kernel density estimate and is used to determine whether the data point is an outlier, and the kernel density estimation formula is expressed as:

[0013]

[0014] Where n represents all data point samples in the first pulse wave signal, ω is the bandwidth parameter, and x represents at least one of the following characteristics of the data points in each window of the first pulse wave signal: mean, variance, skewness, kurtosis, rising amplitude, diphtheria amplitude, falling mid-wave amplitude, and area under the curve.

[0015] According to another aspect of this disclosure, a blood pressure measurement device based on a multidimensional transform neural network model is also provided. The device includes: a signal acquisition unit configured to acquire the pulse wave signal of a subject; and a blood pressure calculation unit configured to input the pulse wave signal into the multidimensional transform neural network model to obtain a blood pressure value. The multidimensional transform neural network model includes an extremum-based dimensional transform block, a fast Fourier transform-based dimensional transform block, and a one-dimensional deep residual contraction network. The extremum-based dimensional transform block converts the pulse wave signal into a first signal, and then converts the first signal into a second signal through the fast Fourier transform-based dimensional transform block, so that the blood pressure value is obtained based on the second signal through the one-dimensional deep residual contraction network.

[0016] Optionally, the blood pressure calculation unit is configured to: convert the pulse wave signal into a two-dimensional pulse wave signal based on the dimensional transformation of the extreme value; input the two-dimensional pulse wave signal into a residual network with the last layer of connections removed to obtain a first output; perform pooling processing on the first output to convert it into a one-dimensional pulse wave signal; and perform residual linking on the one-dimensional pulse wave signal and the pulse wave signal to obtain the first signal.

[0017] Optionally, the blood pressure calculation unit is configured to: identify the maximum and minimum values ​​in each cycle of the pulse wave signal; take the maximum or minimum value in each cycle as the starting point, extract multiple signal segments of length denoted by a first preset threshold from the pulse wave signal; and stack each signal segment as a row of a two-dimensional matrix to obtain the two-dimensional pulse wave signal.

[0018] Optionally, the blood pressure calculation unit is configured to: normalize the first signal output to obtain a second output; perform multiple dimensional transformations based on fast Fourier transform on the second output to obtain a third output; and perform residual linking of the third output with the pulse wave signal to obtain the second signal.

[0019] Optionally, the signal acquisition unit is further configured to: acquire the photoplethysmography (PPG) signal of the subject; perform data cleaning on the PPG signal to obtain a first pulse wave signal; and extract a second pulse wave signal that meets preset conditions by performing a quality check on the first pulse wave signal; wherein, the blood pressure calculation unit is further configured to: input the second pulse wave signal into the multidimensional transform neural network model.

[0020] Optionally, the signal preprocessing unit is configured to: perform low-pass filtering on the photoplethysmography pulse wave signal to obtain a third pulse wave signal; perform discrete wavelet transform on the third pulse wave signal and eliminate the low-frequency component coefficients of the first layer obtained during the discrete wavelet transform process to obtain the first pulse wave signal.

[0021] Optionally, the signal preprocessing unit is configured to: perform outlier detection on the pulse wave signals within each window of the first pulse wave signal based on kernel density estimation, wherein each window is obtained by pre-segmenting the first pulse wave signal; and obtain the second pulse wave signal that satisfies the preset conditions by extracting the pulse wave signals of the windows in the first pulse wave signal that do not contain outliers.

[0022] Optionally, the signal preprocessing unit is configured to: calculate the quality score of each data point within each window of the first pulse wave signal based on a kernel density estimation formula using a smooth and symmetric Gaussian kernel, wherein the quality score is the reciprocal of the kernel density estimate and is used to determine whether the data point is an outlier, and the kernel density estimation formula is expressed as:

[0023]

[0024] Where n represents all data point samples in the first pulse wave signal, ω is the bandwidth parameter, and x represents at least one of the following characteristics of the data points in each window of the first pulse wave signal: mean, variance, skewness, kurtosis, rising amplitude, diphtheria amplitude, falling mid-wave amplitude, and area under the curve.

[0025] According to another aspect of the present disclosure, an electronic device is also provided, comprising: at least one processor; at least one memory storing computer-executable instructions, wherein the computer-executable instructions, when executed by the at least one processor, cause the at least one processor to perform the blood pressure measurement method based on a multidimensional transformation neural network model as described above.

[0026] According to another aspect of the embodiments of this disclosure, a computer-readable storage medium storing instructions that, when executed by at least one processor, cause the at least one processor to perform the blood pressure measurement method based on a multidimensional transformation neural network model as described above.

[0027] According to another aspect of the embodiments of this disclosure, a system is also provided that includes at least one computing device and at least one storage device for storing instructions, wherein the instructions, when executed by the at least one computing device, cause the at least one computing device to perform the blood pressure measurement method based on the multidimensional transformation neural network model as described above.

[0028] According to another aspect of the embodiments of this disclosure, a computer program product is also provided, including a computer program / instructions that, when executed by a processor, implement the blood pressure measurement method based on a multi-dimensional transformation neural network model as described above.

[0029] The technical solutions provided in this disclosure offer at least the following beneficial effects:

[0030] According to the blood pressure measurement method, device, and electronic device based on the multidimensional transformation neural network model disclosed herein, by capturing the pulse wave signal of the subject and using a series of dimension transformation techniques and feature extraction techniques to process the pulse wave signal, it is possible not only to capture the instantaneous fluctuations in the pulse wave signal, but also to grasp the overall trend of the pulse wave signal, thereby achieving rapid and accurate blood pressure measurement.

