Vascular function detection method based on multi-physiological signal fusion and related products

Abstract

This application relates to the field of smart medical technology, and provides a method and related products for vascular function detection based on multi-physiological signal fusion. The method includes: acquiring arterial pulse wave signals from a first and second measurement site of a user using a first sensor and a second sensor configured on a wearable device, respectively; acquiring motion state data of the user during the acquisition of arterial pulse wave signals using an inertial measurement unit configured on the wearable device, and dynamically compensating the arterial pulse wave signals based on the motion state data; extracting feature waveforms from the dynamically compensated arterial pulse wave signals using a signal processing model; calculating the characteristic time difference of pulse wave propagation from the first measurement site to the second measurement site based on the dynamically compensated first and second feature waveforms; and inputting the characteristic time difference, the user's static physiological parameters, and / or at least one waveform morphological parameter into a vascular stiffness assessment model to obtain a vascular stiffness assessment value.

Description

Technical Field

[0001] This application relates to the field of smart medical technology, and in particular to a method and related products for vascular function detection based on the fusion of multiple physiological signals. Background Technology

[0002] Cardiovascular disease is one of the leading causes of death worldwide, and arteriosclerosis is a significant pathological basis for it. In the early stages of arteriosclerosis, the stiffness of the blood vessel walls gradually increases. Therefore, early and convenient detection and assessment of vascular stiffness are crucial for the prevention and early intervention of cardiovascular disease.

[0003] Pulse wave velocity (PWV) is currently one of the important non-invasive reference indicators for assessing vascular stiffness in clinical practice. Its basic principle is to measure the propagation time of a pulse wave between two specific arterial measurement points and calculate the propagation velocity by combining this with the vascular path length between the two points. The higher the vascular stiffness, the faster the pulse wave propagation velocity. Currently, some specialized medical devices can achieve relatively accurate PWV measurements, such as detectors using the carotid-femoral artery measurement path. However, these devices are usually bulky, expensive, and require operation by professionals in specific clinical settings, making them difficult to integrate into users' daily health monitoring and unable to achieve long-term, frequent tracking of vascular health.

[0004] To improve the convenience of testing, some vascular function testing solutions based on wearable devices (such as smart bracelets, watches, etc.) have emerged on the market. Most of these solutions attempt to use a single photoplethysmography (PPG) sensor on the device to acquire pulse wave signals. However, such solutions suffer from inaccurate vascular assessments because a single sensor cannot directly measure time differences and has poor resistance to motion interference. Summary of the Invention

[0005] Based on this, it is necessary to address the technical problems of the existing technology, such as the inability of a single sensor to directly measure time difference and poor resistance to motion interference, which leads to inaccurate vascular assessment. Therefore, a vascular function detection method and related products based on the fusion of multiple physiological signals are proposed.

[0006] Firstly, a method for detecting vascular function based on the fusion of multiple physiological signals is provided, the method comprising:

[0007] The first and second sensors configured on the wearable device collect arterial pulse wave signals from the user's first and second measurement sites, respectively.

[0008] The wearable device uses an inertial measurement unit to collect motion state data of the user during the process of acquiring arterial pulse wave signals, and dynamically compensates the arterial pulse wave signals based on the motion state data.

[0009] Based on the dynamically compensated arterial pulse wave signal, feature waveforms are extracted through a signal processing model to identify waveform feature points strongly correlated with the cardiac cycle in the arterial pulse wave signal. The signal processing model is a neural network model based on an attention mechanism. The feature waveforms include a first feature waveform corresponding to the first measurement site and a second feature waveform corresponding to the second measurement site.

[0010] Based on the first and second characteristic waveforms after dynamic compensation, the characteristic time difference of pulse wave propagation from the first measurement site to the second measurement site is calculated.

[0011] The characteristic time difference, pulse wave propagation velocity, user's static physiological parameters, and at least one waveform morphological parameter extracted from the arterial pulse wave signal are input into the vascular stiffness assessment model to obtain a vascular stiffness assessment value. The waveform morphological parameters include at least one of the following: the slope of the rising branch of the pulse wave and the ratio of the peak value of the reflected wave to the peak value of the main wave.

[0012] Secondly, a vascular function detection device based on multi-physiological signal fusion is provided, the device comprising:

[0013] The acquisition module is used to acquire arterial pulse wave signals at the user's first and second measurement sites respectively through the first and second sensors configured on the wearable device;

[0014] The compensation module is used to collect motion state data of the user during the process of acquiring arterial pulse wave signals through the inertial measurement unit configured on the wearable device, and to dynamically compensate the arterial pulse wave signal based on the motion state data.

[0015] The extraction module is used to extract feature waveforms based on the dynamically compensated arterial pulse wave signal through a signal processing model. It is used to identify waveform feature points that are strongly correlated with the cardiac cycle in the arterial pulse wave signal. The signal processing model is a neural network model based on an attention mechanism. The feature waveforms include a first feature waveform corresponding to the first measurement site and a second feature waveform corresponding to the second measurement site.

[0016] The calculation module is used to calculate the characteristic time difference of the pulse wave from the first measurement site to the second measurement site based on the dynamically compensated first characteristic waveform and the second characteristic waveform;

[0017] The evaluation module is used to input the characteristic time difference, pulse wave conduction velocity, user's static physiological parameters, and at least one waveform morphological parameter extracted from the arterial pulse wave signal into the vascular stiffness evaluation model to obtain a vascular stiffness evaluation value. The waveform morphological parameters include at least one of the following: the slope of the rising branch of the pulse wave and the ratio of the peak value of the reflected wave to the peak value of the main wave.

[0018] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method for vascular function detection based on multi-physiological signal fusion.

[0019] Fourthly, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described method for vascular function detection based on the fusion of multiple physiological signals.

