Cardiovascular risk early warning method based on multi-channel pulse and ai and related equipment

Abstract

This application relates to the fields of medical artificial intelligence and signal processing technology, and provides a cardiovascular risk early warning method and related equipment based on multi-channel pulse imaging and AI. It collects pulse data from multiple arterial regions of a user using intelligent sensors; standardizes the data to generate a pulse time-series matrix; constructs a convolutional neural network based on spatiotemporal attention; extracts spatiotemporal pulse features from the pulse time-series matrix using the convolutional neural network; constructs a health atlas based on the spatiotemporal pulse features and the user's health record simulation graph nodes and edges; generates a cardiovascular risk score for the user based on the feature information in the health atlas; and provides risk warnings based on the risk score. By constructing a health atlas based on multi-channel arterial region pulse data, comprehensively integrating spatiotemporal pulse features to generate a risk score, it comprehensively considers multiple aspects of the cardiovascular system's state, accurately reflects the degree of risk, and provides a basis for early intervention in cardiovascular problem prevention based on this score.

Description

Technical Field

[0001] This application relates to the fields of medical artificial intelligence and signal processing technology, specifically to a cardiovascular risk early warning method and related equipment based on multi-channel pulse imaging and AI. Background Technology

[0002] Traditional cardiovascular risk early warning methods typically rely on static physiological indicators detected at discrete time points, or continuous monitoring of single sites, such as blood pressure, blood lipids, and blood glucose. However, the cardiovascular system is a complex whole, and a single indicator often cannot comprehensively reflect cardiovascular health, making it difficult to simultaneously and multidimensionally capture the dynamic correlation information of the cardiovascular system in space (different arteries) and time. For example, some users may have normal blood pressure and blood lipid levels, but may still have cardiovascular risks. Furthermore, traditional methods are mostly static detections, unable to monitor changes in the cardiovascular system in real time and continuously. The occurrence and development of cardiovascular problems is a dynamic process, requiring dynamic monitoring and assessment to promptly identify potential risks. Therefore, existing technologies suffer from low reliability in cardiovascular risk early warning. Summary of the Invention

[0003] This application provides a cardiovascular risk early warning method and related equipment based on multi-channel pulse imaging and AI, which can at least partially solve the problem of low reliability of cardiovascular risk early warning in the prior art.

[0004] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part from practice of this application.

[0005] According to one aspect of this application, a cardiovascular risk early warning method based on multi-channel pulse imaging and AI is provided, comprising: collecting pulse data of multiple arterial regions of a user through intelligent sensors; standardizing the pulse data to generate a pulse temporal matrix; constructing a convolutional neural network based on spatiotemporal attention based on the spatiotemporal information of the pulse temporal matrix, and extracting spatiotemporal pulse features from the pulse temporal matrix through the convolutional neural network; constructing a health atlas based on the spatiotemporal pulse features and nodes and edges of a simulated health record of the user; generating a cardiovascular risk score for the user based on the feature information in the health atlas, and providing risk warning based on the risk score.

[0006] In this application, based on the aforementioned scheme, the standardization processing of the pulse data to generate a pulse time series matrix includes: interpolating the pulse data by constructing an interpolation function to generate a pulse baseline; decomposing the pulse baseline into different frequency levels to generate frequency domain signals; determining statistical parameters based on the frequency domain signals at each frequency level, processing the frequency domain signals by comparing the statistical parameters with frequency thresholds, and reconstructing the processed frequency domain signals back into time domain signals to obtain denoised pulse data; using the starting point of the carotid pulse wave in the denoised pulse data as a reference point, calculating the cross-correlation function between the radial artery and dorsalis pedis artery pulse signals and the carotid pulse wave respectively, and adjusting the signals in the denoised pulse data by shifting based on the time offset determined by the cross-correlation function to generate a three-channel aligned signal; and storing the aligned signal sequentially into the previously allocated memory space according to the pre-planned format of the standardized pulse time series matrix to generate a pulse time series matrix.

[0007] In this application, based on the aforementioned scheme, the step of constructing a convolutional neural network based on spatiotemporal attention using the spatiotemporal information of the pulse time series matrix, and extracting spatiotemporal pulse features from the pulse time series matrix using the convolutional neural network, includes: extracting local features of each feature channel from the pulse time series matrix; calculating spatial factors between the feature channels based on trainable matrix parameters and the local features, and fusing the local features based on the spatial factors to generate fused spatial features; organizing the fused spatial features into a feature sequence suitable for time series analysis, and extracting the hidden state of the current time step from the feature sequence using a gated recurrent unit; and fusing the hidden states of different time steps using residual connections to obtain spatiotemporal pulse features at different time scales.

[0008] In this application, based on the aforementioned scheme, the step of constructing a health graph by simulating graph nodes and edges based on the spatiotemporal pulse characteristics and the user's health record includes: reading the user's health record and extracting preset health indicators from the health record; simulating the spatiotemporal pulse characteristics and the health indicators as graph nodes and calculating the relationship factors between each graph node; and using the relationship factors between the graph nodes as edges of the health graph to generate the health graph.

[0009] In this application, based on the aforementioned scheme, generating a user's cardiovascular risk score based on the feature information in the health graph includes: performing feature propagation on the feature information of graph nodes and their neighboring nodes in the health graph through a graph convolutional neural network, aggregating the feature information of each graph node and its neighboring nodes to generate a first feature of the graph node; aggregating the first features of the graph nodes and their neighboring nodes in the health graph through global pooling to generate a second feature of the graph node; and mapping the second feature to a preset range to generate a user's cardiovascular risk score.

[0010] In this application, based on the aforementioned scheme, the risk warning based on the risk score includes: classifying the risk level corresponding to the risk score based on a preset threshold; and issuing a risk warning through the warning method corresponding to the risk level.

[0011] In this application, based on the aforementioned scheme, after generating a cardiovascular risk score for the user based on the feature information in the health map and issuing a risk warning based on the risk score, the method further includes: generating a warning report based on user information and the risk level; and sending the warning report to the user terminal.

[0012] According to one aspect of this application, a cardiovascular risk early warning system based on multi-channel pulse imaging and AI is provided, comprising:

[0013] The acquisition module is used to collect pulse data from multiple arterial regions of the user through intelligent sensors;

[0014] The standardization module is used to standardize the pulse data and generate a pulse time series matrix.

