Blood pressure assessment method based on relation attention mechanism

Abstract

The invention discloses a blood pressure assessment method based on a relational attention mechanism, belongs to the technical field of blood pressure assessment, and solves the problems that when blood pressure is assessed by combining photoelectric volume pulse waves and electrocardiogram signals in an existing method, the signal recognition process and the final assessment performance are hindered due to the high sensitivity of the PPG signals to noise. The method comprises the steps of collecting blood pressure related signals in real time, preprocessing the blood pressure related signals, extracting signal key points in the blood pressure related signals, constructing a blood pressure map topology based on the signal key points, recognizing and processing the blood pressure map topology by a map neural network model based on time-space dependence of physiological signals, and outputting a blood pressure evaluation value. According to the embodiment of the invention, the graph neural network model based on the relation attention mechanism is provided, the graph neural network model is combined with signal preprocessing, graph construction and the relation attention mechanism, the hidden information is extracted from the graph topological data by enhancing the feature extraction capability, and the accuracy of blood pressure estimation is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of blood pressure assessment, and particularly relates to a blood pressure assessment method based on a relational attention mechanism. Background Art

[0002] At present, cardiovascular diseases (CVDs) remain one of the main factors leading to high morbidity and mortality of chronic non-communicable diseases globally. Blood pressure (BP), as a key physiological indicator in health monitoring, its accurate measurement is crucial for the early detection and management of cardiovascular diseases. However, existing blood pressure measurement methods have many limitations and urgently need improvement and innovation.

[0003] Currently, the cuff-based non-invasive blood pressure monitoring method is the most widely used blood pressure measurement means. Although its measurement results are relatively accurate, during 24-hour ambulatory blood pressure monitoring, the repeated inflation and deflation of the cuff will not only interfere with the patient's daily activities but also bring discomfort to the patient. In addition, the cuff-type sphygmomanometer cannot provide continuous blood pressure data and is difficult to meet the requirements of clinical and daily health monitoring for continuous blood pressure monitoring. Therefore, developing a non-invasive blood pressure measurement method without a cuff has become an urgent need in the current blood pressure monitoring field.

[0004] In recent years, impedance-based measurement methods have been proposed as a potential solution. However, establishing a reliable, durable, and stable contact quality between the electrode and the skin is a major challenge, which requires continuous attention to the placement and maintenance of the sensor. In addition, millimeter-wave-based blood pressure estimation methods mainly rely on single-channel pulse wave analysis, but due to limitations in accuracy and portability, it is difficult to develop into a widely used technology. The pulse transit time (PTT) method usually combines photoplethysmogram (PPG) and electrocardiogram (ECG) signals to evaluate blood pressure. Although the use of PPG signals is becoming more and more common in smart wearable devices, the high sensitivity of PPG signals to noise (such as motion artifacts and environmental light interference) may seriously hinder the signal recognition process and the final evaluation performance. Moreover, significant individual differences and the influence of physiological and environmental factors still exist, which limit the accuracy of existing methods. To address the above problems, we propose a blood pressure assessment method based on a relational attention mechanism. Summary of the Invention

[0005] The purpose of the present invention is to provide a blood pressure assessment method based on a relational attention mechanism for the deficiencies of the existing technology, and solve the problem that when the existing method combines photoplethysmogram and electrocardiogram signals to evaluate blood pressure, the high sensitivity of the PPG signal to noise hinders the signal recognition process and the final evaluation performance.

[0006] The present invention is implemented as follows. The blood pressure assessment method based on a relational attention mechanism includes: Collect blood pressure related signals in real time, preprocess the blood pressure related signals, and extract signal key points from the blood pressure related signals. Among them, the blood pressure related signals include the original PPG signal and the original ECG signal; Load the signal key points of the blood pressure related signals, and construct a blood pressure map topology based on the signal key points; Pre-construct a graph neural network model based on the relational attention mechanism, iteratively train the graph neural network model, and output a converged graph neural network model; Load the blood pressure map topology, and the graph neural network model identifies and processes the blood pressure map topology based on the spatio-temporal dependence of the physiological signals, and outputs a blood pressure evaluation value.