[0031] Furthermore, based on the acquisition of photoplethysmography (PPG) pulse wave signals, high-quality pulse wave signals are obtained through data processing methods such as cleaning and optimization. After signal quality screening, effective pulse wave signals can be obtained. These effective pulse wave signals undergo a series of dimensional transformations, converting one-dimensional signals into multiple two-dimensional signals through various dimensional transformations. This allows for the extraction of richer information. Combined with feature extraction processing, the instantaneous fluctuations of the pulse wave signal can be captured more accurately, and the overall trend can be grasped. This enables faster and more accurate blood pressure measurement. In addition, it can maintain continuous monitoring capabilities and can be applied to miniaturized, wearable blood pressure monitoring devices for frequent use, improving the portability and comfort of the blood pressure measurement process. Attached Figure Description

[0032] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0033] Figure 1 A flowchart illustrating a blood pressure measurement method based on a multidimensional transformation neural network model in an exemplary embodiment of this disclosure is shown.

[0034] Figure 2 A schematic diagram of the structure of a multi-dimensional transformation neural network model in an exemplary embodiment of this disclosure is shown;

[0035] Figure 3This diagram illustrates the data processing flow of a dimension transformation block based on extrema in an exemplary embodiment of this disclosure.

[0036] Figure 4 This diagram illustrates the data processing flow of a dimension transformation block based on the Fast Fourier Transform in an exemplary embodiment of this disclosure.

[0037] Figure 5 A schematic diagram illustrating the data processing flow of a one-dimensional deep residual shrinkage network in an exemplary embodiment of this disclosure is shown.

[0038] Figure 6 This illustration shows one structural diagram of a multidimensional transformation neural network model in an exemplary embodiment of the present disclosure;

[0039] Figure 7 A block diagram of a blood pressure measurement device based on a multidimensional transformation neural network model is shown in an exemplary embodiment of this disclosure;

[0040] Figure 8 This illustration shows a schematic diagram of one structure of a blood pressure measuring device based on a multidimensional transformation neural network model in an exemplary embodiment of the present disclosure;

[0041] Figure 9 A block diagram illustrating an electronic device according to an exemplary embodiment of the present disclosure. Detailed Implementation

[0042] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0043] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure 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 disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following examples do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0044] It should be noted that the phrase "at least one of several items" in this disclosure refers to three parallel cases: "any one of the several items", "a combination of any number of the several items", and "all of the several items". For example, "including at least one of A and B" includes the following three parallel cases: (1) including A; (2) including B; (3) including A and B. As another example, "performing at least one of step one and step two" indicates the following three parallel cases: (1) performing step one; (2) performing step two; (3) performing both step one and step two.

[0045] Currently, there are two main methods in blood pressure monitoring technology: pressure sensing-based methods and photoelectric sensing-based methods.

[0046] Pressure-sensing methods provide relatively accurate blood pressure readings by directly measuring changes in arterial wall pressure. However, this method has several significant drawbacks, including discomfort when worn, large device size making it difficult to carry, and limited frequency of use. These issues restrict its widespread application in daily life.

[0047] Photoelectric sensing-based methods detect changes in blood flow using optical means, thereby indirectly calculating blood pressure values. This method improves user comfort, portability, and the feasibility of continuous monitoring. However, this method also has some drawbacks, such as the data being easily affected by external interference (e.g., movement, ambient light), and the relatively complex data processing, which may lead to some errors in the final detection results.

[0048] To address one or more problems in related technologies, such as large errors in blood pressure measurement, difficulty in carrying, inconvenience, and infrequent use, this disclosure provides a blood pressure measurement method, device, and electronic device based on a multi-dimensional transformation neural network model. By capturing the pulse wave signal of the subject and using a series of dimensional transformation and feature extraction techniques to process the pulse wave signal, it can not only capture the instantaneous fluctuations in the pulse wave signal but also grasp the overall trend of the pulse wave signal, thereby achieving rapid and accurate blood pressure measurement.

[0049] Below, we will refer to Figures 1 to 9 This disclosure provides a detailed description of the blood pressure measurement method, apparatus, and electronic device based on a multidimensional transformation neural network model.

[0050] Figure 1 A flowchart illustrating a blood pressure measurement method based on a multidimensional transformation neural network model in an exemplary embodiment of this disclosure is shown.

[0051] Reference Figure 1 In step 101, the pulse wave signal of the subject is acquired.

[0052] Specifically, pulse wave signals of the subject can be obtained through methods such as ultrasonic Doppler ultrasound and impedance plethysmography.

[0053] According to an exemplary embodiment of this disclosure, the photoplethysmography (PPG) signal of the subject can be acquired.

[0054] Specifically, photoplethysmography (PPG) signals can be acquired using a photoelectric module. Exemplary embodiments of this disclosure are illustrated using PPG signals as examples.

[0055] According to an exemplary embodiment of this disclosure, the photoplethysmography (PPG) signal can be cleaned to obtain a first PPG signal (i.e., a high-quality PPG signal); and a second PPG signal (i.e., an effective PPG signal) that meets preset conditions can be extracted by performing a quality check on the first PPG signal.

[0056] Specifically, methods based on filtering, thresholding, statistics, and machine learning can be used to clean photoplethysmography (PPG) signals. The quality of the first pulse wave signal can be checked using time-domain analysis, frequency-domain analysis, statistical features, and machine learning.

[0057] According to an exemplary embodiment of this disclosure, the process of data cleaning of the photoplethysmography (PPG) signal may include, but is not limited to, the following operations: the PPG signal may be low-pass filtered to obtain a third pulse signal; the third pulse signal may be discretely wavelet transformed, and the low-frequency component coefficients of the first layer obtained during the discrete wavelet transform may be eliminated to obtain a first pulse signal.