[0015] The extraction module is used to construct a convolutional neural network based on spatiotemporal attention based on the spatiotemporal information of the pulse time series matrix, and extract spatiotemporal pulse features from the pulse time series matrix through the convolutional neural network;

[0016] The graph module is used to construct a health graph based on the spatiotemporal pulse characteristics and the user's health record simulation graph nodes and edges;

[0017] The early warning module is used to generate a cardiovascular risk score for the user based on the feature information in the health map, and to issue a risk warning based on the risk score.

[0018] According to one aspect of this application, a computer-readable medium is provided having a computer program stored thereon, which, when executed by a processor, implements the cardiovascular risk warning method based on multi-channel pulse imaging and AI as described in the above embodiments.

[0019] According to one aspect of this application, an electronic device is provided, comprising: one or more processors; and a storage device for storing one or more programs, which, when executed by the one or more processors, cause the one or more processors to implement the cardiovascular risk warning method based on multi-channel pulse imaging and AI as described in the above embodiments.

[0020] According to one aspect of this application, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the cardiovascular risk warning method based on multi-channel pulse imaging and AI provided in the various alternative implementations described above.

[0021] The main differences and technical effects of the technical solution of this application compared with the prior art are as follows:

[0022] On the one hand, traditional cardiovascular risk early warning methods typically rely on single or a few physiological indicators, such as blood pressure and blood lipids, which cannot comprehensively reflect the complex state of the cardiovascular system. Simultaneously, the use of pulse data is often limited to a single location, ignoring the correlation information between pulses in different arterial regions. However, the cardiovascular system is a complex whole, and its health is influenced by multiple factors. The technical solution in this application uses multi-channel pulse data to reflect various aspects of information such as cardiac pumping function and vascular elasticity from different arterial regions. Combined with information such as age and medical history in the user's health record, it can provide a more comprehensive data foundation, making accurate assessment of cardiovascular risk possible. This solution collects pulse data from multiple arterial regions using intelligent sensors and integrates user health record information to construct a health atlas. It fully utilizes the correlation information between multi-source data, and compared to traditional methods, it can more comprehensively capture the dynamic changes of the cardiovascular system, providing richer data support for risk assessment, thereby improving the accuracy of early warnings.

[0023] On the other hand, existing technologies, when processing pulse data, often neglect the spatial (between different channels) and temporal (within the cycle) correlations of pulse patterns, failing to effectively uncover complex patterns within the data. This leads to the loss of feature information or inaccurate characterization of cardiovascular status. Pulse features contain crucial cardiovascular health information in both spatial and temporal dimensions, and effectively extracting these spatiotemporal features is essential for accurate cardiovascular risk assessment. This solution utilizes spatiotemporal attention-based convolutional neural networks and gated recurrent units to better capture complex patterns and dynamic changes in pulse data. Specifically, it constructs a spatiotemporal attention-based convolutional neural network to extract local features and perform spatial fusion, enhancing key spatial information. Then, it extracts temporal features through gated recurrent units and residual connections, fusing multi-scale temporal information. Compared to traditional methods, this feature extraction method combining spatiotemporal information more effectively uncovers useful features in pulse data, improving feature expressiveness and discriminative power, thereby enhancing the accuracy of cardiovascular risk assessment.

[0024] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0025] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0026] Figure 1 The flowchart of a cardiovascular risk warning method based on multi-channel pulse imaging and AI is illustrated in one embodiment of this application.

[0027] Figure 2 The flowchart illustrating the generation of a pulse timing matrix is ​​shown in one embodiment of this application.

[0028] Figure 3 The illustration shows a schematic diagram of a cardiovascular risk early warning system based on multi-channel pulse imaging and AI in one embodiment of this application.

[0029] Figure 4 A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application is shown. Detailed Implementation

[0030] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this application more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art.

[0031] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this application. However, those skilled in the art will recognize that the technical solutions of this application can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this application.

[0032] It should be noted that the data acquisition or information collection in this embodiment is performed after authorization by the user or the object of collection, and its process and purpose strictly follow the relevant regulations.

[0033] The block diagrams shown in the attached figures are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, or in one or more hardware modules composed of smart chips, smart integrated circuits, or application-specific integrated circuits (ASICs), or in different network and / or processor devices and / or microcontroller devices.

[0034] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily need to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0035] The implementation details of the technical solution of this application are described below:

[0036] Figure 1 A flowchart of a cardiovascular risk warning method based on multi-channel pulse imaging and AI according to an embodiment of this application is shown. (Refer to...) Figure 1 As shown, this cardiovascular risk early warning method based on multi-channel pulse imaging and AI includes at least steps S110 to S150, which are detailed below:

[0037] The S110 uses intelligent sensors to collect pulse data from multiple arterial regions of the user.

[0038] In this embodiment, smart sensors equipped in specific locations are first activated. These sensors are placed on the user's body at locations that can accurately collect pulse signals from different arterial regions, such as the radial artery, carotid artery, and dorsalis pedis artery.

[0039] Subsequently, the intelligent sensor begins operation, continuously sensing pulse fluctuations in the corresponding arterial region through its sensor chip at a preset sampling frequency and precision, converting the physical changes in the pulse into electrical signals. Throughout the acquisition process, the sensor's operating status is monitored in real time to ensure the stability and accuracy of signal acquisition, and finally, the acquired multi-channel pulse data is transmitted to the subsequent processing module.

[0040] In this embodiment, multiple arterial regions include, but are not limited to, multi-channel pulse regions such as the radial artery, carotid artery, and dorsalis pedis artery. The intelligent sensors include, but are not limited to, pulse sensors with three data acquisition channels for the radial artery, carotid artery, and dorsalis pedis artery.

[0041] In the process of collecting pulse data, smart sensors are first deployed, and a communication connection is established between the data storage device and the pulse sensors. After a successful connection is established, data is read in real time from each smart sensor at a fixed sampling rate, such as one thousand times per second. Each reading accurately reflects the pulse information of the corresponding artery. By continuously performing data reading operations, a continuous stream of pulse data is ensured.