[0007] The method for preprocessing the blood pressure related signals includes: Load the blood pressure related signals, denoise and baseline correct the blood pressure related signals to obtain a corrected signal set; Obtain the corrected signal set, and perform segmentation processing on the corrected signal set based on the sliding window method to divide the corrected signal set into at least one group of overlapping time windows; Perform EMD decomposition on the signals within each group of the time windows to separate the intrinsic mode functions; Among them, when using EMD decomposition, the formula is expressed as:

[0008] Among them, represents the original signal at time , is the th intrinsic mode function (IMF) of the signal. In order to remove noise and motion artifacts, the middle IMFs are retained, while the IMFs related to noise and motion artifacts are discarded; Load the corrected signal set after EMD decomposition, and reconstruct the corrected signal set after decomposition and noise reduction to obtain a signal reconstruction set; The signal reconstruction set is expressed as:

[0009] Among them is the set of modal function IMFs retained for reconstruction.

[0010] The method for preprocessing the blood pressure related signals further includes: Extract key points from the signals within the time window. Among them, use the AMPD algorithm to extract key points for the PPG signal in the signal reconstruction set, and use the Neurokit2 library to extract key points for the ECG signal in the signal reconstruction set to obtain PPG signal key points and ECG signal key points; The key points of the PPG signal are denoted as a, b, c, d, e, and the key points of the ECG signal are denoted as P, Q, R, S, T.

[0011] The time characteristics and amplitude characteristics of the key points of the PPG signal and the key points of the ECG signal within the time window are represented by the following formulas:

[0012] where, represents the average time characteristic within the i-th window, is a function for calculating the time difference between consecutive key points, is the timestamp of the key point within the i-th window; represents the average normalized amplitude characteristic within the i-th window, is a function for normalizing and averaging the amplitudes of the key points, is the set of amplitudes of the key points within the i-th window.

[0013] When collecting blood pressure-related signals in real time, the original PPG signal is obtained through a fingertip sensor, the original ECG signal is recorded through a standard lead configuration, and the blood pressure-related signals are sampled at a frequency of 125 Hz and a precision of 12 bits.

[0014] The method for constructing a blood pressure map topology based on signal key points includes: Loading the key points of the PPG signal and the key points of the ECG signal, using the key points of the PPG signal and the key points of the ECG signal as the nodes of the blood pressure map topology, and encoding the time relationship between the key points of the PPG signal and the key points of the ECG signal as the edges of the blood pressure map topology; where, let represent the blood pressure map topology of the i-th window, where is the set of nodes (key points of the PPG signal and key points of the ECG signal), is the set of edges (time relationship between the key points of the PPG signal and key points of the ECG signal), and each node is associated with a feature vector and each edge is associated with a feature vector ; The feature definitions of the nodes and edges of the blood pressure map topology are as follows:

[0015] where, is the timestamp of node , is the amplitude of the key point is the time difference between consecutive key points, is node The timestamp; Update the features of the topological nodes and edges of the blood pressure graph based on the graph-based relational attention mechanism; Node The features are updated by aggregating the information of its neighbor nodes. The aggregation process calculates the mean, maximum, minimum, and standard deviation of the neighbor node features to form a feature vector, which is then passed through a multi-layer perceptron MLP to obtain the updated node features. Edge The features are updated by aggregating the information of its connected nodes. The aggregation process calculates the mean, maximum, minimum, and standard deviation of the connected node features to form a feature vector, which is also passed through the MLP to obtain the updated edge features.

[0016] The graph neural network model includes a first relational attention module, a second relational attention module, an average pooling layer, a softmax layer, and a batch normalization layer. The activation function of the graph neural network model is the Relu function. Two sets of batch normalization layers are provided, which are respectively arranged after the first relational attention module and the second relational attention module.