[0058] Specifically, a low-pass filter based on a Butterworth template from 0 to 8 Hz can be used to low-pass filter the photoplethysmography (PPG) signal. This low-pass filter strictly adheres to the design principles of an Infinite Impulse Response (IIR) filter. The frequency domain formula for the Butterworth analog low-pass filter is as follows:

[0059]

[0060] Where N is the order, ω cThe cutoff frequency is 3dB. After obtaining the S-domain expression H(s) of the analog low-pass filter, the impulse response invariance method can be used to convert the S-domain to the Z-domain. The impulse response invariance method utilizes the principle of time-domain approximation, taking the inverse Laplace transform of the S-domain expression H(s) to obtain the continuous time-domain expression h[t], sampling the time domain to obtain the discrete time-domain expression h[k], and finally using the Z-transform to obtain the Z-domain expression H(z) of the Butterworth digital low-pass filter. The Butterworth filter has advantages such as a flat passband, which can filter out high-frequency interference to a large extent while maintaining the original waveform blood pressure information, thus obtaining the third pulse wave signal.

[0061] Subsequently, the db8 wavelet function can be used to perform a discrete wavelet transform on the third pulse wave signal. Furthermore, the low-frequency component coefficients in the first layer obtained during the discrete wavelet transform can be eliminated to obtain the first pulse wave signal (i.e., the high-quality pulse wave signal). The discrete wavelet transform can achieve high-frequency and low-frequency subdivision of the time-series waveform by stretching and shifting the wavelet function. Moreover, the discrete wavelet transform has the characteristic of energy concentration, which can effectively derive detail functions and approximate functions at different scales of the signal. Finally, baseline drift can be effectively removed to obtain the high-quality pulse wave signal (i.e., the first pulse wave signal).

[0062] According to an exemplary embodiment of this disclosure, the process of quality checking the first pulse wave signal may include, but is not limited to, the following operations: outlier detection may be performed on the pulse wave signal within each window of the first pulse wave signal based on the outlier detection method of kernel density estimation, wherein each window is obtained by pre-segmenting the first pulse wave signal; and the second pulse wave signal that satisfies the preset conditions is obtained by extracting the pulse wave signal from the window in the first pulse wave signal where there are no outliers.

[0063] Specifically, the window size of the first pulse wave signal can be set to 3-5 seconds for segmentation to perform signal quality checks. Furthermore, a quality score can be obtained using an outlier detection method based on kernel density estimation, thereby identifying outliers. The preset condition is that no outliers exist in any window of the pulse wave signal; therefore, the pulse wave signal from each window without outliers can be extracted to obtain the valid pulse wave signal (i.e., the second pulse wave signal), thus achieving quality checks on the first pulse wave signal.

[0064] In addition, the pulse wave signal of each window containing outliers can be extracted to obtain invalid pulse wave signals.

[0065] According to an exemplary embodiment of this disclosure, the outlier detection method based on kernel density estimation, which detects outliers in each window of the first pulse wave signal, may include, but is not limited to, the following operations: The quality score of each data point in each window of the first pulse wave signal can be calculated based on a kernel density estimation formula using a smooth and symmetric Gaussian kernel, wherein the quality score is the reciprocal of the kernel density estimate and is used to determine whether the data point is an outlier. The kernel density estimation formula is expressed as:

[0066]

[0067] Where n represents all data point samples in the first pulse wave signal, ω is the bandwidth parameter, and x represents at least one of the following characteristics of the data points in each window of the first pulse wave signal: mean, variance, skewness, kurtosis, rising amplitude, diphthous wave amplitude, falling mid-wave amplitude, and area under the curve.

[0068] Specifically, in the signal quality inspection process, an outlier detection method based on kernel density estimation can be used. This method employs eight data dimensions for outlier detection: PPG signal mean, variance, skewness, kurtosis, rise amplitude, diabetic wave amplitude, drop-off amplitude, and area under the curve. This allows for determination of whether each cycle within the signal window is normalized. The kernel density estimation formula is as follows:

[0069]

[0070] Where n represents all data samples, and ω is the bandwidth parameter, the formula is essentially a calculation of the relative distance between a certain signal window and all other signal windows. The key to kernel density estimation lies in the selection of the kernel function. In this disclosure, a smooth and symmetric Gaussian kernel can be used. The Gaussian kernel formula is as follows:

[0071]

[0072] Substituting the Gaussian kernel into the kernel density estimation formula yields:

[0073]

[0074] The quality score can be set as the reciprocal of the kernel density estimate to determine whether data points are in low-density regions, thereby identifying outliers.

[0075] In step 102, the pulse wave signal is input into a multidimensional transform neural network model to obtain the blood pressure value. The multidimensional transform neural network model includes a dimension transform block based on extrema, a dimension transform block based on fast Fourier transform, and a one-dimensional deep residual contraction network. The dimension transform block based on extrema is used to convert the pulse wave signal into a first signal, and then the dimension transform block based on fast Fourier transform converts the first signal into a second signal, so that the blood pressure value is obtained based on the second signal through the one-dimensional deep residual contraction network.

[0076] Specifically, the blood pressure estimation process can be achieved through a blood pressure estimation model (i.e., a multidimensional transformation neural network model).

[0077] Figure 2 A schematic diagram of the structure of a multidimensional transformation neural network model in an exemplary embodiment of this disclosure is shown.

[0078] Reference Figure 2 The multidimensional transformation neural network model mainly consists of three parts: the maximum-minimum based dimension transformation block 201 (MMBlock), the fast Fourier transform based dimension transformation 202 (FFTBlock), and the one-dimensional deep residual shrinkage network 203 (DRSNs1D).

[0079] Here, the photoplethysmography (PPG) pulse wave signal, or a first pulse wave signal (high-quality pulse wave signal) after data cleaning, or a second pulse wave signal (effective pulse wave signal) after data cleaning and quality checking, can be input into the multidimensional transformation neural network model, and finally output the corresponding blood pressure value. According to an exemplary embodiment of this disclosure, the extremum-based dimensional transformation block can be used to perform operations including but not limited to: converting the pulse wave signal into a two-dimensional pulse wave signal based on the extremum-based dimensional transformation; inputting the two-dimensional pulse wave signal into a residual network with the last layer of connections removed to obtain a first output; performing pooling processing on the first output to convert it into a one-dimensional pulse wave signal; and performing residual linking on the one-dimensional pulse wave signal and the pulse wave signal to obtain a first signal.