[0042] The above process, through intelligent sensors collecting pulse data from multiple arterial regions of the user, can obtain more comprehensive and richer pulse information. The pulse patterns of different arterial regions reflect different functional states of the cardiovascular system. Multi-channel acquisition can integrate multi-dimensional information, laying a solid foundation for subsequent accurate risk assessment and avoiding the one-sidedness that may be caused by pulse information from a single location.

[0043] S120, The pulse data is standardized to generate a pulse time series matrix.

[0044] In this embodiment, key data points that reflect the overall trend of change are first selected from the collected pulse data. An interpolation function is then constructed to generate a pulse baseline, which is used to eliminate baseline drift in the data. The pulse baseline is decomposed into different frequency ranges, and appropriate frequency thresholds are determined based on the statistical characteristics of the signals at each frequency level to remove noise. The denoised signal is then converted back to the time domain. Next, using the carotid pulse wave origin as a reference, the time offset of pulse signals from other arterial regions is calculated and aligned to synchronize the multi-channel pulse data in time. Finally, the processed signals are stored in an ordered manner in a predetermined format in the allocated memory space, thereby generating a pulse time series matrix.

[0045] like Figure 2As shown, in one embodiment of this application, the pulse data is standardized to generate a pulse time series matrix, including:

[0046] S210, interpolate the pulse data by constructing an interpolation function to generate a pulse baseline;

[0047] S220, the pulse baseline is decomposed into different frequency levels to generate frequency domain signals; statistical parameters are determined based on the frequency domain signals at each frequency level; the frequency domain signals are processed by comparing the statistical parameters with frequency thresholds; and the processed frequency domain signals are reconstructed back into time domain signals to obtain denoised pulse data.

[0048] S230, using the starting point of the carotid pulse wave in the de-corrupted pulse data as the reference point, calculate the cross-correlation function between the radial artery and dorsalis pedis artery pulse signals and the carotid pulse wave respectively. Based on the time offset determined by the cross-correlation function, translate and adjust the signal in the de-corrupted pulse data to generate a three-channel aligned signal.

[0049] S240, the alignment signals are sequentially stored into the previously allocated memory space according to the pre-planned format of the standardized pulse timing matrix to generate the pulse timing matrix.

[0050] Optionally, while starting to collect pulse data, a space of a specific size can be pre-allocated in memory to store the standardized pulse time series matrix after subsequent processing.

[0051] After pulse data is acquired, an interpolation function is constructed. This function is used to interpolate the pulse data, thereby eliminating baseline fluctuations caused by factors such as sensor displacement. This process is similar to finding a smooth baseline in the data and adjusting the pulse data to this baseline to generate a pulse baseline, allowing subsequent analysis to be based on a stable data foundation.

[0052] Next, noise filtering is performed. The corrected pulse baseline is decomposed into different frequency levels to generate frequency domain signals. There can be up to five frequency levels, each representing the signal characteristics at different scales. Based on the statistical parameters of the frequency domain signal at each frequency level, such as the mean, an appropriate frequency threshold is adaptively determined. Statistical parameters below this threshold are considered to be mainly caused by noise, and their frequency domain signals are set to zero; while statistical parameters above the threshold are retained or appropriately adjusted. Finally, the processed frequency domain signal is reconstructed back into the time domain signal through an inverse transform to obtain the denoised pulse data. This step effectively removes high-frequency noise interference from the pulse data, improving signal quality.

[0053] After noise filtering, time alignment is performed. Specifically, using the starting point of the carotid pulse wave in the denoising pulse data as the reference point, the cross-correlation function between the radial and dorsalis pedis pulse signals and the carotid pulse signal is calculated. The cross-correlation function reflects the similarity between the two signals at different time offsets. By finding the maximum position of the cross-correlation function, the time offset of the radial and dorsalis pedis signals relative to the carotid signal is determined. Based on these offsets, the corresponding signals are shifted and adjusted to generate aligned signals, ensuring that the three-channel pulse signals are aligned in time, thus ensuring the accuracy of subsequent multi-dimensional information fusion analysis.

[0054] After the above preprocessing operations, the processed three-channel aligned signals are sequentially stored in the pre-allocated memory space according to the pre-planned standardized pulse time series matrix format. Optionally, the pulse time series matrix is ​​planned with three channels, 1024 sampling points, and a data volume of 60 seconds. This pre-allocation of memory is to improve the efficiency of data storage and processing, and avoid performance loss caused by dynamic memory allocation during data processing. The data of each channel is arranged sequentially according to the order of sampling points, ultimately forming a complete, standardized pulse time series matrix, providing high-quality data input for subsequent risk identification and analysis.

[0055] The above process generates a pulse baseline by constructing an interpolation function, then decomposes the baseline into different frequency levels to generate frequency domain signals. After denoising, the frequency threshold is adaptively determined based on statistical parameters, and the signal is reconstructed back into the time domain. This effectively eliminates noise interference in the pulse data, making the data purer and more accurate. Simultaneously, constructing a pulse baseline helps eliminate baseline fluctuations caused by factors such as sensor displacement, improving data quality and stability. Using the carotid pulse wave origin as a reference, the time offset is determined by calculating the cross-correlation function between the radial and dorsalis pedis pulse signals and the carotid pulse wave. The signals are then shifted and adjusted to generate a three-channel aligned signal. Signal alignment ensures the temporal consistency of multi-channel pulse data, allowing subsequent analysis to be based on synchronized data, improving the accuracy and reliability of data analysis. The aligned signal is stored in a planned format to generate a pulse time series matrix, providing a standardized data structure for subsequent feature extraction, facilitating efficient computer processing and analysis.

[0056] S130, Based on the spatiotemporal information of the pulse time series matrix, a convolutional neural network based on spatiotemporal attention is constructed, and spatiotemporal pulse features are extracted from the pulse time series matrix through the convolutional neural network.

[0057] In this embodiment, specific feature channels are first selected from the pulse time-series matrix to extract local pulse feature information. Next, trainable parameters are used to analyze the spatial correlation between different feature channels. By fusing these local features with spatial correlation information, a fused spatial feature is formed. Subsequently, these fused spatial features are arranged chronologically into a sequence suitable for analyzing time changes, and a gated recurrent unit is used to extract the hidden state information at each time point sequentially. Finally, residual connections are used to integrate the hidden state information from different time points, thereby extracting rich pulse features covering both time and space dimensions.