[0017] The method for the graph neural network model to identify and process the topology of the blood pressure graph based on the spatio-temporal dependence of physiological signals includes: Load the blood pressure graph topology and input it into the graph neural network model. The first relational attention module updates the node features by aggregating information from neighboring nodes and considering the attributes of the edges. The updated features are normalized to stabilize the training, and a non-linear activation function is used to enhance the expressive power of the graph neural network model; The second relational attention module further refines the representation of the node features. The refined node features are obtained through the mean pooling operation of the average pooling layer. The pooling operation aggregates the node-level features into a single vector representation; Load the vector representation and perform identification processing on the vector representation through the softmax layer to predict and generate a blood pressure evaluation value; Among them, the prediction and generation of the blood pressure evaluation value are calculated by the following formula:

[0018] Among them, X represents the input node features, A is the adjacency matrix, And Are trainable weight matrices, Represents the activation function, And Are used to map the pooled representation to the final predicted blood pressure evaluation value.

[0019] Compared with the prior art, the embodiments of the present application mainly have the following beneficial effects: In the embodiments of the present invention, a graph neural network model based on a relational attention mechanism is provided. The graph neural network model combines signal preprocessing, graph construction, and a relational attention mechanism. By enhancing the feature extraction ability, hidden information is extracted from graph topology data, improving the accuracy of blood pressure estimation. When predicting and generating blood pressure evaluation values, the mean absolute errors (MAEs) in diastolic blood pressure (DBP) and systolic blood pressure (SBP) estimations are 1.3520 mmHg and 2.9862 mmHg respectively, meeting the standards of the American Association for the Advancement of Medical Instrumentation (AAMI) and the Grade A standard of the British Hypertension Society (BHS).

[0020] In the embodiments of the present invention, the AMPD algorithm detects local maxima through a multi-scale sliding window and can handle complex signal structures, which is very useful for analyzing multiple frequency components and non-linear features in the PPG signal. Combining the key point extraction of the PPG signal and the ECG signal enables more comprehensive physiological signal analysis.

[0021] In the embodiments of the present invention, by accurately extracting the key points of the PPG signal and the ECG signal, important features in the signal can be captured more precisely. At the same time, by constructing a blood pressure graph topology, the complex relationships between signals can be comprehensively captured, improving the accuracy of signal analysis. The anti-noise and stability characteristics of the AMPD algorithm and the Neurokit2 library make the key point extraction more robust. The relational attention mechanism of the graph can effectively handle noise and outliers, further improving the robustness of signal processing. By constructing a blood pressure graph topology, more comprehensive physiological signal analysis can be achieved, providing strong support for the early detection and management of cardiovascular diseases. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a schematic diagram of the implementation process of the blood pressure assessment method based on the relational attention mechanism provided by the present invention.

[0023] Figure 2 It shows a Bland-Altman plot of the DBP estimation of the graph neural network model in the embodiments of the present invention.

[0024] Figure 3 It shows a Bland-Altman plot of the SBP estimation of the graph neural network model in the embodiments of the present invention.

[0025] Figure 4 It shows a histogram analysis of the DBP estimation error of the graph neural network model in the embodiments of the present invention.

[0026] Figure 5 It shows a histogram analysis of the SBP estimation error of the graph neural network model in the embodiments of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0027] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the technical field to which this application belongs; the terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" and any variations thereof in the specification and claims of this application and the above drawings are intended to cover non-exclusive inclusion. The terms "first", "second", etc. in the specification and claims of this application or the above drawings are used to distinguish different objects and not to describe a specific order.

[0028] When existing methods combine photoplethysmogram (PPG) and electrocardiogram (ECG) signals to evaluate blood pressure, the high sensitivity of PPG signals to noise hinders the signal recognition process and the final evaluation performance. To address the above problems, we propose a blood pressure evaluation method based on a relational attention mechanism. Briefly, when implementing the method, blood pressure-related signals are first collected in real time, preprocessed, signal key points in the blood pressure-related signals are extracted, a blood pressure map topology is constructed based on the signal key points, a graph neural network model based on the relational attention mechanism is pre-constructed, and finally the graph neural network model processes the blood pressure map topology based on the spatio-temporal dependence of physiological signals to output a blood pressure evaluation value. In the embodiments of the present invention, a graph neural network model based on the relational attention mechanism is provided. The graph neural network model combines signal preprocessing, graph construction, and the relational attention mechanism, extracts hidden information from graph topology data by enhancing the feature extraction ability, and improves the accuracy of blood pressure estimation.