[0080] Specifically, in the first stage, the input photoplethysmography (PPG) signal (or high-quality pulse wave signal, or effective pulse wave signal) is first converted into two-dimensional data (i.e., two-dimensional pulse wave signal) through a dimension transformation based on extrema. Subsequently, this data can be fed into a modified ResNet model (residual network), which removes the connections of the last layer to obtain the first output while maintaining the shape of the two-dimensional data (two-dimensional pulse wave signal). After the first output is processed by a pooling layer, the two-dimensional data (two-dimensional pulse wave signal) can be flattened into one dimension (one-dimensional pulse wave signal) and formed with the original input (PPG signal, or high-quality pulse wave signal, or effective pulse wave signal) to form a residual structure, thereby obtaining the first signal to enhance feature representation.

[0081] It is understood that the original input referred to in this disclosure refers to the input of the multidimensional transformation neural network model.

[0082] According to an exemplary embodiment of the present disclosure, converting a pulse wave signal into a two-dimensional pulse wave signal based on the dimensional transformation of the maximum and minimum values ​​may include, but is not limited to, the following operations: identifying the maximum and minimum values ​​in each cycle of the pulse wave signal; taking the maximum or minimum value in each cycle as the starting point, extracting multiple signal segments of length denoted by a first preset threshold from the pulse wave signal; and stacking each signal segment as a row of a two-dimensional matrix to obtain a two-dimensional pulse wave signal.

[0083] Specifically, Figure 3 This diagram illustrates the data processing flow of a dimension transformation block based on extrema in an exemplary embodiment of this disclosure.

[0084] Reference Figure 3 Based on the extrema-based dimension transformation block (MMBlock), the extrema points of high-quality pulse wave signals (or photoplethysmography pulse wave signals, or effective pulse wave signals; the following explanation uses high-quality pulse wave signals as an example) can be identified and arranged alternately, i.e., X. mm ={X max1 ,X min1 ,X max2 ,X min2 ,X max3 ,X min3 Before dimensional transformation, a high-quality pulse wave signal X can be obtained. 1D The expansion is performed, padding with zeros L times. Next, a sliding window length of L (a first preset threshold) can be set to extract X from the high-quality pulse wave signal. mm Begin by extracting a high-quality pulse wave signal X of length L from the beginning. 1dEach segment is taken as a row of two-dimensional data (two-dimensional pulse wave signal). Then, each row can be stacked to form a two-dimensional matrix, thus obtaining the two-dimensional pulse wave signal. The dimensionality transformation process can be explained by the following formula:

[0085] X 2D = append{slide(X 1D ), size = L, head = X mm}

[0086] Since the number of extreme points in the data varies, the matrix can be expanded by padding with zeros to include n rows. This two-dimensional matrix can then be fed into a two-dimensional deep neural network for training; for example, this deep neural network could be a ResNet model. Finally, the two-dimensional model (two-dimensional pulse wave signal) can be converted back to one-dimensional (one-dimensional pulse wave signal), and a residual structure can be established between the one-dimensional pulse wave signal and the original input (photoplethysmography pulse wave signal, or high-quality pulse wave signal, or effective pulse wave signal) to form the first signal, which is also the first output.

[0087] According to an exemplary embodiment of this disclosure, a dimension transformation block based on Fast Fourier Transform can be used to perform operations including but not limited to: normalizing a first signal output to obtain a second output; performing multiple dimension transformations based on Fast Fourier Transform on the second output to obtain a third output; and residually linking the third output with a pulse wave signal to obtain a second signal.

[0088] Specifically, after obtaining the second output through the first-stage one-dimensional data normalization operation, multiple dimensional transformations based on Fast Fourier Transform (FFTBlock) can be performed on the second output to obtain the third output.

[0089] Unlike MMBlock, FFTBlock does not rely on the explicit periodicity of the signal, and therefore can be reused to deepen the network structure and learn richer features. Thus, the signal is processed through an FFT block and combined with the original input using residual connections. To fully extract features, the multidimensional transform neural network model can reuse FFTBlock multiple times, for example, up to three times.

[0090] Specifically, Figure 4 This diagram illustrates the data processing flow of a dimension transformation block based on the Fast Fourier Transform in an exemplary embodiment of this disclosure.

[0091] Reference Figure 4The Fast Fourier Transform (FFT) can be introduced as the basis for dividing the data. A dimensionality transform block (FFTBlock) based on the FFT can be used to handle the input of the first signal. In this case, the step of normalizing the output of the first signal can be integrated into the FFTBlock. By performing FFT analysis on the input signal, the amplitude A = {A...} can be extracted during the FFT transform stage. f1,..., A fk The first k prominent frequencies f = {f1,...,f} k Based on these frequencies, the one-dimensional data can be decomposed k times, with each decomposition having a length of f, padded with zeros if necessary. These fragments can then be recombine to form k sets of two-dimensional data. The process can be illustrated by the following formula:

[0092] A = Amp(FFT(X) 1D ),{f1,f2,...,f k} = argTopk(A)

[0093]

[0094] Next, two-dimensional convolutional kernels can be used to extract signal features. This method can simultaneously capture the dependencies within a time period and the dependencies between adjacent time periods. In this process, the model can use four convolutional kernels of different sizes (1x1, 3x3, 5x5, and 7x7, respectively) (Inception Block) for convolution operations. Subsequently, a Flatten layer can be used to convert the two-dimensional data back into a one-dimensional form. To represent the differences between different time periods, a set of programmable coefficients can be introduced into the Adaptive Aggregation layer, and combined with a softmax function to determine the weighted sum of k one-dimensional data segments. The length of this weighted sum can be equal to k.