[0020] As can be seen from the technical solution provided in this application, on the one hand, by synchronously acquiring arterial pulse wave signals through at least two sensing units set at different arterial measurement points on the user's body surface, and calculating the characteristic time difference of the pulse wave between the two measurement points, the key physiological parameter reflecting vascular stiffness—pulse wave conduction time—can be directly obtained. This provides a direct and reliable data basis for subsequent PWV calculation and vascular stiffness assessment, overcoming the deficiency of a single sensor solution that cannot directly measure pulse wave conduction time, thus making the assessment results more accurate. On the other hand, by introducing an inertial measurement unit to acquire motion state data, and dynamically compensating the extracted feature waveforms based on this, the solution can actively identify and suppress motion artifacts caused by the user's daily activities (e.g., arm swinging, gesture changes, etc.). To address interference, dynamic compensation of the characteristic waveform effectively purifies the signal source used to calculate the characteristic time difference, thus ensuring the accuracy of the characteristic time difference calculation and ultimately improving the reliability and robustness of the entire system in real-life scenarios. Thirdly, the technical solution of this application does not simply calculate PWV, but rather inputs the calculated characteristic time difference, the user's static physiological parameters, and waveform morphological parameters directly extracted from the arterial pulse wave signal into the vascular stiffness assessment model for fusion analysis. The pulse wave morphological parameters themselves contain rich vascular state information. This multi-parameter fusion analysis strategy can comprehensively assess vascular function from both wave propagation velocity and waveform morphology dimensions, avoiding the limitations of a single indicator and facilitating the acquisition of a more comprehensive vascular stiffness assessment value that better reflects individual physiological characteristics. In summary, the technical solution of this application improves the accuracy and reliability of vascular assessment in everyday scenarios by using dual sensors to measure time difference and combining motion compensation with multi-parameter fusion. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0022] Figure 1 This is an application scenario diagram of a vascular function detection method based on multi-physiological signal fusion in one embodiment;

[0023] Figure 2 This is a flowchart of a vascular function detection method based on multi-physiological signal fusion in one embodiment;

[0024] Figure 3 This is a structural block diagram of a vascular function detection device based on multi-physiological signal fusion in one embodiment;

[0025] Figure 4 This is a structural block diagram of a computer device in one embodiment. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Pulse wave velocity (PWV) is currently one of the important reference indicators for non-invasive assessment of vascular stiffness recognized in clinical practice. Its basic principle is to measure the conduction time of a pulse wave between two specific arterial measurement points and calculate the conduction velocity by combining this with the vascular path length between the two points. The higher the vascular stiffness, the faster the pulse wave conduction velocity. In existing technologies, some specialized medical devices can achieve relatively accurate PWV measurements, such as detectors using the carotid-femoral artery measurement path. However, these devices are usually bulky, expensive, and require operation by professionals in specific clinical environments, making them difficult to integrate into users' daily health monitoring and unable to achieve long-term, frequent tracking of vascular health. To improve the convenience of testing, some vascular function testing solutions based on wearable devices (such as smart bracelets, watches, etc.) have also emerged on the market. Most of these solutions attempt to use a single photoplethysmography (PPG) sensor on the device to acquire pulse wave signals. However, such solutions have the following obvious limitations: 1) Since there is only one measurement point, it is impossible to directly calculate the pulse wave propagation time on the blood vessel segment, thus making it difficult to calculate the PWV value directly and accurately; 2) In daily activities, the movement of the user's limbs is very likely to interfere with the PPG signal (i.e., motion artifacts), and the single sensor solution lacks an effective signal compensation mechanism, resulting in poor signal quality and inaccurate feature point identification, ultimately leading to insufficient reliability and accuracy of the assessment results of blood vessel status.

[0028] To address the aforementioned problems in existing technologies, this application proposes a vascular function detection method based on multi-physiological signal fusion, which can be applied to... Figure 1 The example application scenarios mainly involve entities such as wearable devices 101 and smart computing devices (APP 102) or cloud servers 103. Figure 1In the example application scenario, a user initiates a vascular health screening using a wearable device 101 on their wrist. Two high-precision sensors built into the wearable device 101 then begin working: one attached to the radial artery and the other to the digital artery, simultaneously collecting pulsation signals from both arteries. At the same time, motion sensors within the wearable device 101 continuously record subtle hand movements, laying the foundation for subsequent precise analysis. The collected raw signals are securely transmitted to the user's smart computing device APP 102 (e.g., a mobile app) or directly uploaded to a cloud server 103. An attention-based intelligent algorithm begins deep processing of the signals, identifying feature points strongly correlated with the heartbeat cycle from the complex waveforms and generating two clear feature waveforms. The smart computing device APP 102 or the cloud server 103 dynamically compensates for these two waveforms based on the motion data simultaneously recorded by the wearable device 101, effectively filtering out interference from the user's recent keyboard typing or mouse movements, ensuring signal purity. Subsequently, the intelligent computing device APP 102 or cloud server 103 compares the two purified waveforms to accurately calculate the characteristic time difference of the pulse wave transmission from the wearable device 101 to the base of the hand. At this point, all key physiological parameters are ready: not only this characteristic time difference, but also static information such as the user's age and height, as well as key morphological indicators extracted from the pulse wave pattern. This information is aggregated into a multi-task learning model trained on a large amount of data on the cloud server 103. This model, through its shared feature extraction layer, fuses and analyzes this multi-source information, ultimately generating a comprehensive report: its main task branch outputs a professional vascular stiffness assessment value, objectively reflecting the physical state of the blood vessels; while its auxiliary task branch outputs a more easily understood "vascular age" assessment, intuitively telling the user the level of their vascular condition relative to their physiological age. This report is finally presented on the user's mobile phone screen, including not only the results but also historical trend comparisons, helping the user to understand subtle changes in their health.

[0029] Please see Figure 2 As shown, Figure 2 A flowchart illustrating a vascular function detection method based on multi-physiological signal fusion provided in an embodiment of the present invention is shown, mainly including steps S201 to S205, which are detailed below:

[0030] Step S201: Collect arterial pulse wave signals from the user's first and second measurement sites using the first and second sensors configured on the wearable device, respectively.

[0031] Achieving high-quality synchronous signal acquisition is fundamental to all subsequent analyses. In this embodiment, the wearable device can be a smartwatch or a wristband, and the first sensor can be configured on the back of the wearable device to contact the radial artery. The second sensor is an independent finger clip probe or a probe integrated into the wristband and extending to the digital artery. The first measurement site corresponds to the radial artery, and the second measurement site corresponds to the digital artery, thus forming a clinically significant arterial path. To accurately capture the details of the pulse wave, both sensors should operate at a sampling rate of not less than 500Hz. To ensure the accuracy of the calculation, the time synchronization accuracy between the two sensors needs to be controlled within milliseconds. The arterial pulse wave signals from the user's first and second measurement sites acquired in step S201 provide a high-quality, synchronized raw signal source for subsequent calculations of pulse wave conduction time.

[0032] Step S202: The user's motion state data during the acquisition of arterial pulse wave signals is collected through the inertial measurement unit configured on the wearable device, and the arterial pulse wave signals are dynamically compensated based on the motion state data.