[0029] Embodiments of the present invention provide a blood pressure evaluation method based on a relational attention mechanism. Figure 1 The schematic implementation flowchart of the blood pressure evaluation method based on the relational attention mechanism is shown. The blood pressure evaluation method based on the relational attention mechanism includes: S10, Collect blood pressure-related signals in real time, preprocess the blood pressure-related signals, and extract signal key points in the blood pressure-related signals, where the blood pressure-related signals include the original PPG signal and the original ECG signal; It should be noted that when collecting blood pressure-related signals in real time, the original PPG signal is obtained through a fingertip sensor, the original ECG signal is recorded through a standard lead configuration, and the blood pressure-related signals are sampled at a frequency of 125 Hz and a precision of 12 bits.

[0030] S20, Load the signal key points of the blood pressure-related signals and construct a blood pressure map topology based on the signal key points; S30, Pre-construct a graph neural network model based on the relational attention mechanism, iteratively train the graph neural network model, and output a converged graph neural network model; S40. Load the blood pressure map topology. The graph neural network model identifies and processes the blood pressure map topology based on the spatio-temporal dependence of physiological signals, and outputs a blood pressure assessment value.

[0031] In the embodiments of the present invention, the method for preprocessing blood pressure-related signals includes: S101. Load the blood pressure-related signals, perform denoising and baseline correction on the blood pressure-related signals, and obtain a corrected signal set; In the embodiments of the present invention, the original PPG signal and the original ECG signal are preprocessed to remove noise and artifacts, ensuring high-quality input to the graph neural network model. The preprocessing steps include, but are not limited to, denoising, baseline correction, and exclusion of irregular or discontinuous signal segments. In addition, the EMD decomposition algorithm is applied to further reduce motion artifacts and enhance signal quality.

[0032] S102. Obtain the corrected signal set, and perform segmentation processing on the corrected signal set based on the sliding window method, and segment the corrected signal set into at least one group of overlapping time windows; It should be noted that the width of each time window is 3 seconds, and the sliding step is 1.5 seconds.

[0033] S103. Use EMD decomposition on the signals within each group of the time windows to separate the intrinsic mode functions; Among them, when using EMD decomposition, the formula is expressed as:

[0034] Among them, represents the original signal at time , is the th intrinsic mode function (IMF) of the signal. In order to remove noise and motion artifacts, the middle IMFs are retained, while the IMFs related to noise and motion artifacts are discarded; S104. Load the corrected signal set after EMD decomposition, and reconstruct the corrected signal set after decomposition and noise reduction to obtain a signal reconstruction set; The signal reconstruction set is expressed as:

[0035] Among them is the set of modal functions IMFs reserved for reconstruction.

[0036] S105. Extract key points from the signals within the time window. Among them, for the PPG signal in the signal reconstruction set, the AMPD algorithm is used to extract key points, and for the ECG signal in the signal reconstruction set, the Neurokit2 library is used to extract key points, obtaining PPG signal key points and ECG signal key points; It should be noted that in the embodiments of the present invention, when extracting key points of the PPG signal based on the Automatic Multi-scale Peak Detection (AMPD) algorithm, the automatic multi-scale peak detection is an efficient peak detection algorithm, which is particularly suitable for periodic or quasi-periodic signals. This algorithm detects local maxima through a multi-scale sliding window and has good adaptability and anti-noise ability. For the PPG signal, the AMPD algorithm can accurately extract the key points (such as peaks and troughs) of the signal, thereby improving the accuracy of signal analysis. Automatic multi-scale peak detection.

[0037] The key points of the PPG signal are denoted as a, b, c, d, e, and the key points of the ECG signal are denoted as P, Q, R, S, T.