[0095] Finally, these k weighted one-dimensional data can be combined with the original input to form a residual structure, which in turn forms the second signal. Here, k can be 3 to 5.

[0096] According to an exemplary embodiment of this disclosure, in the third stage, a one-dimensional deep residual contraction network (DRSNs1D) can be used to convert the one-dimensional data second signal into a blood pressure value. Specifically, a modified DRSNs (Deep Residual Contraction Networks) adapted to one-dimensional data, namely a one-dimensional deep residual contraction network (DRSNs1D), can outperform the two-dimensional version. After operations such as convolution, soft thresholding, and pooling, the model can finally compress the extracted information to obtain the blood pressure label value. For systolic and diastolic blood pressure, the same model structure can be used to build personalized models respectively.

[0097] Specifically, Figure 5 This diagram illustrates the data processing flow of a one-dimensional deep residual shrinkage network in an exemplary embodiment of the present disclosure.

[0098] Reference Figure 5 The one-dimensional deep residual shrinking network includes key components such as batch normalization (BN), rectified linear units (ReLU), convolution operations (Conv), global average pooling (GAP), and fully connected layers (FC), as well as components such as absolute value operations, average operations, activation functions (ReLU, Sigmoid), soft thresholding, and backpropagation (BP). The one-dimensional deep residual shrinking network can be responsible for receiving the second signal and outputting the final blood pressure value. Furthermore, DRSNs1D can introduce soft thresholding as a nonlinear transformation layer. The threshold used can be automatically determined in the deep architecture, and the function of soft thresholding can be expressed as:

[0099]

[0100] Here, x represents the input feature, y represents the output feature, and τ is the threshold, i.e., a positive parameter. The soft thresholding method does not set the negative features in the ReLU activation function to zero; instead, it sets features close to zero to zero, thus preserving useful negative features. This effectively prevents the vanishing and exploding gradient problems.

[0101] Figure 6 This diagram illustrates one structural schematic of a multidimensional transformation neural network model in an exemplary embodiment of the present disclosure.

[0102] Reference Figure 6 The multidimensional transform neural network model can include an extremum-based dimensional transform block (MMblock), three fast Fourier transform blocks (FFTBlock), and a one-dimensional deep residual shrinking network (DRSNs). The input can be a valid pulse wave signal that has been cleaned and quality checked. After the input is processed by the extremum-based dimensional transform block, the three fast Fourier transform blocks, and the one-dimensional deep residual shrinking network, the output, i.e., the blood pressure value, can be obtained.

[0103] According to exemplary embodiments of this disclosure, a blood pressure monitoring method integrating multiple dimensional transformation models is proposed. By capturing the optical volumetric pulse signal of the subject, optimizing it into a high-quality pulse wave signal, and after signal quality screening, a series of dimensional transformation techniques are used to process the pulse wave signal. This not only captures instantaneous fluctuations but also grasps the overall trend, thereby achieving rapid and accurate blood pressure measurement and maintaining continuous monitoring capabilities. Furthermore, this blood pressure monitoring method can be applied to portable devices.

[0104] Figure 7 A block diagram of a blood pressure measurement device based on a multidimensional transformation neural network model is shown in an exemplary embodiment of this disclosure.

[0105] Reference Figure 7 The exemplary embodiments of this disclosure also provide a blood pressure measuring device 700 based on a multidimensional transformation neural network model, which may include, but is not limited to, a signal acquisition unit 701 and a blood pressure calculation unit 702.

[0106] The signal acquisition unit 701 can acquire the pulse wave signal of the subject.

[0107] The blood pressure calculation unit 702 can input the pulse wave signal into a multidimensional transform neural network model to obtain the blood pressure value. The multidimensional transform neural network model includes a dimension transform block based on extrema, a dimension transform block based on fast Fourier transform, and a one-dimensional deep residual contraction network. The dimension transform block based on extrema is used to convert the pulse wave signal into a first signal, and then the dimension transform block based on fast Fourier transform converts the first signal into a second signal, so that the blood pressure value can be obtained based on the second signal through the one-dimensional deep residual contraction network.

[0108] According to an exemplary embodiment of the present disclosure, the blood pressure calculation unit 702 can convert the pulse wave signal into a two-dimensional pulse wave signal based on the dimensional transformation of the extreme value; input the two-dimensional pulse wave signal into a residual network with the last layer of connection removed to obtain a first output; perform pooling processing on the first output to convert it into a one-dimensional pulse wave signal; and perform residual linking on the one-dimensional pulse wave signal and the pulse wave signal to obtain a first signal.

[0109] According to an exemplary embodiment of the present disclosure, the blood pressure calculation unit 702 can identify the maximum and minimum values ​​in each cycle of the pulse wave signal; take the maximum or minimum value in each cycle as the starting point, extract multiple signal segments with a length of a first preset threshold from the pulse wave signal; and stack each signal segment as a row of a two-dimensional matrix to obtain a two-dimensional pulse wave signal.

[0110] According to an exemplary embodiment of this disclosure, the blood pressure calculation unit 702 can normalize the first signal output to obtain a second output; perform multiple dimensional transformations based on fast Fourier transform on the second output to obtain a third output; and perform residual linking of the third output with the pulse wave signal to obtain a second signal.

[0111] According to an exemplary embodiment of the present disclosure, the signal acquisition unit 701 may also acquire the photoplethysmography (PPG) signal of the subject; perform data cleaning on the PPG signal to obtain a first pulse wave signal; and extract a second pulse wave signal that meets preset conditions by performing a quality check on the first pulse wave signal; wherein, the blood pressure calculation unit 702 may also input the second pulse wave signal into a multidimensional transformation neural network model.

[0112] According to an exemplary embodiment of the present disclosure, the signal acquisition unit 701 may further perform low-pass filtering on the photoplethysmography pulse wave signal to obtain a third pulse wave signal; perform discrete wavelet transform on the third pulse wave signal, and eliminate the low-frequency component coefficients of the first layer obtained by decomposition during the discrete wavelet transform process to obtain a first pulse wave signal.