[0033] Although the original signal has been acquired synchronously, in real-world scenarios, the movement of the user's limbs can easily interfere with the Photo PlethysmoGraph (PPG) signal (i.e., motion artifacts). To eliminate this interference, this application introduces motion state data as a compensation reference. As one embodiment of this application, dynamic compensation of the arterial pulse wave signal based on motion state data can be achieved by: parsing the user's limb posture change information and / or acceleration information from the motion state data; generating a compensation signal related to motion artifacts based on the user's limb posture change information and / or acceleration information; and filtering out the components corresponding to the compensation signal from the arterial pulse wave signal. Specifically, the generation of the compensation signal related to motion artifacts based on the user's limb posture change information and / or acceleration information is implemented as follows: inputting the user's limb posture change information and / or acceleration information into a pre-trained time-series generation model; simulating typical motion artifact signal patterns corresponding to the motion state data through the time-series generation model, and using the typical motion artifact signal patterns as the compensation signal.

[0034] Step S203: Based on the dynamically compensated arterial pulse wave signal, feature waveforms are extracted through a signal processing model to identify waveform feature points strongly correlated with the cardiac cycle in the arterial pulse wave signal. The signal processing model is a neural network model based on the attention mechanism, and the feature waveforms include a first feature waveform corresponding to the first measurement site and a second feature waveform corresponding to the second measurement site.

[0035] The acquired raw PPG signals often contain various noises, such as baseline drift, power line interference, and complex motion artifacts caused by limb movement. Traditional fixed-parameter filtering or thresholding methods have limited effectiveness in dealing with these non-stationary noises that overlap with the signal spectrum, easily leading to feature point identification errors and severely affecting the accuracy of subsequent pulse wave propagation time calculations. To address this issue, this application abandons the simple linear filtering approach and introduces a more intelligent, data-driven feature extraction method. As one embodiment of this application, based on the dynamically compensated arterial pulse wave signal, the feature waveform extracted through a signal processing model can be a neural network model based on an attention mechanism. By encoding the input time-series signal, its internal attention layer adaptively focuses on signal segments related to the cardiac cycle and suppresses noise interference, outputting a purified feature waveform.

[0036] Step S204: Calculate the characteristic time difference of pulse wave propagation from the first measurement site to the second measurement site based on the first characteristic waveform and the second characteristic waveform after dynamic compensation.

[0037] The first characteristic waveform, which is pure and reflects the pulsation of the radial and digital arteries, was obtained. Second characteristic waveform Then, the core task is to accurately calculate the time difference of pulse wave propagation between these two locations, i.e., the characteristic time difference (UTC). However, due to subtle variations in individual heart rate and each heartbeat, the two waveforms are not simply translated along the time axis, and directly searching for the global maximum correlation point may introduce errors. To address the potential time scaling and local deformation issues between waveforms, this invention employs a more refined calculation strategy. As a specific embodiment of this application, the characteristic time difference of pulse wave propagation from the first measurement site to the second measurement site, based on the dynamically compensated first and second characteristic waveforms, can be calculated through steps S2041 to S2043, as detailed below:

[0038] Step S2041: Perform phase alignment processing on the first feature waveform and the second feature waveform.

[0039] Phase alignment aims to eliminate coarse offsets in the overall time scale between two waveforms, aligning the corresponding cardiac cycles approximately on the time axis. One approach is to calculate the cross-correlation function of the two waveforms. And find the time delay that maximizes the cross-correlation coefficient. :

[0040]

[0041]

[0042] In the above calculation formula, the cross-correlation function Used to measure the time delay between two signals Similarity, and These are the first and second characteristic waveforms after dynamic compensation at discrete time points. t amplitude, This indicates the search for a function that makes The time delay corresponding to the maximum value The second characteristic waveform is shifted along the time axis. This allows for approximate phase alignment. More refined alignment can be achieved using the Dynamic Time Warping (DTW) algorithm, which handles the nonlinear time distortion between two time-series signals, finding the optimal matching path and thus achieving alignment at the sample point level. After this step, the corresponding heartbeat cycles in the two waveforms are effectively matched, laying the foundation for subsequently finding precise feature point time differences at smaller time scales.

[0043] Step S2042: Dynamically determine the preset feature points used to calculate the time difference.

[0044] The clarity and physiological significance of different feature points on the pulse wave (e.g., the onset point, the main peak, and the dicrotic notch) are influenced by vascular condition. The selection of feature points is dynamically determined based on waveform morphological parameters. Specifically, these parameters include the slope of the rising limb of the pulse wave. Peak value of reflected wave With the peak value of the main wave The ratio of these parameters. This application utilizes these parameters to intelligently select the most stable and clearest feature points.

[0045] Pulse wave rising limb slope ( The slope reflects the rate of arterial expansion during the initial ejection of blood from the heart. A steeper slope generally indicates a sharper waveform start point and makes it easier to identify accurately.

[0046] The ratio of the peak value of the reflected wave to the peak value of the main wave ( The ratio (ratio) is a key morphological indicator for assessing arterial stiffness and wave reflection intensity. A smaller ratio generally indicates a more prominent main wave peak and a potentially deeper dicrotic notch.

[0047] Based on the above parameters, the selection of waveform feature points can be dynamically determined based on waveform morphological parameters as follows: if the slope of the rising branch of the pulse wave is greater than a first threshold, then the waveform starting point is selected as the waveform feature point; if the ratio of the reflected wave peak value to the main wave peak value is less than a second threshold, then the main wave peak value is selected as the waveform feature point. The first and second thresholds are obtained based on statistical analysis of a large amount of data from healthy individuals, or dynamically adjusted through user-personalized calibration data. As one implementation method, the following rules 1) to 4) can be set:

[0048] 1) For a heartbeat cycle, first calculate the slope of its ascending branch. Sum and ratio .

[0049] 2) If (in, (This can be set by statistically analyzing a large amount of health data), then the starting point of the periodic waveform is considered to be... Clear and reliable waveform; point F is selected as the waveform feature point.

[0050] 3) Otherwise, if (For example, =0.6), then the main peak is considered to be Distinctive features, selection P Points are used as waveform feature points.

[0051] 4) If none of the above conditions are met, the dicrotic notch (DN) point, which is more stable in most cases, is selected as the waveform feature point.

[0052] The aforementioned dynamic selection mechanism can adapt to different vascular states and signal quality, always selecting the most robust waveform feature points for calculation, thereby improving the measurement accuracy and repeatability of the feature time difference.