[0058] In one embodiment of this application, a convolutional neural network based on spatiotemporal attention is constructed based on the spatiotemporal information of the pulse time series matrix. The convolutional neural network extracts spatiotemporal pulse features from the pulse time series matrix, including:

[0059] Local features of each feature channel are extracted from the pulse time series matrix;

[0060] Based on the trainable matrix parameters and the local features, the spatial factor between the feature channels is calculated, and the local features are fused based on the spatial factor to generate fused spatial features;

[0061] The fused spatial features are organized into a feature sequence suitable for time series analysis, and the hidden state of the current time step is extracted from the feature sequence through a gated loop unit.

[0062] By using residual connections, the hidden states at different time steps are fused to obtain spatiotemporal pulse features at different time scales.

[0063] First, necessary memory resources are allocated for the entire spatiotemporal feature extraction process, and relevant trainable parameter matrices are initialized. These parameter matrices include trainable parameters involved in subsequent spatial and temporal attention modules. The initial values ​​are set based on a certain random distribution or empirical values, providing a starting point for subsequent model training and optimization. In the spatial attention module, the features of different channels in the standardized pulse time-series matrix are analyzed pairwise. The feature information of any two channels is interactively processed, comparing and combining pulse information from different perspectives. In this way, the inherent connections and mutual influences between pulses in different channels are discovered.

[0064] Initial values ​​are allocated in memory for the trainable parameter matrices, which are continuously adjusted by optimization algorithms during subsequent training to adapt to different pulse data characteristics. For any two feature channels i and j in the parameter matrix, local features extracted from the standardized pulse time series matrix are used as their respective feature vectors. .

[0065] Based on the results of feature interactions, the spatial factor of each feature channel relative to other feature channels is dynamically determined, and the spatial factor of feature channel i relative to feature channel j is calculated. for:

[0066]

[0067] in, Let i, j, and k represent the trainable parameter matrices, respectively, and let i, j, and k represent the identifiers of the feature channels. Let represent the feature vectors of feature channel i and feature channel j, respectively. Through the above calculation process, among multiple information sources, different levels of attention are assigned to each information source based on the degree of correlation between them and the needs of the current task. For example, if the features of a certain channel are closely related to the features of other channels and are more helpful in distinguishing different risk states, then a higher weight is assigned to it.

[0068] In addition, spatial factors can be determined by calculating the correlation coefficients between feature channels.

[0069] After calculating the spatial factors, the features of different channels are weighted and fused according to these spatial factors. This strengthens the information of key channels, so that the fused features can more prominently reflect pulse characteristics that are important for cardiovascular risk warning, while suppressing interference from relatively unimportant information.

[0070] Optionally, before processing by the temporal attention module, the fused spatial features processed by the spatial attention module are first organized into a feature sequence suitable for time series analysis. By organizing the fused spatial features based on data order and dividing them into time steps, it is ensured that subsequent temporal attention calculations can be performed in the correct temporal order.

[0071] The feature sequence is processed step-by-step using a Gated Recurrent Unit (GRU). At each time step, the GRU determines how to update the current state based on the current input features and the state information retained from the previous time step, through an internal gating mechanism (update gate and reset gate). The update gate controls the proportion of state information retained from the previous time step, while the reset gate determines whether to ignore some state information from the previous time step. In this way, the GRU can effectively capture the dynamic changes and long-term dependencies of pulse features over time.

[0072] To fully utilize pulse information across different time scales, residual connections are used to fuse the outputs of the GRU at different time steps. This allows the model to learn not only feature changes within a local timeframe but also feature trends over a longer time span. This multi-scale feature fusion helps to more comprehensively characterize the dynamic changes in the pulse. The temporal attention module effectively captures the dynamic changes of the pulse signal over time using the GRU, while residual connections further fuse multi-scale features, enabling the model to more comprehensively understand the temporal characteristics of the pulse and improving its ability to represent dynamic changes in the pulse.

[0073] After processing by the spatial attention module and the temporal attention module, the feature information enhanced in the spatial dimension and modeled in the temporal dimension is aggregated to form a fixed-dimensional (256-dimensional) spatiotemporal pulse feature. This integrates the complex changes in the pulse in space (between channels) and time (within the cycle), such as key features like pulse wave propagation speed and reflected wave intensity, providing highly condensed and discriminative feature inputs for subsequent cardiovascular risk warning, greatly improving the performance and accuracy of the warning model.

[0074] The above process extracts local features from preset feature channels from the pulse time series matrix, calculates spatial factors between feature channels based on trainable matrix parameters, and fuses local features to generate fused spatial features. This process can uncover spatial correlation information between different feature channels, strengthen key features, and make the extracted features better reflect the spatial characteristics of the pulse, improving the expressiveness and discriminative power of the features. The fused spatial features are organized into a feature sequence suitable for time series analysis. The hidden state of the current time step is extracted through a gated recurrent unit, and then the hidden states of different time steps are fused using a residual connection method to obtain spatiotemporal pulse features at different time scales. The gated recurrent unit can effectively capture the dynamic changes and long-term dependencies of pulse features in the time dimension, while the residual connection helps to fuse multi-scale features, making the extracted spatiotemporal pulse features more comprehensive and rich, and better able to represent the dynamic change patterns of the pulse.

[0075] S140, construct a health graph based on the spatiotemporal pulse characteristics and the nodes and edges of the user's health record simulation graph.

[0076] In this embodiment, the user's health record is read, and preset key health indicators are extracted. Then, the previously extracted spatiotemporal pulse characteristics and these health indicators are simulated as graph nodes in a health atlas. Simultaneously, the correlation degree between each graph node, i.e., the relationship factor, is calculated, and these relationship factors are used as edges connecting the graph nodes. In this way, health information from different sources is integrated in the form of graph nodes, and edges are used to represent the relationships between them, ultimately successfully constructing a health atlas.

[0077] In one embodiment of this application, a health graph is constructed based on the spatiotemporal pulse characteristics and the user's health record simulation graph nodes and edges, including:

[0078] Read the user's health record and extract preset health indicators from the health record;

[0079] The spatiotemporal pulse characteristics and the health indicators are simulated as graph nodes, and the relationship factors between the graph nodes are calculated.