[0113] According to an exemplary embodiment of the present disclosure, the signal acquisition unit 701 can also perform outlier detection on the pulse wave signal in each window of the first pulse wave signal based on the outlier detection method of kernel density estimation, wherein each window is obtained by pre-segmenting the first pulse wave signal; by extracting the pulse wave signal of the window in the first pulse wave signal that does not contain outliers, a second pulse wave signal that meets the preset conditions is obtained.

[0114] According to an exemplary embodiment of this disclosure, the signal preprocessing unit can calculate the quality score of each data point within each window of the first pulse wave signal based on a kernel density estimation formula employing a smooth and symmetric Gaussian kernel. The quality score is the reciprocal of the kernel density estimate and is used to determine whether the data point is an outlier. The kernel density estimation formula is expressed as:

[0115]

[0116] Where n represents all data point samples in the first pulse wave signal, ω is the bandwidth parameter, and x represents at least one of the following characteristics of the data points in each window of the first pulse wave signal: mean, variance, skewness, kurtosis, rising amplitude, diphtheria amplitude, falling mid-wave amplitude, and area under the curve.

[0117] It is understood that the specific implementation process of the blood pressure measuring device 700 based on the multi-dimensional transformation neural network model described above is largely the same as that of the blood pressure measuring method based on the multi-dimensional transformation neural network model described above, and will not be repeated here. The blood pressure measuring device 700 based on the multi-dimensional transformation neural network model can be configured as software, hardware, firmware, or any combination thereof to perform specific functions. For example, these devices may correspond to dedicated integrated circuits, pure software code, or modules combining software and hardware. Furthermore, one or more functions implemented by these devices may also be uniformly executed by components in a physical entity device (e.g., a processor, client, or server).

[0118] According to exemplary embodiments of this disclosure, Figure 8 This diagram illustrates one structural schematic of a blood pressure measuring device based on a multidimensional transformation neural network model, as shown in an exemplary embodiment of this disclosure. (Refer to...) Figure 8 The device may include a data collection unit 801, a signal preprocessing unit 802, a signal quality inspection unit 803, a blood pressure calculation unit 804, and an information storage unit 805 connected in sequence.

[0119] The data collection unit 801 can be used to acquire the photoplethysmography (PPG) signal of the person being tested and transmit the PPG signal to the signal preprocessing unit 802.

[0120] The signal preprocessing unit 802 can be used to clean the photoplethysmography (PPG) signal to obtain a high-quality PPG signal, and then transmit the high-quality PPG signal to the signal quality inspection unit 803.

[0121] The signal quality inspection unit 803 can be used to inspect the signal quality of high-quality pulse wave signals and transmit high-quality pulse wave signals that meet the inspection conditions to the blood pressure calculation unit 804.

[0122] The blood pressure calculation unit 804 can be used to analyze the features and data correlation in the high-quality pulse wave signal, perform blood pressure detection on the subject, and transmit the high-quality pulse wave signal and blood pressure calculation results to the information storage unit 805.

[0123] The information storage unit 805 can be used to save high-quality pulse wave signals and blood pressure calculation results to a storage device.

[0124] The data collection unit 801 can be built based on a photoelectric module for acquiring photoplethysmography (PPG) signals. This photoelectric module can include a photoelectric sensing device and a light source device; the former can be a photoelectric sensor, and the latter can be a device capable of emitting green light.

[0125] After capturing these PPG signals, the data collection unit 801 can pass them to the connected signal preprocessing unit 802. The signal preprocessing unit 802 can receive the PPG signals from the data collection unit 801 and preprocess them to ensure that the signals meet the input data requirements of the subsequent blood pressure estimation algorithm (i.e., the multidimensional transformation neural network model), thereby generating high-quality pulse wave signals.

[0126] Figure 9 A block diagram illustrating an electronic device according to an exemplary embodiment of the present disclosure.

[0127] Reference Figure 9 The electronic device 900 includes at least one memory 901 and at least one processor 902. The at least one memory 901 stores a set of computer-executable instructions. When the set of computer-executable instructions is executed by the at least one processor 902, a blood pressure measurement method based on a multidimensional transform neural network model according to an exemplary embodiment of the present disclosure is executed.

[0128] As an example, electronic device 900 may be a PC, tablet, personal digital assistant, smartphone, or other device capable of executing the aforementioned set of instructions. Here, electronic device 900 is not necessarily a single electronic device, but may be any collection of devices or circuits capable of executing the aforementioned instructions (or instruction sets) individually or in combination. Electronic device 900 may also be part of an integrated control system or system manager, or may be configured to interconnect with a portable electronic device locally or remotely (e.g., via wireless transmission) through an interface.

[0129] In electronic device 900, processor 902 may include a central processing unit (CPU), a graphics processing unit (GPU), a programmable logic device, a dedicated processor system, a microcontroller, or a microprocessor. By way of example and not limitation, processor may also include analog processors, digital processors, microprocessors, multi-core processors, processor arrays, network processors, etc.

[0130] The processor 902 can execute instructions or code stored in the memory 901, which can also store data. Instructions and data can also be sent and received via a network through a network interface device, which can employ any known transmission protocol.

[0131] The memory 901 may be integrated with the processor 902, for example, by placing RAM or flash memory within an integrated circuit microprocessor. Alternatively, the memory 901 may include a separate device, such as an external disk drive, a storage array, or other storage device usable by any database system. The memory 901 and the processor 902 may be operatively coupled, or may communicate with each other, for example, via I / O ports, network connections, etc., enabling the processor 902 to read files stored in the memory.

[0132] In addition, the electronic device 900 may also include a video display (such as a liquid crystal display) and a user interaction interface (such as a keyboard, mouse, touch input device, etc.). All components of the electronic device 900 can be interconnected via a bus and / or network.