[0053] Step S2043: Calculate the time difference between the aligned waveform and the selected waveform feature points.

[0054] After completing phase alignment and dynamically determining the waveform feature points for each heartbeat cycle, the characteristic time difference is calculated. It becomes direct and precise. For the aligned first... k For each heartbeat cycle, locate the waveform feature points (e.g., the starting point) selected in step S2042 on the first and second characteristic waveforms respectively. F ), and record its timestamp as and The characteristic time difference of a single period is then... To improve the robustness of the results, it is usually necessary to take multiple consecutive (e.g., 5-10) stable heartbeat cycles. The mean or median is used as the feature time difference in the final output. The above embodiments, through precise alignment and intelligent waveform feature point selection, ensure... It can reflect the true propagation time of the pulse wave in the blood vessel segment with high fidelity.

[0055] Step S205: Input the characteristic time difference of pulse wave transmission from the first measurement site to the second measurement site, pulse wave transmission velocity, user's static physiological parameters, and at least one waveform morphology parameter extracted from the arterial pulse wave signal into the vascular stiffness assessment model to obtain the vascular stiffness assessment value. The waveform morphology parameter includes at least one of the following: the slope of the rising branch of the pulse wave and the ratio of the peak value of the reflected wave to the peak value of the main wave.

[0056] Thus far, several characteristics reflecting vascular status have been obtained: the characteristic time difference that directly characterizes the basis of wave propagation velocity. The waveform morphological parameters calculated in step S2042 (e.g., , ), pulse wave velocity, and the user's static physiological parameters (e.g., age, gender, height). A single While morphological parameters can indicate vascular stiffness, they are easily confused by other factors (e.g., blood pressure, heart rate), limiting their effectiveness. To integrate multi-source information and obtain a more robust, personalized, and clinically meaningful assessment, this application introduces an advanced machine learning model—a vascular stiffness assessment model—for fusion decision-making. This vascular stiffness assessment model is a multi-task learning model. The advantage of this model architecture lies in learning the common feature representations of all tasks through a shared layer, and then completing different but related prediction tasks through specific branches. This typically improves the generalization performance of the main task.

[0057] Specifically, the input feature vector of the vascular stiffness assessment model It can be constructed as follows: ,in, , and These represent the user's age, gender, and height, respectively. First, the input vector undergoes non-linear transformation and feature abstraction through a shared feature extraction layer (which can consist of multiple fully connected layers) to obtain a high-dimensional fused feature representation. For shared layer parameters, This is a nonlinear transformation function of the shared feature extraction layer (e.g., composed of fully connected layers and activation functions). Then, the features are fused. It is fed into two branches: the main task branch and the auxiliary task branch.

[0058] 1) Main task branch: A regression layer that outputs a continuous vascular stiffness assessment value (e.g., a value similar to baPWV or a standardized stiffness index). , It is a function of the main task branch.

[0059] 2) Auxiliary task branch: A classification layer that outputs vascular age classification (e.g., younger than actual age, consistent with actual age, older than actual age) or cardiovascular event risk classification (e.g., low risk, intermediate risk, high risk). , It is a function that assists in the task branch.

[0060] During the training phase, the vascular stiffness assessment model optimizes its parameters by minimizing the joint loss function. :

[0061]

[0062] in, It is a joint loss function. It is the regression loss function used for the main task (e.g., mean squared error, MSE). It is a classification loss function used for auxiliary tasks (e.g., cross-entropy loss). It is a true label of the actual value of blood vessel stiffness. It is a true label of vascular age (or risk level). and It is a hyperparameter that balances the losses of the two tasks. This multi-task learning paradigm allows the model to utilize more diverse supervision signals during training, thereby learning feature representations that better reflect the health status of blood vessels, significantly improving the accuracy, robustness, and clinical relevance of vascular stiffness assessment values.

[0063] Following the core vascular stiffness assessment, this application also includes a series of enhancement steps designed to improve the system's long-term suitability, result reliability, and user experience. These steps, combined with the core assessment process, constitute a complete and robust daily vascular health monitoring solution.

[0064] The initial vascular stiffness assessment model was a general model built upon large-scale population data. However, individual users have different vascular physiological characteristics, which can change over time or with lifestyle habits. To make the assessment results more relevant to individual users, this application designed a model update mechanism, namely... Figure 2The example method may also include updating the vascular stiffness assessment model, namely: periodically collecting multiple sets of data from users in resting and preset exercise load states; based on these multiple sets of data, jointly optimizing and personally calibrating the model parameters of the main and auxiliary task branches of the vascular stiffness assessment model to improve the individualized accuracy of the user's vascular health status assessment. More specific implementation methods are as follows: 1) and 2).

[0065] 1) Triggering the Update: Updates can be triggered periodically (e.g., monthly) or when a significant trend of change in the user's physiological state is detected. The data required for the update consists of multiple sets of data collected in both resting and post-exercise recovery states (e.g., after standardized stepping exercises). These data should include the characteristic time differences calculated through steps S201 to S205 described above. Waveform morphological parameters and reference vascular stiffness values, which may be obtained through calibration equipment, can be used as monitoring signals, for example.

[0066] 2) Joint Optimization and Personalized Calibration: This process does not involve retraining the model, but rather performing personalized calibration. Specifically, transfer learning or incremental learning techniques can be used. A pre-trained general model can be used as a base, fixing most of the parameters of its shared feature extraction layer (…). Only the last one or a few levels of parameters in the main task branch and auxiliary task branches ( , Fine-tuning is then performed. The objective function for optimization is to minimize a weighted loss on the user's new data.

[0067]

[0068] in, It is the loss function in the personalized calibration stage. This is the loss function for the main task branch, which can be measured using mean squared error or mean absolute error. It is the loss function for auxiliary task branches, and is usually measured by cross-entropy. and This is either a reference value for the user during this calibration or a stable value obtained through statistical analysis of multiple measurements. and These are the weighting coefficients. Through this joint optimization, the model can adaptively adjust its output without forgetting general knowledge, making it more consistent with the user's individual characteristics. The scheme in the above embodiment significantly reduces the systematic error introduced by individual differences, making long-term tracked trends more valuable.