[0080] The relationship factors between the graph nodes are used as edges in the health graph to generate a health graph.

[0081] The spatiotemporal pulse characteristics calculated based on the aforementioned steps contain complex features of the pulse in both spatial and temporal dimensions. Simultaneously, health records are retrieved, encompassing information such as age, gender, blood pressure, blood lipids, blood glucose, and medication history. These input data undergo preliminary checks and organization to ensure completeness and correct formatting, preparing for subsequent atlas construction.

[0082] Spatiotemporal pulse characteristics and every indicator in the health record, such as systolic blood pressure and low-density lipoprotein cholesterol (LDL-C), are simulated as graph nodes. This unifies and abstracts data from different sources and with different properties into basic units within a graph structure, enabling subsequent correlation analysis of these data in graph form. Each graph node carries specific data information, laying the foundation for analyzing the relationships between them.

[0083] Relationship factors are used to calculate the relationship factors between graph nodes, thereby constructing the edge weights of a sparsely connected graph. In practice, all graph node pairs are traversed, and the degree of data association between each pair is analyzed. Specifically, first, sets X and Y of all graph nodes are determined (where X and Y are different representations of the sets of all graph nodes used for formula calculation). For each pair of graph nodes x and y in the set, the joint probability distribution p(x,y) and the marginal probability distributions p(x) and p(y) are calculated. In actual calculations, these probability distributions can be approximated by statistically analyzing the frequency of node values. Then, the relationship factors between each graph node are calculated. for:

[0084]

[0085] in, x, y These represent two graph nodes, molecule. This represents the probability that two graph nodes simultaneously take a specific value, while the denominator p(x)p(y) represents the product of the probabilities that the two graph nodes independently take the corresponding values. The calculation yields a numerical value that measures the correlation between two nodes; the larger the correlation factor, the stronger the correlation between the two nodes.

[0086] In this embodiment, the relation factor measures the degree of interdependence between two variables. Through extensive data statistics and analysis, the magnitude of the relation factor between each pair of graph nodes is determined. The larger the relation factor value, the stronger the association between the two graph nodes, and the higher the corresponding edge weight in the constructed sparse connected graph. In this way, a healthy graph that reflects the relationships between graph nodes is established, providing connection information for subsequent graph convolution propagation.

[0087] The above process integrates various aspects of a user's health information by reading their health records and extracting preset health indicators, providing data support for constructing a comprehensive health graph. Through graph node simulation and relational factor calculation, spatiotemporal pulse characteristics and health indicators are simulated as graph nodes, and the relational factors between these nodes are calculated as edges of the health graph. The health graph constructed in this way can intuitively represent the correlations between different health information, providing a structured data model for subsequent risk assessment and facilitating in-depth analysis of the impact of various factors on cardiovascular risk.

[0088] S150, Based on the feature information in the health map, generate a cardiovascular risk score for the user, and issue a risk warning based on the risk score.

[0089] In this embodiment, the process begins by focusing on each graph node in the health graph, aggregating the feature information of each node and its neighboring nodes to form a more representative first feature. Next, the first features of the graph node and its neighboring nodes are further aggregated to obtain a second feature that better reflects the overall condition. This second feature is then mapped to a specific range to generate a cardiovascular risk score for the user. Finally, based on pre-defined risk level classification standards, the risk level corresponding to the risk score is determined, and a risk warning is issued to the user or relevant personnel according to the warning method corresponding to this risk level.

[0090] In one embodiment of this application, a cardiovascular risk score for a user is generated based on feature information in the health atlas, including:

[0091] The feature information of graph nodes and their neighboring nodes in the health graph is propagated through a graph convolutional neural network to aggregate the feature information of each graph node and its neighboring nodes, thereby generating the first feature of the graph node.

[0092] The first features of the graph nodes and their neighboring nodes in the health graph are aggregated by global pooling to generate the second features of the graph nodes.

[0093] The second feature is mapped to a preset range to generate a cardiovascular risk score for the user.

[0094] In one embodiment of this application, initial values ​​are assigned to the trainable parameter matrix in the graph convolutional network (GCN). These parameters will be continuously adjusted and optimized during subsequent training to adapt to different data characteristics and risk assessment needs.

[0095] Based on the constructed health graph, graph convolution operations are performed. For each graph node in the health graph, information about all its neighboring nodes is collected. Here, neighboring nodes refer to other nodes connected to the current node via edges. The information of neighboring nodes is aggregated based on their features and their connection weights to the current node. During the computation, an activation function (such as ReLU) is introduced to increase the model's non-linear expressive power, enabling the model to learn more complex patterns. After the first layer of graph convolution, the feature representation of each node is updated, incorporating the information from neighboring nodes, and serves as the first feature of the graph node.

[0096] The output of the first layer of graph convolution is used as the input of the second layer, and graph convolution is performed again to aggregate neighbor node information. Trainable parameters and activation functions are then applied to calculate and generate the second feature of the graph node. Through two layers of graph convolution, the node can acquire information from a wider range, further enriching the node's feature representation.

[0097] After two layers of graph convolution, a linear transformation is performed on the final second feature of the graph nodes, mapping the result to a specific preset range (0~100) to generate the user's cardiovascular risk score. S for:

[0098]

[0099] in, Let represent the trainable parameter matrices, Represents graph nodes n, m Relationship factors Represents graph nodes n, m degree of nodes, Represents graph nodes n, m The feature vector. This score integrates information from multimodal data and reflects the user's current cardiovascular risk level.

[0100] This embodiment effectively aggregates information from neighboring nodes through a graph convolutional network, ensuring that each node's feature representation not only includes its own information but also incorporates information from its associated nodes. Through two layers of graph convolution, the model can capture a wider range of data relationships and complex patterns, thereby more accurately assessing cardiovascular risk. The final output risk score provides doctors and users with an intuitive risk indicator, helping them assess users' cardiovascular health and facilitating timely intervention and prevention strategies.

[0101] The above process involves two aggregations of feature information from graph nodes and their neighboring nodes in the health graph, generating the first and second features of the graph nodes. Through multiple aggregation operations, the information from graph nodes and their neighboring nodes can be fully integrated, uncovering deeper feature relationships and making the generated features more reflective of the user's overall health status and cardiovascular risk. The second feature is then mapped to a preset range to generate the user's cardiovascular risk score. The risk score presents the user's cardiovascular risk level in an intuitive numerical form, improving the accuracy and reliability of the risk score and providing strong support for the early detection and intervention of cardiovascular diseases.