[0133] According to exemplary embodiments of the present disclosure, a computer-readable storage medium storing instructions may also be provided, wherein when the instructions are executed by at least one computing device, they cause at least one computing device to perform the blood pressure measurement method based on the multidimensional transformation neural network model described above.

[0134] Examples of computer-readable storage media herein include: read-only memory (ROM), random access programmable read-only memory (PROM), electrically erasable programmable read-only memory (EEPROM), random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), flash memory, non-volatile memory, CD-ROM, CD-R, CD+R, CD-RW, CD+RW, DVD-ROM, DVD-R, DVD+R, DVD-RW, DVD+RW, DVD-RAM, BD-ROM, BD-R, BD-R LTH, BD-RE, Blu-ray or optical disc storage, hard disk drive (HDD), solid-state drive (SSD), card storage (such as multimedia cards, secure digital (SD) cards, or ultra-fast digital (XD) cards), magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid-state drive, and any other device configured to store a computer program and any associated data, data files, and data structures in a non-transitory manner and to provide the computer program and any associated data, data files, and data structures to a processor or computer so that the processor or computer can execute the computer program. The computer program in the aforementioned computer-readable storage medium can run in an environment deployed in computer devices such as clients, hosts, agent devices, servers, etc. Furthermore, in one example, the computer program and any associated data, data files, and data structures are distributed across a networked computer system, such that the computer program and any associated data, data files, and data structures are stored, accessed, and executed in a distributed manner through one or more processors or computers. It should be noted that the instructions can also be used to perform additional steps in addition to those described above, or to perform more specific processing while performing the above steps. The details of these additional steps and further processing have already been mentioned in the description of the relevant methods, so they will not be repeated here to avoid repetition.

[0135] Another embodiment of this disclosure relates to a system including at least one computing device and at least one storage device for storing instructions, wherein the instructions, when executed by at least one computing device, cause at least one computing device to perform the blood pressure measurement method based on the multidimensional transformation neural network model described above.

[0136] It should be noted that the system according to the exemplary embodiments of this disclosure may rely entirely on the operation of computer programs or instructions to achieve the corresponding functions. That is, each unit corresponds to each step in the functional architecture of the computer program, so that the entire system is called through a special software package (e.g., a lib library) to achieve the corresponding functions.

[0137] On the other hand, when the above system is implemented in software, firmware, middleware or microcode, the program code or code segment used to perform the corresponding operation can be stored in a computer-readable medium such as a storage medium, so that at least one processor or at least one computing device can perform the corresponding operation by reading and running the corresponding program code or code segment.

[0138] According to exemplary embodiments of this disclosure, the storage device may be integrated with the computing device, for example, by arranging RAM or flash memory within an integrated circuit microprocessor. Alternatively, the storage device may include a separate device, such as an external disk drive, a storage array, or other storage device usable by any database system. The storage device and the computing device may be operatively coupled, or may communicate with each other, for example, via I / O ports, network connections, etc., enabling the computing device to read instructions stored in the storage device.

[0139] Another embodiment of this disclosure relates to a computer program product, including a computer program / instructions that, when executed by a processor, implement the blood pressure measurement method based on a multidimensional transformation neural network model as described above.

[0140] According to the blood pressure measurement method, device, and electronic device based on the multidimensional transformation neural network model provided in this disclosure, by capturing the pulse wave signal of the subject and using a series of dimension transformation techniques and feature extraction techniques to process the pulse wave signal, it is possible not only to capture the instantaneous fluctuations in the pulse wave signal, but also to grasp the overall trend of the pulse wave signal, thereby achieving rapid and accurate blood pressure measurement.

[0141] Furthermore, based on the photoplethysmography (PPG) pulse wave signal, after obtaining a high-quality pulse wave signal through data processing methods such as cleaning and optimization, and then undergoing signal quality screening, an effective pulse wave signal can be obtained. This effective pulse wave signal undergoes a series of dimensional transformations, converting a one-dimensional signal into multiple two-dimensional signals through various dimensional transformations. This allows for the extraction of richer information. Combined with feature extraction processing, it can further and more accurately capture the instantaneous fluctuations of the pulse wave signal and grasp the overall trend, thereby achieving faster and more accurate blood pressure measurement. In addition, it can maintain continuous monitoring capabilities, making it suitable for application in miniaturized, wearable blood pressure monitoring devices, facilitating frequent use and improving the portability and comfort of the blood pressure measurement process.

[0142] The foregoing has described various exemplary embodiments of this disclosure. It should be understood that the foregoing description is exemplary only and not exhaustive, and this disclosure is not limited to the disclosed exemplary embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of this disclosure. Therefore, the scope of protection of this disclosure should be determined by the scope of the claims.