[0069] The accurate calculation of pulse wave velocity (PWV) depends on the path length (L) and the time difference (L / W). ),Right now Therefore, accurately estimating L is crucial when the arterial path within the body cannot be directly measured. To address this issue, the static physiological parameters in the above embodiments include at least the user's age, gender, and height. Figure 2 The example method may further include, prior to step S205: estimating the personalized arterial path length between the first and second measurement sites based on the user's height and utilizing a database containing anatomical mappings of major human arterial pathways; dividing the personalized arterial path length by the characteristic time difference to obtain the pulse wave velocity, and inputting it as one of the intermediate parameters into the vascular stiffness assessment model. In specific implementations, the database stores regression relationships between the path length from the radial artery to the fingertip artery and parameters such as height and arm length, established based on extensive human anatomical studies. For example, a simplified linear regression model could be:

[0070]

[0071] in, , and b These are the coefficients obtained through data fitting. After the user inputs their height and arm length, the system can calculate the personalized arterial path length. Then, calculations were performed. The calculated PWV, as a strongly correlated intermediate parameter, is input into the vascular stiffness assessment model along with the original features, providing the model with more direct physical and physiological information and helping to improve the accuracy of the assessment. It should be noted that, in this embodiment, PWV refers to the pulse wave velocity from the radial artery to the digital artery, i.e., RAD-PWV.

[0072] After completing the core vascular stiffness assessment in steps S201 to S205 of the above embodiments, to further improve the reliability of the system in real-world usage scenarios, the comprehensive value of the results, and the accuracy of long-term tracking, this application also includes the following important enhancement steps. These steps work in conjunction with the core assessment process to form a complete, robust, and intelligent vascular health monitoring solution.

[0073] The vascular stiffness assessment value output in step S205 primarily reflects the static vascular state of the user at the time of measurement. However, blood vessels are dynamic organs, and their responsiveness to physiological load is an important dimension for assessing their overall functional health. Therefore, after step S205, the dynamic regulation index of vascular function is calculated and comprehensively assessed. Specifically, the calculation and comprehensive assessment of the dynamic regulation index of vascular function can be achieved through steps S2061 to S2063, as detailed below:

[0074] Step S2061: Obtain at least two sets of detection data of the user under different physiological states, including resting state and post-exercise recovery state.

[0075] For example, users can be guided to first perform resting measurements, then complete a set of standardized light exercises (e.g., marching in place for 1 minute), and then take measurements again at specific moments during the recovery period after the exercise, such as at the 1st minute and the 3rd minute.

[0076] Step S2062: Calculate the dynamic regulation index of vascular function based on the vascular stiffness assessment values ​​under different physiological conditions.

[0077] The Vascular Function Dynamics Index (VFA) is used to quantify the ability of blood vessels to recover from a loaded state to a resting state. One calculation method is: VFA = (Post-exercise stiffness value - Resting stiffness value) / Recovery time. The smaller the absolute value of the VFA or the faster the recovery to baseline, the better the vascular regulatory function.

[0078] Step S2063: Use the vascular function dynamic regulation index and vascular stiffness assessment value together as the basis for comprehensive assessment of vascular health status.

[0079] The static stiffness value combined with the dynamic adjustment index provides a more comprehensive picture of vascular function, which helps to detect the trend of declining vascular elasticity at an early stage.

[0080] Because wearable devices face complex dynamic environments in daily use, single measurements may suffer from poor data quality due to accidental interference. Directly using low-quality data for analysis will produce unreliable evaluation results and mislead users. Therefore, after completing the core signal processing and feature extraction, a reliability assessment of the measurement process is performed, and a re-examination is guided when necessary. This reliability assessment and re-examination mechanism is a crucial step in ensuring the reliability of the results. As an embodiment of this application, the reliability assessment and re-examination can be implemented through steps S2071 to S2074, as detailed below:

[0081] Step S2071: Calculate the morphological similarity between the first feature waveform and the second feature waveform.

[0082] Ideally, the pulse waveforms from two locations should be highly similar in shape, differing only in time. Low morphological similarity may indicate severe interference in one channel or poor sensor contact. Similarity can be quantified by calculating the correlation coefficient in the time domain or the coherence coefficient in the frequency domain between the two waveforms.

[0083] Step S2072: Calculate the coefficient of variation of the characteristic time difference obtained in multiple consecutive cardiac cycles.

[0084] Within a short period, the pulse wave conduction time should remain relatively stable. The coefficient of variation of the characteristic time difference. ,in, and These are consecutive multiple cycles The standard deviation and mean of the CV. A smaller CV indicates more stable measurement; a higher CV indicates inaccurate feature point recognition or signal instability. If any indicator exceeds a preset threshold, the reliability of the detection is considered low.

[0085] Step S2073: When the morphological similarity between the first feature waveform and the second feature waveform is lower than the similarity threshold, or the coefficient of variation of the feature time difference is higher than the coefficient of variation threshold, the reliability of this detection is determined to be low.

[0086] Step S2074: In response to the low confidence level, a re-test guidance instruction is generated based on motion state data and / or signal quality indicators, wherein the re-test guidance instruction is used to instruct the user to adjust limb posture or to automatically adjust the sensing parameters of the wearable device.

[0087] This application further responds to a low reliability determination by generating a re-detection guidance instruction based on motion state data and / or signal quality indicators. This involves the following intelligent evaluation and feedback process:

[0088] If the motion data indicates that the limb is swaying continuously: "Continuous swaying" here can be defined as: during the data acquisition period, the acceleration or angular velocity signal measured by the inertial measurement unit continuously exceeds a certain threshold (for example, the acceleration signal of the inertial measurement unit continuously exceeds 2 m / s² and the duration is greater than 3 seconds), and the spectral energy is concentrated in the low frequency band (for example, 1~5Hz), indicating that it is not a brief accidental movement. At this time, the system generates an instruction to guide the user to lean the limb against a stable support, for example, the screen displays: "Please gently place your arm on the table, keep it stable and then measure again";

[0089] If the signal quality index indicates insufficient signal perfusion: "Insufficient signal perfusion" here can be characterized as: the AC component amplitude of the photoplethysmography (PPG) signal is consistently below the normal range or the DC component fluctuates too much (for example, the AC component amplitude of the PPG signal is consistently below 0.5% or the DC component fluctuates more than 10%). In this case, the system generates instructions to guide the user to move their limbs and then remeasure or automatically increase the sensor's light intensity, such as prompting: "Please rub your fingers or gently move your wrist to improve blood circulation and then remeasure", or the device automatically increases the light power of the PPG sensor to obtain a stronger signal.