[0102] In one embodiment of this application, risk warning based on the risk score includes:

[0103] The risk scores are classified into risk levels based on preset thresholds;

[0104] Risk warnings are issued using the warning methods corresponding to the aforementioned risk levels.

[0105] In one embodiment of this application, threshold ranges corresponding to different risk levels are pre-set during system initialization or based on clinical experience and historical data analysis. These thresholds are key criteria for classifying risk levels and are typically determined by medical experts based on the risk characteristics of cardiovascular diseases and clinical practice. For example, three levels—low risk, intermediate risk, and high risk—are set, each corresponding to a specific risk score range. The purpose of pre-setting thresholds is to provide clear boundaries for subsequent risk level classification, enabling the user's risk level to be quickly and accurately determined based on the risk score.

[0106] After calculating the risk score, a risk level classification process is performed. The current risk score is compared one by one with preset thresholds. First, it checks if the risk score meets the threshold conditions for a high-risk level. If it does, the user is classified as high-risk. If not, it checks if it meets the threshold conditions for a medium-risk level. If it does, the user is classified as medium-risk. If neither of these conditions is met, the user is classified as low-risk. This process is similar to a multi-branch decision logic, mapping continuous risk scores to discrete risk levels through simple comparison operations, making the risk status more intuitive and understandable.

[0107] After the risk level is determined, a corresponding early warning method is matched for each risk level according to the pre-set rules. Different risk levels correspond to different levels of urgency and methods of early warning.

[0108] Among them, the early warning method corresponding to low risk is a relatively mild health education, which can extract relevant suggestions on diet, exercise, and lifestyle habits from a pre-set health knowledge base and push them to users through user applications, text messages, etc., to remind users to pay attention to their own health and take preventive measures.

[0109] The warning method for medium-risk individuals has been upgraded to include remote monitoring. This allows communication with remote monitoring devices, such as smart blood pressure monitors and heart rate monitoring bracelets, with the ability to set monitoring frequencies, such as daily blood pressure measurements, and to receive monitoring data in real time. Simultaneously, monitoring reports are generated for medical personnel to view and analyze, enabling timely detection of changes in the user's health status.

[0110] Among these, high-risk situations represent the most urgent circumstances, and the warning method involves seeking immediate medical attention and receiving notifications of nearby hospitals. Emergency medical attention alerts are sent to users and their families through various channels, such as SMS, app notifications, and voice calls, providing the addresses and contact information of nearby hospitals capable of treating cardiovascular diseases. Simultaneously, relevant emergency medical services can also be notified to prepare for patient reception.

[0111] Specific alert actions are executed according to the pre-defined alert method. For health education, an alert report is generated from the compiled health advice, user information, and risk level, and then sent to the user's terminal. For remote monitoring, a stable communication connection is established with the monitoring equipment to ensure real-time data transmission and storage. For emergency medical alerts, the information is presented in a prominent manner to ensure that users and their families receive it promptly and take action. Throughout the entire alert execution process, information such as the alert time, method, and content is recorded for subsequent effectiveness evaluation and system optimization.

[0112] Through the above steps, risk scores are categorized into risk levels based on preset thresholds, discretizing continuous risk scores into different levels. This makes the risk status clearer and easier to understand, facilitating appropriate early warning and intervention measures based on different levels. Risk warnings are issued through risk level-based alerts, providing personalized warning services for different risk levels, such as health education for low-risk cases, remote monitoring for medium-risk cases, and emergency medical alerts for high-risk cases. This timely and effective communication of risk information to users and medical personnel encourages appropriate action, reducing the risk and severity of cardiovascular diseases and contributing to improved health management and treatment outcomes for users.

[0113] This application's technical solution collects pulse data from multiple arterial regions of a user using intelligent sensors; standardizes the pulse data to generate a pulse temporal matrix; constructs a spatiotemporal attention-based convolutional neural network based on the spatiotemporal information of the pulse temporal matrix, and extracts spatiotemporal pulse features from the pulse temporal matrix through the convolutional neural network; constructs a health atlas based on the spatiotemporal pulse features and the user's health record simulation graph nodes and edges; generates a cardiovascular risk score for the user based on the feature information in the health atlas, and provides risk warnings based on the risk score. This application's technical solution, by collecting pulse data from multiple arterial regions to construct a health atlas, can comprehensively integrate spatiotemporal pulse features and user health information. The risk score generated based on this comprehensively considers multiple aspects of the cardiovascular system's state and can accurately reflect the degree of risk. Risk warnings based on this score can provide appropriate prompts for different risk levels, helping to prevent cardiovascular problems in a timely manner, providing a basis for early intervention, and improving the effectiveness of cardiovascular disease prevention and control.

[0114] The following describes embodiments of the cardiovascular risk early warning system based on multi-channel pulse imaging and AI, which can be used to execute the cardiovascular risk early warning method based on multi-channel pulse imaging and AI described in the above embodiments of this application. It is understood that the cardiovascular risk early warning system based on multi-channel pulse imaging and AI can be a computer program (including program code) running on a computer device. For example, the cardiovascular risk early warning system based on multi-channel pulse imaging and AI can install application software or health management software to achieve cloud computing of the big data generated during the detection process through an industrial cloud platform. The cardiovascular risk early warning system based on multi-channel pulse imaging and AI can be used to execute the corresponding steps in the method provided in the embodiments of this application. For details not disclosed in the embodiments of the cardiovascular risk early warning system based on multi-channel pulse imaging and AI of this application, please refer to the embodiments of the cardiovascular risk early warning method based on multi-channel pulse imaging and AI described above.

[0115] Figure 3 A block diagram of a cardiovascular risk early warning system based on multi-channel pulse imaging and AI according to an embodiment of this application is shown.

[0116] Reference Figure 3 As shown, a cardiovascular risk early warning system based on multi-channel pulse imaging and AI according to an embodiment of this application includes:

[0117] The acquisition module 310 is used to collect pulse data of multiple arterial regions of the user through a smart sensor;

[0118] Standardization module 320 is used to standardize the pulse data and generate a pulse time series matrix.