[0038] In the embodiments of the present invention, the AMPD algorithm detects local maxima through a multi-scale sliding window and can handle complex signal structures, which is very useful for analyzing multiple frequency components and non-linear features in the PPG signal. Combining the extraction of key points of the PPG signal and the ECG signal can achieve a more comprehensive physiological signal analysis. For example, by analyzing the key points of the PPG and ECG signals, the functional state of the cardiovascular system can be evaluated more accurately.

[0039] In the embodiments of the present invention, the time characteristics and amplitude characteristics of the key points of the PPG signal and the ECG signal within the time window are represented by the following formulas:

[0040] where represents the average time characteristic within the i-th window, is a function for calculating the time difference between consecutive key points, is the time stamp of the key points within the i-th window; represents the average normalized amplitude characteristic within the i-th window, is a function for normalizing and averaging the amplitudes of the key points

[0041] The embodiments of the present invention provide a method for constructing a blood pressure map topology based on signal key points. The method for constructing a blood pressure map topology based on signal key points specifically includes: S201, loading the key points of the PPG signal and the key points of the ECG signal, using the key points of the PPG signal and the key points of the ECG signal as the nodes of the blood pressure map topology, and encoding the time relationship between the key points of the PPG signal and the key points of the ECG signal as the edges of the blood pressure map topology; where, let represent the blood pressure map topology of the i-th window, where is the set of nodes (key points of the PPG signal and key points of the ECG signal), is a set of edges (the time relationship between PPG signal key points and ECG signal key points), and each node is associated with a feature vector Each edge is associated with a feature vector associated; The feature definitions of the blood pressure map topology nodes and edges are as follows:

[0042] Among them, is the timestamp of node , is the amplitude of the key point is the time difference between consecutive key points, is the timestamp of node ; S202, update the features of the blood pressure map topology nodes and edges based on the graph-based relational attention mechanism; Among them, the features of node are updated by aggregating the information of its neighbor nodes. The aggregation process calculates the mean, maximum, minimum, and standard deviation of the neighbor node features to form a feature vector, which is then passed through a multi-layer perceptron MLP to obtain the updated node features. The features of edge are updated by aggregating the information of its connected nodes. The aggregation process calculates the mean, maximum, minimum, and standard deviation of the connected node features to form a feature vector, which is also passed through the MLP to obtain the updated edge features.

[0043] It should be noted that the blood pressure map topology construction process involves mapping the key points identified from the PPG and ECG signals to nodes and encoding the time relationship between these points as edges in the time graph. At the same time, updating the features of the blood pressure map topology nodes and edges based on the graph-based relational attention mechanism is different from the standard transformer attention, which only calculates the query (Q), key (K), and value (V) vectors from the node features. Relational attention incorporates edge features into the calculation of Q, K, and V. This is achieved by concatenating the edge vector with the node vector before applying the linear transformation.

[0044] In the embodiments of the present invention, by accurately extracting the key points of the PPG signal and the ECG signal, important features in the signals can be captured more precisely. At the same time, by constructing the blood pressure map topology, the complex relationships between the signals can be comprehensively captured, improving the accuracy of signal analysis. The anti-noise and stability characteristics of the AMPD algorithm and the Neurokit2 library make the key point extraction more robust. The relational attention mechanism of the graph can effectively handle noise and outliers, further improving the robustness of signal processing. By constructing the blood pressure map topology, a more comprehensive physiological signal analysis can be achieved, providing strong support for the early detection and management of cardiovascular diseases.

[0045] In the embodiments of the present invention, the graph neural network model includes a first relational attention module, a second relational attention module, an average pooling layer, a softmax layer, and a batch normalization layer. The activation function of the graph neural network model is the Relu function. Two groups of batch normalization layers are provided, which are respectively arranged after the first relational attention module and the second relational attention module.

[0046] Specifically, as Figure 1 shown, the first relational attention module is connected to the batch normalization layer. The batch normalization layer is connected to the second relational attention module through the Relu function. The second relational attention module is connected to the other group of batch normalization layers. The batch normalization layer is connected to the average pooling layer, and the average pooling layer is connected to the softmax layer.