[0090] To ensure the reliability of the results, this application also introduces a cross-validation mechanism for the evaluation results. Specifically, this includes inputting the vascular stiffness assessment value, waveform morphology parameters, and characteristic time difference into a pre-set rule base or lightweight validation model. When the deviation between the vascular stiffness assessment value and the estimated value calculated based on the waveform morphology parameters and / or characteristic time difference through the rule base or lightweight validation model exceeds a reasonable range, the current evaluation result is marked or a new round of testing is initiated to ensure the reliability of the vascular function evaluation results. For example, a simple rule could be: if the vascular stiffness assessment value is high, but the waveform morphology parameter ratio is also high (usually indicating acceptable elasticity), there may be a contradiction. The system will mark the result or suggest a retest, which is equivalent to adding a safety valve to the model output.

[0091] To enable the vascular stiffness assessment model to adapt to long-term physiological changes in users and avoid assessment biases caused by individual differences and physiological drift, this application also introduces a personalized calibration mechanism based on a baseline model. Specifically, the joint optimization and personalized calibration of the model parameters of the main task branch and auxiliary task branch of the vascular stiffness assessment model based on multiple sets of data from the user in a resting state and under a preset exercise load state, as detailed in steps S2081 to S2084, are as follows:

[0092] Step S2081: Using data collected in a resting state, establish a personalized vascular function baseline model for the user, wherein the personalized vascular function baseline model is used to characterize the normal fluctuation range of the characteristic time difference and waveform morphological parameters of the user in a resting state.

[0093] A personalized vascular function baseline model is essentially a multivariate Gaussian model used to describe the user's characteristic time differences and normal fluctuation range of waveform morphological parameters at rest. For example, the mean vector of these parameters can be calculated. Given the covariance matrix Σ, the normal fluctuation range can be defined as follows: The space centered at a certain standard deviation.

[0094] Step S2082: Analyze the dynamic response pattern of the user's vascular function under load using data collected under a preset exercise load state.

[0095] The dynamic response pattern of a user's vascular function under load can be characterized as a curve of changes in vascular stiffness assessment values ​​after exercise, such as peak value, time to reach peak value, time required to recover to half, and other characteristic parameters.

[0096] Step S2083: In subsequent testing, the measured resting state parameters are compared with the prediction range of the personalized vascular function baseline model, and the measured load state response pattern is analyzed in conjunction with the dynamic response pattern of the user's vascular function under load.

[0097] Step S2084: When the resting state parameters detected in multiple consecutive tests continuously deviate from their predicted range, and / or the load state response pattern changes significantly, the user-related personalized parameters in the vascular stiffness assessment model are fine-tuned.

[0098] When resting state parameters consistently deviate from their predicted range across multiple consecutive (e.g., 5 times) measurements, and / or the load-state response pattern changes significantly (e.g., the change exceeds twice the standard deviation of historical fluctuations), this indicates a genuine and persistent change in the user's physiological baseline, rather than a random fluctuation. In this case, the system triggers fine-tuning of the user-related personalized parameters in the vascular stiffness assessment model. The fine-tuning process, as described above, uses the user's latest valid data to incrementally learn the model's main and auxiliary task branches with a small learning rate, realigning the model output with the user's new physiological state.

[0099] From the above appendix Figure 2The example of a vascular function detection method based on multi-physiological signal fusion demonstrates two key advantages. First, by simultaneously acquiring arterial pulse wave signals through at least two sensing units placed at different arterial measurement points on the user's body surface, and calculating the characteristic time difference of the pulse wave between the two measurement points, the method can directly obtain the key physiological parameter reflecting vascular stiffness—pulse wave conduction time. This provides a direct and reliable data foundation for subsequent PWV calculation and vascular stiffness assessment, overcoming the limitation of single-sensor solutions that cannot directly measure wave conduction time, thus making the assessment results more accurate. Second, by introducing an inertial measurement unit to acquire motion state data and dynamically compensating for the extracted feature waveforms based on this data, the method can actively identify and suppress motion caused by the user's daily activities (e.g., arm swinging, gesture changes, etc.). Artifact interference can be effectively eliminated by dynamically compensating for characteristic waveforms, thus ensuring the accuracy of characteristic time difference calculation and ultimately improving the reliability and robustness of the entire system in real-life scenarios. Thirdly, the technical solution of this application does not simply calculate PWV, but rather inputs the calculated characteristic time difference, the user's static physiological parameters, and waveform morphological parameters directly extracted from arterial pulse wave signals into a vascular stiffness assessment model for fusion analysis. The pulse wave morphological parameters themselves contain rich vascular state information. This multi-parameter fusion analysis strategy can comprehensively assess vascular function from both wave propagation velocity and waveform morphology dimensions, avoiding the limitations of a single indicator and facilitating the acquisition of more comprehensive vascular stiffness assessment values ​​that better reflect individual physiological characteristics. In summary, the technical solution of this application improves the accuracy and reliability of vascular assessment in everyday scenarios by using dual sensors to measure time differences and combining motion compensation with multi-parameter fusion.

[0100] Please see Figure 3 As shown, in one embodiment, a vascular function detection device based on multi-physiological signal fusion is provided. This device may include an acquisition module 301, a compensation module 302, an extraction module 303, a calculation module 304, and an evaluation module 305, as detailed below:

[0101] The acquisition module 301 is used to acquire arterial pulse wave signals at the first and second measurement sites of the user, respectively, through the first and second sensors configured on the wearable device.

[0102] The compensation module 302 is used to collect motion state data of the user during the process of collecting arterial pulse wave signals through the inertial measurement unit configured on the wearable device, and to dynamically compensate the arterial pulse wave signals based on the motion state data.

[0103] The extraction module 303 is used to extract feature waveforms based on the dynamically compensated arterial pulse wave signal through a signal processing model. It is used to identify waveform feature points that are strongly correlated with the cardiac cycle in the arterial pulse wave signal. The signal processing model is a neural network model based on the attention mechanism. The feature waveforms include a first feature waveform corresponding to the first measurement site and a second feature waveform corresponding to the second measurement site.

[0104] The calculation module 304 is used to calculate the characteristic time difference of the pulse wave from the first measurement site to the second measurement site based on the first characteristic waveform and the second characteristic waveform after dynamic compensation.

[0105] The evaluation module 305 is used to input the characteristic time difference of the pulse wave from the first measurement site to the second measurement site, the user's static physiological parameters, and at least one waveform morphology parameter extracted from the arterial pulse wave signal into the vascular stiffness evaluation model to obtain a vascular stiffness evaluation value. The waveform morphology parameter includes at least one of the following: the slope of the rising branch of the pulse wave and the ratio of the peak value of the reflected wave to the peak value of the main wave.