[0119] The extraction module 330 is used to construct a convolutional neural network based on spatiotemporal attention based on the spatiotemporal information of the pulse time series matrix, and extract spatiotemporal pulse features from the pulse time series matrix through the convolutional neural network;

[0120] The graph module 340 is used to construct a health graph based on the spatiotemporal pulse characteristics and the nodes and edges of the user's health record simulation graph;

[0121] The early warning module 350 is used to generate a cardiovascular risk score for the user based on the feature information in the health map, and to provide risk warnings based on the risk score.

[0122] In this application, based on the aforementioned scheme, the standardization processing of the pulse data to generate a pulse time series matrix includes: interpolating the pulse data by constructing an interpolation function to generate a pulse baseline; decomposing the pulse baseline into different frequency levels to generate frequency domain signals; determining statistical parameters based on the frequency domain signals at each frequency level, processing the frequency domain signals by comparing the statistical parameters with frequency thresholds, and reconstructing the processed frequency domain signals back into time domain signals to obtain denoised pulse data; using the starting point of the carotid pulse wave in the denoised pulse data as a reference point, calculating the cross-correlation function between the radial artery and dorsalis pedis artery pulse signals and the carotid pulse wave respectively, and adjusting the signals in the denoised pulse data by shifting based on the time offset determined by the cross-correlation function to generate a three-channel aligned signal; and storing the aligned signal sequentially into the previously allocated memory space according to the pre-planned format of the standardized pulse time series matrix to generate a pulse time series matrix.

[0123] In this application, based on the aforementioned scheme, the step of constructing a convolutional neural network based on spatiotemporal attention using the spatiotemporal information of the pulse time series matrix, and extracting spatiotemporal pulse features from the pulse time series matrix using the convolutional neural network, includes: extracting local features of each feature channel from the pulse time series matrix; calculating spatial factors between the feature channels based on trainable matrix parameters and the local features, and fusing the local features based on the spatial factors to generate fused spatial features; organizing the fused spatial features into a feature sequence suitable for time series analysis, and extracting the hidden state of the current time step from the feature sequence using a gated recurrent unit; and fusing the hidden states of different time steps using residual connections to obtain spatiotemporal pulse features at different time scales.

[0124] In this application, based on the aforementioned scheme, the step of constructing a health graph by simulating graph nodes and edges based on the spatiotemporal pulse characteristics and the user's health record includes: reading the user's health record and extracting preset health indicators from the health record; simulating the spatiotemporal pulse characteristics and the health indicators as graph nodes and calculating the relationship factors between each graph node; and using the relationship factors between the graph nodes as edges of the health graph to generate the health graph.

[0125] In this application, based on the aforementioned scheme, generating a user's cardiovascular risk score based on the feature information in the health graph includes: performing feature propagation on the feature information of graph nodes and their neighboring nodes in the health graph through a graph convolutional neural network, aggregating the feature information of each graph node and its neighboring nodes to generate a first feature of the graph node; aggregating the first features of the graph nodes and their neighboring nodes in the health graph through global pooling to generate a second feature of the graph node; and mapping the second feature to a preset range to generate a user's cardiovascular risk score.

[0126] In this application, based on the aforementioned scheme, the risk warning based on the risk score includes: classifying the risk level corresponding to the risk score based on a preset threshold; and issuing a risk warning through the warning method corresponding to the risk level.

[0127] In this application, based on the aforementioned scheme, after generating a cardiovascular risk score for the user based on the feature information in the health map and issuing a risk warning based on the risk score, the method further includes: generating a warning report based on user information and the risk level; and sending the warning report to the user terminal.

[0128] This application's technical solution collects pulse data from multiple arterial regions of a user using intelligent sensors; standardizes the pulse data to generate a pulse temporal matrix; constructs a spatiotemporal attention-based convolutional neural network based on the spatiotemporal information of the pulse temporal matrix, and extracts spatiotemporal pulse features from the pulse temporal matrix through the convolutional neural network; constructs a health atlas based on the spatiotemporal pulse features and the user's health record simulation graph nodes and edges; generates a cardiovascular risk score for the user based on the feature information in the health atlas, and provides risk warnings based on the risk score. This application's technical solution, by collecting pulse data from multiple arterial regions to construct a health atlas, can comprehensively integrate spatiotemporal pulse features and user health information. The risk score generated based on this comprehensively considers multiple aspects of the cardiovascular system's state and can accurately reflect the degree of risk. Risk warnings based on this score can provide appropriate prompts for different risk levels, helping to prevent cardiovascular problems in a timely manner, providing a basis for early intervention, and improving the effectiveness of cardiovascular disease prevention and control.

[0129] Figure 4A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application is shown.

[0130] It should be noted that the computer system of the electronic device in this embodiment is only an example and should not impose any limitations on the function and scope of use of the embodiments of this application.

[0131] In this embodiment, the computer system includes a central processing unit 401, which can perform various appropriate actions and processes based on programs stored in read-only memory 402 or programs loaded from storage section 408 into random access memory 403, such as executing the cardiovascular risk warning method based on multi-channel pulse and AI described in the above embodiment. The random access memory 403 also stores various programs and data required for system operation, thereby realizing big data storage and big data management. The central processing unit 401, read-only memory 402, and random access memory 403 are interconnected via bus 404. Input / output interface 405 is also connected to bus 404.

[0132] The following components are connected to the input / output interface 405: an input section 406 including a keyboard, mouse, etc.; an output section 407 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 408 including a hard disk, etc.; and a communication section 409 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 409 performs communication processing via a network such as the Internet. A drive 410 is also connected to the input / output interface 405 as needed. A removable medium 411, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 410 as needed so that computer programs read from it can be installed into the storage section 408 as needed.

[0133] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including a computer program for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 409, and / or installed from removable medium 411. When the computer program is executed by central processing unit 401, it performs various functions defined in the system of this application.

[0134] It should be noted that the computer-readable medium shown in the embodiments of this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying a computer-readable computer program. The transmitted data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The computer program contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.

[0135] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0136] The units described in the embodiments of this application can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the specific unit itself.

[0137] According to one aspect of this application, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in the various alternative implementations described above.