[0047] When pre-constructing a graph neural network model based on the relational attention mechanism and iteratively training the graph neural network model, the training data set of the graph neural network model comes from the UCI-BP data set. The PPG signal is obtained through a fingertip sensor, and the ECG signal is recorded through a standard lead configuration. All signals are sampled at a frequency of 125 Hz and a precision of 12 bits. This data set has been preliminarily preprocessed, including denoising, baseline correction, and excluding irregular or discontinuous signal segments. This data set is used to verify the performance of the proposed graph neural network model, including the comparison of the results of physiological data preprocessing and the discussion of the effectiveness of the blood pressure estimation algorithm to support the effectiveness of the proposed method.

[0048] The performance of the proposed graph neural network model is evaluated through regression analysis, Bland-Altman analysis, and error distribution histograms.

[0049] Specifically, the graph neural network model is used to estimate DBP estim and SBP estim .

[0050] Regression analysis: For DBP estim and SBP estimPerform a regression analysis and compare them with the reference values. The one-dimensional linear fitting results of the data are shown below:

[0051] Compared with the ideal linear equation, it can be observed that DBP estim shows the strongest linear correlation, and its proportionality coefficient (kkk) is 0.9709. Followed by SBP estim , whose kkk value is 0.9232. Table 1 shows the performance evaluation comparison results between the graph neural network model and the benchmark model of the embodiments of the present invention.

[0052] Table 1

[0053] Bland-Altman analysis: To evaluate the consistency and variability of blood pressure measurement, Figures 2 - 3 shows the results of the Bland-Altman analysis, in which, Figure 2 shows the Bland-Altman plot of the DBP estimation of the graph neural network model in the embodiments of the present invention, Figure 3 shows the Bland-Altman plot of the SBP estimation of the graph neural network model in the embodiments of the present invention. Among them, the annotations in the figure are:

[0054] And, the mean blood pressure ( ), and the blood pressure difference ( ) are calculated as follows:

[0055] The statistical values of the blood pressure estimation error indicate that the standard deviation (STD) follows the pattern:

[0056] This is consistent with the relevant trend observed in blood pressure measurement. The Bland-Altman plot further verifies the consistency of the estimated blood pressure values, and most of the points are concentrated around , indicating a strong consistency between the estimated blood pressure values and the reference blood pressure values. In addition, within the range of the mean error plus 1.96 times the standard deviation, the sample percentages of DBP and SBP are 95.94% and 95.09% respectively, both exceeding the 95% standard. In addition, the results obtained using the UCI-BP dataset meet the AAMI standard, which stipulates that the mean absolute error (MAE) should not exceed 5 mmHg and the standard deviation should be less than 8 mmHg.

[0057] Estimation error histogram analysis: In the present invention, the BP estimation error distribution of 1200 data points is analyzed through a histogram, and the results are as follows Figures 4 - 5 shown Figure 4 showing the histogram analysis of the DBP estimation error of the graph neural network model in the embodiments of the present invention Figure 5 showing the histogram analysis of the SBP estimation error of the graph neural network model in the embodiments of the present invention. In addition, Table 1 provides the cumulative percentage (CP) of the absolute value of the estimation error, which is evaluated according to the British Hypertension Society (BHS) standard. According to the BHS standard, the proposed BP estimation method is consistent with the A-level accuracy standard. Specifically, for SBP and DBP estimations, the CP of this method exceeds 60% within ±5 mmHg, exceeds 85% within ±10 mmHg, and exceeds 95% within ±15 mmHg. The error percentages of SBP and DBP are significantly higher than the values specified by the BHS standard.

[0058] The embodiments of the present invention also provide a method for the graph neural network model to identify and process the topology of the blood pressure map based on the spatio-temporal dependence of physiological signals. The method for the graph neural network model to identify and process the topology of the blood pressure map based on the spatio-temporal dependence of physiological signals specifically includes: S301, load the blood pressure map topology, input the blood pressure map topology into the graph neural network model. The first relational attention module updates the node features by aggregating information from neighboring nodes and considering the attributes of the edges at the same time. The updated features are normalized to stabilize the training, and the expression ability of the graph neural network model is enhanced through a non-linear activation function; S302, the second relational attention module further refines the representation of the node features. The refined node features are obtained through the mean pooling operation of the average pooling layer. The pooling operation aggregates the node-level features into a single vector representation; S303, load the vector representation, and perform identification processing on the vector representation through the softmax layer to predict and generate a blood pressure evaluation value; Among them, the predicted blood pressure evaluation value is calculated by the following formula:

[0059] where X represents the input node features, A is the adjacency matrix, and are trainable weight matrices, represents the activation function, and are used to map the pooled representation to the final predicted blood pressure evaluation value.