[0106] From the above appendix Figure 3As illustrated by the example of a vascular function detection device based on multi-physiological signal fusion, firstly, by simultaneously acquiring arterial pulse wave signals through at least two sensing units placed at different arterial measurement points on the user's body surface, and calculating the characteristic time difference of the pulse wave between the two measurement points, the key physiological parameter reflecting vascular stiffness—pulse wave conduction time—can be directly obtained. This provides a direct and reliable data foundation for subsequent PWV calculation and vascular stiffness assessment, overcoming the deficiency of single-sensor solutions in directly measuring wave conduction time, thus making the assessment results more accurate. Secondly, by introducing an inertial measurement unit to acquire motion state data and dynamically compensating for the extracted feature waveforms based on this data, the solution can actively identify and suppress motion caused by the user's daily activities (e.g., arm swinging, gesture changes, etc.). Artifact interference can be effectively eliminated by dynamically compensating for characteristic waveforms, thus ensuring the accuracy of characteristic time difference calculation and ultimately improving the reliability and robustness of the entire system in real-life scenarios. Thirdly, the technical solution of this application does not simply calculate PWV, but rather inputs the calculated characteristic time difference, the user's static physiological parameters, and waveform morphological parameters directly extracted from arterial pulse wave signals into a vascular stiffness assessment model for fusion analysis. The pulse wave morphological parameters themselves contain rich vascular state information. This multi-parameter fusion analysis strategy can comprehensively assess vascular function from both wave propagation velocity and waveform morphology dimensions, avoiding the limitations of a single indicator and facilitating the acquisition of more comprehensive vascular stiffness assessment values ​​that better reflect individual physiological characteristics. In summary, the technical solution of this application improves the accuracy and reliability of vascular assessment in everyday scenarios by using dual sensors to measure time differences and combining motion compensation with multi-parameter fusion.

[0107] In one embodiment, a computer device is provided, the computer device may be... Figure 1 The internal structure diagram of the example intelligent recovery solution generation system 102 can be shown as follows: Figure 4 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used for communication with external clients via a network connection. When executed by the processor, the computer program implements the functions or steps of a vascular function detection method based on multi-physiological signal fusion.

[0108] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, performs the following steps:

[0109] The first and second sensors configured on the wearable device collect arterial pulse wave signals from the user's first and second measurement sites, respectively.

[0110] The inertial measurement unit configured on the wearable device collects the user's motion state data during the process of collecting arterial pulse wave signals, and dynamically compensates the arterial pulse wave signals based on the motion state data.

[0111] Based on the dynamically compensated arterial pulse wave signal, feature waveforms are extracted through a signal processing model to identify waveform feature points strongly correlated with the cardiac cycle in the arterial pulse wave signal. The signal processing model is a neural network model based on the attention mechanism, and the feature waveforms include a first feature waveform corresponding to the first measurement site and a second feature waveform corresponding to the second measurement site.

[0112] Based on the first and second characteristic waveforms after dynamic compensation, the characteristic time difference of pulse wave propagation from the first measurement site to the second measurement site is calculated.

[0113] The characteristic time difference of pulse wave propagation from the first measurement site to the second measurement site, pulse wave propagation velocity, user's static physiological parameters, and at least one waveform morphology parameter extracted from the arterial pulse wave signal are input into the vascular stiffness assessment model to obtain the vascular stiffness assessment value. The waveform morphology parameter includes at least one of the following: the slope of the rising branch of the pulse wave and the ratio of the peak value of the reflected wave to the peak value of the main wave.

[0114] The aforementioned computer program improves the accuracy and reliability of vascular assessment in everyday scenarios by using dual sensors to measure time difference and combining motion compensation with multi-parameter fusion.

[0115] In one embodiment, a computer-readable storage medium is provided that stores a computer program, which, when executed by a processor, performs the following steps:

[0116] The first and second sensors configured on the wearable device collect arterial pulse wave signals from the user's first and second measurement sites, respectively.

[0117] The inertial measurement unit configured on the wearable device collects the user's motion state data during the process of collecting arterial pulse wave signals, and dynamically compensates the arterial pulse wave signals based on the motion state data.

[0118] Based on the dynamically compensated arterial pulse wave signal, feature waveforms are extracted through a signal processing model to identify waveform feature points strongly correlated with the cardiac cycle in the arterial pulse wave signal. The signal processing model is a neural network model based on the attention mechanism, and the feature waveforms include a first feature waveform corresponding to the first measurement site and a second feature waveform corresponding to the second measurement site.

[0119] Based on the first and second characteristic waveforms after dynamic compensation, the characteristic time difference of pulse wave propagation from the first measurement site to the second measurement site is calculated.

[0120] The characteristic time difference of pulse wave propagation from the first measurement site to the second measurement site, pulse wave propagation velocity, user's static physiological parameters, and at least one waveform morphology parameter extracted from the arterial pulse wave signal are input into the vascular stiffness assessment model to obtain the vascular stiffness assessment value. The waveform morphology parameter includes at least one of the following: the slope of the rising branch of the pulse wave and the ratio of the peak value of the reflected wave to the peak value of the main wave.

[0121] When the aforementioned computer program is executed by the processor, it measures the time difference using dual sensors and combines motion compensation with multi-parameter fusion, thereby improving the accuracy and reliability of vascular assessment in everyday scenarios.