[0138] In another aspect, this application also provides a computer-readable medium, which may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into the electronic device. The computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to implement the cardiovascular risk warning method based on multi-channel pulse imaging and AI described in the above embodiments.

[0139] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0140] Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, touch terminal, or network device, etc.) to execute the methods according to the embodiments of this application.

[0141] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the embodiments disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

[0142] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A cardiovascular risk early warning method based on multi-channel pulse imaging and AI, characterized in that, include: The system collects pulse data from multiple arterial regions of the user using intelligent sensors. The pulse data is standardized to generate a pulse time series matrix; Based on the spatiotemporal information of the pulse time series matrix, a convolutional neural network based on spatiotemporal attention is constructed, and spatiotemporal pulse features are extracted from the pulse time series matrix through the convolutional neural network. Based on the spatiotemporal pulse characteristics and the nodes and edges of the user's health record simulation graph, a health graph is constructed. Based on the feature information in the health map, a cardiovascular risk score is generated for the user, and a risk warning is given based on the risk score. The process of standardizing the pulse data to generate a pulse time-series matrix includes: The pulse data is interpolated using an interpolation function to generate a pulse baseline. The pulse baseline is decomposed into different frequency levels to generate frequency domain signals; statistical parameters are determined based on the frequency domain signals at each frequency level; the frequency domain signals are processed by comparing the statistical parameters with frequency thresholds; and the processed frequency domain signals are reconstructed back into time domain signals to obtain denoised pulse data. Using the starting point of the carotid pulse wave in the denoised pulse data as the reference point, the cross-correlation function between the radial artery and dorsalis pedis artery pulse signals and the carotid pulse wave is calculated respectively. Based on the time offset determined by the cross-correlation function, the signal in the denoised pulse data is shifted and adjusted to generate a three-channel aligned signal. The alignment signals are sequentially stored into the previously allocated memory space according to the pre-planned format of the standardized pulse timing matrix to generate the pulse timing matrix.

2. The cardiovascular risk early warning method based on multi-channel pulse imaging and AI according to claim 1, characterized in that, Based on the spatiotemporal information of the pulse time series matrix, a convolutional neural network based on spatiotemporal attention is constructed. This convolutional neural network extracts spatiotemporal pulse features from the pulse time series matrix, including: Local features of each feature channel are extracted from the pulse time series matrix; Based on the trainable matrix parameters and the local features, the spatial factor between the feature channels is calculated, and the local features are fused based on the spatial factor to generate fused spatial features; The fused spatial features are organized into a feature sequence suitable for time series analysis, and the hidden state of the current time step is extracted from the feature sequence through a gated loop unit. By using residual connections, the hidden states at different time steps are fused to obtain spatiotemporal pulse features at different time scales.

3. The cardiovascular risk early warning method based on multi-channel pulse imaging and AI according to claim 1, characterized in that, Based on the aforementioned spatiotemporal pulse characteristics and the nodes and edges of the user's health record simulation graph, a health graph is constructed, including: Read the user's health record and extract preset health indicators from the health record; The spatiotemporal pulse characteristics and the health indicators are simulated as graph nodes, and the relationship factors between the graph nodes are calculated. The relationship factors between the graph nodes are used as edges in the health graph to generate a health graph.

4. The cardiovascular risk early warning method based on multi-channel pulse imaging and AI according to claim 1, characterized in that, Based on the feature information in the health atlas, a cardiovascular risk score for the user is generated, including: The feature information of graph nodes and their neighboring nodes in the health graph is propagated through a graph convolutional neural network to aggregate the feature information of each graph node and its neighboring nodes, thereby generating the first feature of the graph node. The first features of the graph nodes and their neighboring nodes in the health graph are aggregated by global pooling to generate the second features of the graph nodes. The second feature is mapped to a preset range to generate a cardiovascular risk score for the user.

5. The cardiovascular risk early warning method based on multi-channel pulse imaging and AI according to claim 1, characterized in that, Risk warnings are issued based on the aforementioned risk scores, including: The risk scores are classified into risk levels based on preset thresholds; Risk warnings are issued using the warning methods corresponding to the aforementioned risk levels.

6. The cardiovascular risk early warning method based on multi-channel pulse imaging and AI according to any one of claims 1-5, characterized in that, Based on the feature information in the health map, a cardiovascular risk score is generated for the user. After issuing a risk warning based on the risk score, the process further includes: Generate early warning reports based on user information and risk levels; The warning report is sent to the user terminal.

7. A cardiovascular risk early warning system based on multi-channel pulse imaging and AI, characterized in that, include: The acquisition module is used to collect pulse data from multiple arterial regions of the user through intelligent sensors; The standardization module is used to standardize the pulse data and generate a pulse time series matrix. The extraction module is used to construct a convolutional neural network based on spatiotemporal attention based on the spatiotemporal information of the pulse time series matrix, and extract spatiotemporal pulse features from the pulse time series matrix through the convolutional neural network; The graph module is used to construct a health graph based on the spatiotemporal pulse characteristics and the user's health record simulation graph nodes and edges; The early warning module is used to generate a cardiovascular risk score for the user based on the feature information in the health map, and to issue a risk warning based on the risk score; The process of standardizing the pulse data to generate a pulse time-series matrix includes: The pulse data is interpolated using an interpolation function to generate a pulse baseline. The pulse baseline is decomposed into different frequency levels to generate frequency domain signals; statistical parameters are determined based on the frequency domain signals at each frequency level; the frequency domain signals are processed by comparing the statistical parameters with frequency thresholds; and the processed frequency domain signals are reconstructed back into time domain signals to obtain denoised pulse data. Using the starting point of the carotid pulse wave in the denoised pulse data as the reference point, the cross-correlation function between the radial artery and dorsalis pedis artery pulse signals and the carotid pulse wave is calculated respectively. Based on the time offset determined by the cross-correlation function, the signal in the denoised pulse data is shifted and adjusted to generate a three-channel aligned signal. The alignment signals are sequentially stored into the previously allocated memory space according to the pre-planned format of the standardized pulse timing matrix to generate the pulse timing matrix.

8. A computer-readable medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the cardiovascular risk early warning method based on multi-channel pulse and AI as described in any one of claims 1 to 6.

9. An electronic device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the cardiovascular risk warning method based on multi-channel pulse imaging and AI as described in any one of claims 1 to 6.

Images