[0060]

[0061] The first relational attention module and the second relational attention module iteratively refine the node embeddings by considering the structural and relational dependencies, ensuring that temporal and context information is effectively captured. In the embodiments of the present invention, a method for identifying and processing the topology of a blood pressure map based on the spatio-temporal dependencies of physiological signals in a graph neural network model is proposed, which is used for cuffless blood pressure estimation using synchronized PPG and ECG signals. Moreover, the proposed graph neural network model utilizes graph-based representations and relational attention mechanisms to effectively capture spatio-temporal dependencies, achieving superior accuracy and robustness under the AAMI standard.

[0062] In summary, the present invention provides a blood pressure assessment method based on a relational attention mechanism. In the embodiments of the present invention, a graph neural network model based on a relational attention mechanism is provided. The graph neural network model combines signal preprocessing, graph construction, and a relational attention mechanism, extracts hidden information from graph topology data by enhancing the feature extraction ability, and improves the accuracy of blood pressure estimation. When predicting and generating blood pressure assessment values, the mean absolute errors (MAEs) in the estimation of diastolic blood pressure (DBP) and systolic blood pressure (SBP) are 1.3520 mmHg and 2.9862 mmHg respectively, meeting the standards of the American Association for the Advancement of Medical Instrumentation (AAMI) and the British Hypertension Society (BHS) Grade A standard.

[0063] It should be noted that for the foregoing embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should be aware that the present invention is not limited by the described action sequence, because according to the present invention, certain steps may be performed in other sequences or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0064] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the protection scope of the invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on these embodiments, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art can still, without conflict, make combinations, additions, deletions, or other adjustments to the features in the embodiments of the present invention according to the situation without creative efforts, so as to obtain different technical solutions that essentially do not deviate from the concept of the present invention, and these technical solutions also fall within the scope of protection of the present invention.

Claims

1. A blood pressure assessment method based on a relational attention mechanism, characterized in that Comprising: Real-time collect blood pressure related signals, preprocess the blood pressure related signals, and extract signal key points in the blood pressure related signals, wherein the blood pressure related signals include the original PPG signal and the original ECG signal; Load the signal key points of the blood pressure related signals, and construct a blood pressure map topology based on the signal key points; Pre-construct a graph neural network model based on the relational attention mechanism, iteratively train the graph neural network model, and output a converged graph neural network model; Load the blood pressure map topology, and the graph neural network model identifies and processes the blood pressure map topology based on the spatio-temporal dependence of physiological signals, and outputs a blood pressure evaluation value.

2. The blood pressure assessment method based on the relational attention mechanism according to claim 1, wherein: The method for preprocessing the blood pressure related signals includes: Load the blood pressure related signals, perform denoising and baseline correction processing on the blood pressure related signals to obtain a corrected signal set; Obtain the corrected signal set, and perform segmentation processing on the corrected signal set based on the sliding window method to divide the corrected signal set into at least one group of overlapping time windows; Perform EMD decomposition on the signals within each group of the time windows to separate the intrinsic mode functions; Wherein, when using EMD decomposition, the formula is expressed as: Among them, indicating the moment of the original signal, is the th intrinsic mode function of the signal. In order to remove noise and motion artifacts, the middle IMFs are retained, while the IMFs related to noise and motion artifacts are discarded; Load the corrected signal set after EMD decomposition, reconstruct the corrected signal set after decomposition and noise reduction to obtain a signal reconstruction set; The signal reconstruction set is expressed as: Among them is the set of modal functions IMFs reserved for reconstruction.