Claims

1. A blood pressure measurement method based on a multi-dimensional transform neural network model, characterized in that, The method includes: Acquire the pulse wave signal of the subject; The pulse wave signal is input into the multidimensional transform neural network model to obtain the blood pressure value. The multidimensional transform neural network model includes an extremum-based dimensional transform block, a fast Fourier transform-based dimensional transform block, and a one-dimensional deep residual contraction network. The extremum-based dimensional transform block converts the pulse wave signal into a first signal, and then the fast Fourier transform-based dimensional transform block converts the first signal into a second signal. The one-dimensional deep residual contraction network then uses the second signal to obtain the blood pressure value. The extremum-based dimensional transformation block performs the following operations: converting the pulse wave signal into a two-dimensional pulse wave signal based on the extremum-based dimensional transformation; inputting the two-dimensional pulse wave signal into a residual network with the last layer of connections removed to obtain a first output; performing pooling processing on the first output to convert it into a one-dimensional pulse wave signal; and performing residual linking on the one-dimensional pulse wave signal and the pulse wave signal to obtain the first signal. The dimensional transformation block based on Fast Fourier Transform (FFT) is used to perform the following operations: normalize the first signal output to obtain a second output; perform multiple dimensional transformations based on FFT on the second output to obtain a third output; and perform residual linking of the third output with the pulse wave signal to obtain the second signal. The step of acquiring the pulse wave signal of the subject includes: acquiring the photoplethysmography (PPG) signal of the subject; performing data cleaning on the PPG signal to obtain a first pulse wave signal; and extracting a second pulse wave signal that meets preset conditions by performing a quality check on the first pulse wave signal. The step of inputting the pulse wave signal into the multidimensional transform neural network model includes: inputting the second pulse wave signal into the multidimensional transform neural network model. The step of extracting a second pulse wave signal that meets preset conditions by performing a quality check on the first pulse wave signal includes: The outlier detection method based on kernel density estimation is used to detect outliers in the pulse wave signal within each window of the first pulse wave signal, wherein each window is obtained by pre-segmenting the first pulse wave signal. By extracting the pulse wave signal from the window containing no outliers in the first pulse wave signal, a second pulse wave signal that satisfies the preset conditions is obtained. The outlier detection method based on kernel density estimation detects outliers in each window of the first pulse wave signal, including: Based on the kernel density estimation formula using a smooth and symmetric Gaussian kernel, the quality score of each data point within each window of the first pulse wave signal is calculated. The quality score is the reciprocal of the kernel density estimate and is used to determine whether the data point is an outlier. The kernel density estimation formula is expressed as follows: in, n This refers to all data point samples in the first pulse wave signal. ω For bandwidth parameters, x The following features are defined as at least one of the following for each data point within each window of the first pulse wave signal: mean, variance, skewness, kurtosis, rising amplitude, diphtheria amplitude, falling mid-wave amplitude, and area under the curve.

2. The blood pressure measurement method based on a multi-dimensional transform neural network model as described in claim 1, characterized in that, The dimensional transformation based on extrema converts the pulse wave signal into a two-dimensional pulse wave signal, including: Identify the maximum and minimum values ​​in each cycle of the pulse wave signal; Taking the maximum or minimum value in each cycle as the starting point, multiple signal segments with a length of a first preset threshold are extracted from the pulse wave signal. Each of the signal segments is stacked as a row of a two-dimensional matrix to obtain the two-dimensional pulse wave signal.

3. The blood pressure measurement method based on a multi-dimensional transform neural network model as described in claim 1, characterized in that, The step of data cleaning the photoplethysmography (PPG) signal to obtain the first PPG signal includes: The photoplethysmography pulse wave signal is low-pass filtered to obtain the third pulse wave signal; The third pulse wave signal is subjected to discrete wavelet transform, and the low-frequency component coefficients of the first layer obtained during the discrete wavelet transform process are eliminated to obtain the first pulse wave signal.

4. A blood pressure measuring device based on a multi-dimensional transform neural network model, characterized in that, The device includes: The signal acquisition unit is configured to acquire the pulse wave signal of the subject. A blood pressure calculation unit is configured to: input the pulse wave signal into the multidimensional transform neural network model to obtain a blood pressure value. The multidimensional transform neural network model includes an extremum-based dimensional transform block, a fast Fourier transform-based dimensional transform block, and a one-dimensional deep residual contraction network. The extremum-based dimensional transform block converts the pulse wave signal into a first signal, and then the fast Fourier transform-based dimensional transform block converts the first signal into a second signal, so that the one-dimensional deep residual contraction network obtains the blood pressure value based on the second signal. The blood pressure calculation unit is configured to: convert the pulse wave signal into a two-dimensional pulse wave signal based on a dimensional transformation of the maximum and minimum values; input the two-dimensional pulse wave signal into a residual network with the last layer of connections removed to obtain a first output; perform pooling processing on the first output to convert it into a one-dimensional pulse wave signal; and perform residual linking on the one-dimensional pulse wave signal and the pulse wave signal to obtain the first signal. The blood pressure calculation unit is configured to: normalize the first signal output to obtain a second output; perform multiple dimensional transformations based on Fast Fourier Transform on the second output to obtain a third output; and perform residual linking of the third output with the pulse wave signal to obtain the second signal. The signal acquisition unit is configured to: acquire the photoplethysmography (PPG) signal of the subject; perform data cleaning on the PPG signal to obtain a first pulse wave signal; and extract a second pulse wave signal that meets preset conditions by performing a quality check on the first pulse wave signal. The blood pressure calculation unit is configured to input the second pulse wave signal into the multidimensional transform neural network model. The signal acquisition unit is configured to: perform outlier detection on the pulse wave signals within each window of the first pulse wave signal using a kernel density estimation-based outlier detection method, wherein each window is obtained by pre-segmenting the first pulse wave signal; and obtain the second pulse wave signal that satisfies the preset conditions by extracting the pulse wave signals from the windows of the first pulse wave signal that do not contain outliers. The signal acquisition unit is configured to: calculate the quality score of each data point within each window of the first pulse wave signal based on a kernel density estimation formula using a smooth and symmetric Gaussian kernel, wherein the quality score is the reciprocal of the kernel density estimate and is used to determine whether the data point is an outlier; the kernel density estimation formula is expressed as: in, n This refers to all data point samples in the first pulse wave signal. ω For bandwidth parameters, x The following features are defined as at least one of the following for each data point within each window of the first pulse wave signal: mean, variance, skewness, kurtosis, rising amplitude, diphtheria amplitude, falling mid-wave amplitude, and area under the curve.

5. An electronic device, characterized in that, include: At least one processor; At least one memory that stores computer-executable instructions. Wherein, when the computer-executable instructions are executed by the at least one processor, the at least one processor causes the at least one processor to execute the blood pressure measurement method based on a multidimensional transformation neural network model as described in any one of claims 1-3.

Images