[0122] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0123] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0124] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0125] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for detecting vascular function based on the fusion of multiple physiological signals, characterized in that, Includes the following steps: The first and second sensors configured on the wearable device collect arterial pulse wave signals from the user's first and second measurement sites, respectively. The wearable device uses an inertial measurement unit to collect motion state data of the user during the acquisition of arterial pulse wave signals, and performs dynamic compensation on the arterial pulse wave signals based on the motion state data. This dynamic compensation includes: parsing the user's limb posture change information and / or acceleration information from the motion state data; generating a compensation signal related to motion artifacts based on the posture change information and / or acceleration information; and filtering out components corresponding to the compensation signal from the arterial pulse wave signals. Based on the dynamically compensated arterial pulse wave signal, feature waveforms are extracted through a signal processing model to identify waveform feature points strongly correlated with the cardiac cycle in the arterial pulse wave signal. The signal processing model is a neural network model based on an attention mechanism. The feature waveforms include a first feature waveform corresponding to the first measurement site and a second feature waveform corresponding to the second measurement site. Based on the dynamically compensated first and second characteristic waveforms, the characteristic time difference of pulse wave propagation from the first measurement site to the second measurement site is calculated. This calculation includes: performing phase alignment processing on the first and second characteristic waveforms; calculating the time difference between the aligned two characteristic waveforms at preset characteristic points, where the preset characteristic points include at least one of the waveform start point, the main wave peak point, and the diatonic wave notch point, and the selection of the characteristic points is dynamically determined based on waveform morphological parameters. The characteristic time difference, pulse wave velocity, user's static physiological parameters, and at least one waveform morphological parameter extracted from the arterial pulse wave signal are input into a vascular stiffness assessment model to obtain a vascular stiffness assessment value. The waveform morphological parameters include at least one of the following: the slope of the rising limb of the pulse wave and the ratio of the peak value of the reflected wave to the peak value of the main wave. The vascular stiffness assessment model is a multi-task learning model. The process of inputting the characteristic time difference, pulse wave velocity, user's static physiological parameters, and at least one waveform morphological parameter extracted from the arterial pulse wave signal into the vascular stiffness assessment model to obtain a vascular stiffness assessment value includes: processing the characteristic time difference, pulse wave velocity, static physiological parameters, and waveform morphological parameters through the shared feature extraction layer of the multi-task learning model. The data are fused and encoded to obtain fused features; the fused features are input into a main task branch and at least one auxiliary task branch, wherein the main task branch regresses and outputs the vascular stiffness assessment value, and the auxiliary task branch outputs a vascular age classification or cardiovascular event risk grading for assessing vascular status; the static physiological parameters include at least age, gender, and height, and the method further includes, before obtaining the vascular stiffness assessment value: based on the height and using a database containing anatomical mapping relationships of major human arterial pathways, estimating the personalized arterial path length between the first measurement site and the second measurement site; dividing the personalized arterial path length by the feature time difference to obtain the pulse wave conduction velocity, and inputting it as one of the intermediate parameters into the vascular stiffness assessment model.

2. The vascular function detection method based on multi-physiological signal fusion according to claim 1, characterized in that, After obtaining the vascular stiffness assessment value, the following is also included: Acquire at least two sets of detection data from the user under different physiological states, including resting state and post-exercise recovery state; Based on the vascular stiffness assessment values ​​under different physiological conditions, the dynamic regulation index of vascular function was calculated. The vascular function dynamic regulation index and the vascular stiffness assessment value are used together as the basis for comprehensively assessing vascular health status.

3. The vascular function detection method based on multi-physiological signal fusion according to claim 2, characterized in that, The method also includes a cross-validation step for evaluating the results: Input the vascular stiffness assessment value, waveform morphology parameters, and characteristic time difference into a preset rule base or lightweight validation model; When the deviation between the assessed vascular stiffness value and the estimated value exceeds a reasonable range, the current assessment result is marked or a new round of testing is initiated to ensure the reliability of the vascular function assessment result. The estimated value is an estimated value calculated based on the waveform morphological parameters and / or feature time difference through the rule base or an estimated value calculated through a lightweight validation model.

4. A vascular function detection device based on multi-physiological signal fusion, characterized in that, The device includes: The acquisition module is used to acquire arterial pulse wave signals at the user's first and second measurement sites respectively through the first and second sensors configured on the wearable device; The compensation module is used to collect motion state data of the user during the acquisition of arterial pulse wave signals through an inertial measurement unit configured on the wearable device, and to dynamically compensate the arterial pulse wave signals based on the motion state data. The dynamic compensation of the arterial pulse wave signals based on the motion state data includes: parsing the user's limb posture change information and / or acceleration information from the motion state data; generating a compensation signal related to motion artifacts based on the posture change information and / or acceleration information; and filtering out components corresponding to the compensation signal from the arterial pulse wave signals. The extraction module is used to extract feature waveforms based on the dynamically compensated arterial pulse wave signal through a signal processing model. It is used to identify waveform feature points that are strongly correlated with the cardiac cycle in the arterial pulse wave signal. The signal processing model is a neural network model based on an attention mechanism. The feature waveforms include a first feature waveform corresponding to the first measurement site and a second feature waveform corresponding to the second measurement site. The calculation module is used to calculate the characteristic time difference of pulse wave propagation from the first measurement site to the second measurement site based on the dynamically compensated first characteristic waveform and the second characteristic waveform. The calculation of the characteristic time difference of pulse wave propagation from the first measurement site to the second measurement site based on the first characteristic waveform and the second characteristic waveform includes: performing phase alignment processing on the first characteristic waveform and the second characteristic waveform; calculating the time difference between the two characteristic waveforms after alignment at a preset characteristic point, wherein the preset characteristic point includes at least one of the waveform start point, the main wave peak point, and the diatonic wave notch point, and the selection of the characteristic point is dynamically determined based on waveform morphological parameters. The evaluation module is used to input the characteristic time difference, pulse wave velocity, user's static physiological parameters, and at least one waveform morphological parameter extracted from the arterial pulse wave signal into a vascular stiffness evaluation model to obtain a vascular stiffness evaluation value. The waveform morphological parameters include at least one of the following: the slope of the rising limb of the pulse wave and the ratio of the peak value of the reflected wave to the peak value of the main wave. The vascular stiffness evaluation model is a multi-task learning model. The process of inputting the characteristic time difference, pulse wave velocity, user's static physiological parameters, and at least one waveform morphological parameter extracted from the arterial pulse wave signal into the vascular stiffness evaluation model to obtain a vascular stiffness evaluation value includes: processing the characteristic time difference, pulse wave velocity, static physiological parameters, and waveform morphological parameters through the shared feature extraction layer of the multi-task learning model. Morphological parameters are fused and encoded to obtain fused features. These fused features are then input into a main task branch and at least one auxiliary task branch. The main task branch regresses and outputs the vascular stiffness assessment value, while the auxiliary task branch outputs a vascular age classification or cardiovascular event risk grading for assessing vascular status. The static physiological parameters include at least age, sex, and height. Before obtaining the vascular stiffness assessment value, the model further includes: estimating the personalized arterial path length between the first and second measurement sites based on the height and using a database containing anatomical mappings of major human arterial pathways; dividing the personalized arterial path length by the feature time difference to obtain the pulse wave velocity, which is then input into the vascular stiffness assessment model as one of the intermediate parameters.

5. A computer device, characterized in that, The computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the vascular function detection method based on multi-physiological signal fusion as described in any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the vascular function detection method based on multi-physiological signal fusion as described in any one of claims 1 to 3.

Images