3. The blood pressure assessment method based on the relational attention mechanism according to claim 2, wherein: The method for preprocessing the blood pressure related signals further includes: Extract key points of the signals within the time window. Among them, use the AMPD algorithm to extract key points for the PPG signal in the signal reconstruction set, and use the Neurokit2 library to extract key points for the ECG signal in the signal reconstruction set to obtain PPG signal key points and ECG signal key points; The PPG signal key points are expressed as a, b, c, d, e, and the ECG signal key points are expressed as P, Q, R, S, T.

4. The blood pressure assessment method based on the relational attention mechanism according to claim 3, wherein: The time characteristics and amplitude characteristics of the PPG signal key points and ECG signal key points within the time window are expressed by the following formula: Among them, represents the average time feature within the i-th window, is a function for calculating the time difference between consecutive key points, is the timestamp of the key points within the i-th window; represents the average normalized amplitude feature within the i-th window, is a function that normalizes and averages the key-point amplitudes, is the set of amplitudes of the key points within the i-th window.

5. The blood pressure assessment method based on the relational attention mechanism according to claim 2, characterized in that: When real-time collecting blood pressure related signals, the original PPG signal is obtained through a fingertip sensor, the original ECG signal is recorded through a standard lead configuration, and the blood pressure related signals are sampled at a frequency of 125Hz and a precision of 12 bits.

6. The blood pressure assessment method based on the relational attention mechanism according to claim 4, wherein: The method for constructing a blood pressure map topology based on signal key points includes: Load the PPG signal key points and the ECG signal key points, use the PPG signal key points and the ECG signal key points as the nodes of the blood pressure map topology, and encode the time relationship between the PPG signal key points and the ECG signal key points as the edges of the blood pressure map topology; Among them, let represent the blood pressure map topology of the i-th window, where is a set of nodes, is a set of edges, and each node is associated with a feature vector , and each edge is associated with a feature vector . The characteristics of the nodes and edges of the blood pressure map topology are defined as follows: Among them, is the timestamp of the node , is the amplitude of the key point is the time difference between consecutive key points, is the timestamp of the node . Update the characteristics of the nodes and edges of the blood pressure map topology based on the relational attention mechanism of the graph; Node is updated by aggregating the information of its neighbor nodes. The aggregation process calculates the mean, maximum, minimum, and standard deviation of the neighbor node features to form a feature vector, which is then passed through a multi-layer perceptron (MLP) to obtain the updated node feature. Edge is updated by aggregating the information of its connected nodes. The aggregation process calculates the mean, maximum, minimum, and standard deviation of the connected node features to form a feature vector, which is also passed through the MLP to obtain the updated edge feature.

7. The blood pressure assessment method based on the relational attention mechanism according to claim 6, characterized in that: The graph neural network model includes a first relational attention module, a second relational attention module, an average pooling layer, a softmax layer, and a batch normalization layer. The activation function of the graph neural network model is the Relu function, and two groups of batch normalization layers are provided, which are respectively arranged after the first relational attention module and the second relational attention module.

8. The blood pressure assessment method based on the relational attention mechanism according to claim 7, wherein: The method for the graph neural network model to identify and process the blood pressure map topology based on the spatio-temporal dependence of physiological signals includes: Load the blood pressure graph topology and input it into the graph neural network model. The first relational attention module updates the node features by aggregating information from neighboring nodes and considering the attributes of the edges. The updated features are normalized to stabilize the training, and the expression ability of the graph neural network model is enhanced through a non-linear activation function; The second relational attention module further refines the representation of the node features. The refined node features are obtained through the mean pooling operation of the average pooling layer, and the pooling operation aggregates the node-level features into a single vector representation; Load the vector representation and perform recognition processing on the vector representation through the softmax layer to predict and generate the blood pressure evaluation value; Among them, the predicted blood pressure evaluation value is calculated by the following formula: Among them, X represents the input node features, and A is the adjacency matrix, and are trainable weight matrices, represents the activation function, and are used to map the pooled representation to the final predicted blood pressure assessment value.

Images